Evaluating AIDS Prevention Programs: Expanded Edition (1991)

Below is the uncorrected machine-read text of this chapter, intended to provide our own search engines and external engines with highly rich, chapter-representative searchable text of each book. Because it is UNCORRECTED material, please consider the following text as a useful but insufficient proxy for the authoritative book pages.

Methodological Issues in AIDS Surveys To supplement our discussion of evaluation measurement in Chapter 2, this appendix presents an excerpt on measurement issues in AIDS research from AIDS, The Second Decade (Miller, Turner, and Moses, 1990:3594711. Readers my also wish to consult the discussion of sampling and related issues in the prior report, AIDS, Sexual Behavior, and Intravenous Drug Use (Turner, Miller, arm Moses, 1989:147-157, 214-225J. INTRODUCTION Surveys or, more generally, He memos of asking questions and recording answers, continue to be one of the most important methods for obtaining essential information about Be epidemiology of AIDS and ~V, He be- haviors Hat spread ~V, and the effectiveness of AIDS prevention efforts. Previous reports of our committee have included numerous examples of surveys and obsenabons about the methodological difficulties that often attend these measurements. Because of the central role surveys play in research on AIDS and ~V, this appendix focuses on methodological aspects of this data- gather~;ng method Hat have important consequences for He usefulness of survey data. This appendix contains much technical material and methodological detail. Our aim in presenting this material is to provide researchers conducting AIDS surveys or analyzing data collected In such surveys win a detailed review of the current state of methodological re- search In this area. Readers who seek only a synopsis of our conclusions and recommendations may wish to consult pages 27-34 of the summary chapter of the report AIDS: The Second Decade (Miller, Turner, and Moses, 19901. 207

208 ~ APPENDIX C Before honing to specifics, it may be useful to consider data gathering in general and the types of problems that may compromise the collection of accurate and informative data. One may usefully distinguish five aspects of survey data collection: (~) the definition of the population to be studied and the drawing of a target sample from that population; (2) the execution of the sample design, that is, finding the persons in the target sample and enlisting their cooperation in the survey; (3) the posing of questions to elicit Me desired information; (4) the answering of those questions by the respondent; and (5) the recording of those answers (~d subsequent data processing and analysis). To examine these elements, let us consider a hypothetical survey (much like the decennial census) that targeted all households in a par- ticular junsdiction of the state of Texas. Let us suppose further that the information to be obtained concerned automobile ownership (e.g., how many automobiles were owned by each household, the make and year of the autos, etc.~. This survey, although not simple to conduct, would nev- er~eless be considerably less difficult to conduct than a survey seeking to assess behaviors that transmit REV; In particular, · the survey involves matters of fact that are both open to direct observation and matters of public record (e.g., make and year of automobiles owned by household); · We topic is unlikely to be regarded as sensitive or "nri- vate" by respondents, although as in any survey—some respondents may not wish to take the time to respond; · developing questions about this topic can draw on a widely shared vocabulary (i.e., there is little ambiguity about what constitutes a "car" or "ownership"; · respondents who are not wed informed can consult win other household members or check records (e.g., registration certificates); checks of survey accuracy can be made at Me group leve] (by comparing the rates of auto ownership found In Me survey and in registration records) and at the individual level (by checking individual regis~ationsi); and · census data on income and statewide auto reg~s~ation data are available to target Me survey efficiently toward seg- ments of Me population of particular interest (e.g., current or "potential" owners of Belchfire 500s). Ache survey may, however, produce detailed information Bat cannot be verified from public records for example, die proportion of Belchfire 500s owned by persons with 16+ years of education.

METHODOLOGICAL ISSUES ~ 209 Surveys that inquire about sexual behaviors or IV Mug use differ in several ways from He foregoing example, and these differences pro- vide a much greater challenge to the survey (or question-and-answer) method. First, many of the logical "target populations" for drug use and sexual behavior surveys cannot be identified reliably from officio] statistics. There are few reliable data on the distnbution across the na- tion of persons who engage In behaviors that risk HIV transmission. Furthermore, the behaviors in question occur In private and cannot be verified by direct observation or public records. Many of the behaviors are actively concealed because they are considered illicit (IV drug use is illegal throughout the nation, and many sexual behaviors of interest in preventing HIV transmission are illegal in some states). Thus, the topics these surveys cover are likely to be highly sensitive, which may create difficulties in enlisting the cooperation of persons ~ a target sample and in obtaining permission from "gatekeepers" (e.g., high school authorities) who control access to particular populations (e.g., high school students). This appendix considers these problems and reviews the available empirical evidence gathered from surveys of sexual and drug use behav- iors. Before beginning this review, however, some cautionary words are in order. The evidence presented here regarding errors In data about sen- sitive behaviors might lead some readers to unwarranted and wholesale rejection of survey findings on these important topics.2 Indeed, consid- er~ng the litany of difficulties presented In this appendix, some readers may ask whether anything at all can be learned from surveys or whether surveys have a useful role to play in research on AIDS and REV trans- mission. The following considerations prompt the committee to answer "yes" to these questions. The most important consideration arises directly from the nature of the disease. HIV infection occurs through the joint operation of the biology of this particular infectious virus and the human behaviors that transmit it. In the absence of vaccines, all interventions that seek to retard the spread of HIV infection focus on changing human behaviors to diminish the probability that the virus will be transmitted. Data on these behaviors are needed for a number of important purposes for example, to understand the factors that motivate and shape the behaviors and to determine whether behaviors that transmit HIV are becoming less frequent in the population. It might, of course, be argued that merely mo~to~g changes in He prevalence of HIV would be sufficient to determine whether behaviors 2The following pages borrow heavily from He discussion of errors in survey measurements in linger and Martin (1984:Vol. 1, 1016) and Tumer (1989).

210 ~ APPENDIX C change was occulting. Although this argument is true to some extent, there are important deficiencies in any strategy that eschews direct mea- surement of the behaviors themselves. Reliable data on HIV prevalence and incidence, although of great value for many purposes, are only a final accounting of the number of infected and uninfected persons in He population. From the viewpoint of prevention, such statistics serve best as a catalog of failures. Yet, those who are uninfected are not necessary successes. For example, the very low rate of HIV infection in states like Wyoming does not necessarily imply that the population has adopted protective behaviors. Instead, the low rate of HIV prevalence could be attnbutable to an epidemiological happenstance (e.g., isolation—in teens of sexual contacts and injection equipment sharing from populations with high HIV prevalence.) Determining whether protective behavioral changes have occurred fin Wyoming or anywhere else) requires asking questions about these risky behaviors. This activity, In tum, raises a host of methodological issues that are germane to survey research of aU types plus some questions Hat are specific to surveys of drug use and sexual behavior. The questions may be quite basic: Are the respondents telling the truth? Do Hey understand the meaning of the survey questions In He same way the investigator does? Simple or complex, such questions inevitably introduce a degree of uncertainty into the interpretation of aD survey data. Grappling with these issues forces an appreciation of the human interac cons that produce survey measurements. process Elsewhere it has been argued that fundamental aspects of the survey are quintessentially social psychological in character. They arise from a complex interpersonal exchange, they embody the subjectivities of both interviewer and interviewee, and they present their interpreter with an analytical challenge that requires a multitude of assumptions concerning, among other things, how respondents experience the reality of the interview situation, decode the "meaning" of survey questions, and respond to the social presence of the interviewer and the demand characteristics of the interview. (Tu~ner, 1984:202) Although this "analytical challenge" may be substantial, researchers are aided in their task by several decades of methodological research (see, for example, Sudman and Bradbunn [1974], Bradburn and Sudman [1979], Rossi, Wright, and Anderson [1983], Turner and Martin [1984], and Catania et al. [199Oa,bl). A further reason for not abandoning behavioral measurement is that many of the problems encountered in this arena are not unique. Useful lessons may thus be learned from other disciplines that also confront such challenges.

METHODOLOGICAL ISSUES ~ 21t Fallibility of Measurement in Other Sciences Fallibility and error are not confined to behaviors measurements, as evidenced by the decade-Ion" controversy suuTounding the population statistics produced by the decennial censuses.3 Furthermore, just as falli- bility of measurement is not limited to behavioral measurements, neither is it I:m~ted to surveys or social statistics. For example, Hunter (1977) and Lide (1981) have noted the variability among measurements of such elementary physical phenomena as the thermal conductivity of copper (Figure C-1~. As Hunter observed, "although each analyst measured a physical quality that did not vary with location or time, it is clear that a remarkable vanability attended He measurements" (1977:2~. He concluded: "The variation in attempting to evaluate the same physi- cal constant is obvious. This example is not unusual. Similar plots of thermal conductivity as a function of temperature for approximately 400 common metals and materials can be found in a supplement to the lour- nal (Ho, Powell, and Liley, 1974~. Nor is the observed variation in the measurement of 'thermal conductivity' unique among physical paramet- ers ... ." Common biological measurements have shown similar fallibility. Examples include data collected by CDC Hat show substantial variation in the est~rnates made by different laboratories of the amount of lead in identical samples of blood. For a sample of blood with a putative lead concentration of 41 milligrams per deciliter (mg/Dl), 100 cooperating laboratories produced measurements that ranged from 33 to 55 mg/Dl; this result prompted He reviewer to observe: "Clearly, whatever He true amount of lead in a sample, the variability demonstrated [in these measurements] guarantees numerous false alanns Perhaps more ~m- portant when the true level is high nonalanns" (Hunter, 1980:870~. Another category of fallibility in the physical sciences involves "dis- coveries" that are later shown to be experimental artifacts. For example, between 1963 and 1974 more than 500 journal articles (including some ~ Science and Nature ~ discussed a supposed new substance: anomalous water, or polywater. Although it resembled ordinary water, polywater allegedly had a greater density, a reduced freezing point, and an elevated boiling point, among over anomalous properties. In He end, however, it was discovered Hat this "new substance" was nothing more than an impure solution of ordinary water (Franks, 1981; Eisenberg, 19811. 3By October 1981, more than 50 lawsuits had been filed challenging die accuracy of the 1980 Cen- sus results and their use in legislative apportionment and fund allocation decisions (Citro and Cohen, 1985:9).

212 Q Q o , C) · < · a: L'J I . ~ _ ~ O _ ' ~ .D ' D co - ~ ~ C) _ _ ~ ~ ' ~ (L) - OF ~ ~ g C) ~ ~ o Ct ~ ~ C) it_ 0 04 .= O ;^ - I ~ _ o —~ {_ ,_> ,_~ M ·=l~lOnO NO: l~WB3H1 o ~ ~ ;` _1 . ~ ~ G vie 3 A O ~ ~ ~ ~ D ~ , ~ i` Z ~ I ~ ~ 1 ;-, .; .: ;

METHODOLOGICAL ISSUES ~ 2~3 Such examples indicate that the problems AIDS researchers confront when they seek to assess sexual and drug-us~ng behavior are not unique in the annals of scientific measurement. As Quinn McNemar observed more than 40 years ago, "bat! measurement is befuddled with error. About this the scientist can and does do something; he ascertains the possible extent of the error, determines whether it is constant (biasing) or vanable, or both, and ever strives to improve his instruments and techniques" (1946:294~. ~ the following pages the committee reviews what is currently known about the errors that afflict measures of sexual and drug-us~ng be- havior and offers some prescriptions for how future measurements might be improved. The first section of the appendix- reviews the experience to date In mounting surveys and obtaining responses from We public. The second section considers the reliability and validity of responses obtained In surveys of sexual and drug use behaviors. The final section reviews the use of anthropological research strategies that may provide important complementary information to that obtained In surveys and that may also be crucial in questionnaire development to improve the accuracy and completeness of responses. RECRUITMENT OF RESPONDENTS IN SEX AND SEROPREVALENCE SURVEYS Much of what is now known about the epidemiology of AIDS has come from smaR-scale, local studies among subgroups thought to be at high risk for infection. Participants In these studies were recruited through venous sources and means from He clientele of local clinics or treat- ment facilities or the membership rosters of local organizations, Trough newspaper advertisements and physician referrals, and occasionally from "sweet sampling." The yield from this research has been remarkably rich. From these studies, researchers have identified the principal mecha- nisms of REV infection (i.e., transmission Trough sexual contact, sharing injection equipment In {V drug use, transfusion of contaminated blood products); verified sexual transmission of HIV from male to male, female to male, and male to female; measured the efficiency of transmission In specific kinds of sexual contacts; and discovered some of the basic fea- tures of the long natural history of this devastating disease. As valuable as these studies are, however, the data drawn from them cannot address many over important public health questions that arise because of AIDS, such as: How large is He epidemic? What is the potential for general spread of HIV infection? Can an HIV epidemic be sustained Trough het- erosexual contact alone? To answer questions like these, the knowledge

214 ~ APPENDIX C gained from measurements and observations cattier out in local studies of special subgroups must be applied in large-scale investigations of pop- ulations chosen not because of convenient or ready access but because of their importance in understanding the genera] course of the epidemic. This section considers the feasibility of sex and seroprevalence sur- veys as a means of measuring the distribution of sexual behaviors that risk HIV transmission and the distnbution of HIV itself in general pop- ulations. Although such surveys may be designed in a variety of ways, all of the studies discussed here employ the same general procedures for participant selection: an unambiguous definition of the population to be studied and a form of sampling from this population that allows the probabilities of selection to be known. The potential advantages of a probability sampling program for selecting survey participants are well known. In principle, probability sampling permits the use of a large body of statistical theory to make inferences from the sample to the larger pop- ulation and avoids the possible biases inherent in recruitment by other means. The suggestion to use probability sampling for surveys of sexual behavior was made more than three decades ago in connection with a review of the statistical methods used In Kinsey, Pomeroy, and Mar- tin's Sexual Behavior in the Human Male (1948) (Cochran, Mosteller, and Tukey, 19531. The authors of this suggestion were a committee of the Commission on Statistical Standards of the American Statistical Association. At the invitation of Dr. Kinsey and the National Research Council's Committee for Research on Problems of Sex, they were asked to provide course! on ways to improve the statistical methods used In the Kinsey research. They recommended a step-by-step program of prob- ability sampling, beginning with a small pilot effort. They argued that research of this kind would provide a check on the results obtained with Kinsey's large, nonprobability sample. The committee was aware that problems of cost and potentially high rates of nonparticipation in such surveys would present special challenges. Their comments about the limits of this approach are worth quoting at length because the issues they raised more Han 30 years ago In relation to Kinsey's work remain germane in evaluating He potential value of contemporary surveys of sexual behavior. In our opinion, no sex study of a broad human population can expect to present incidence data for reported behavior that are known to be correct to within a few percentage points. Even with the best available sampling techniques, there will be a certain percentage of the population who refuse to give histories. If the percentage of refusals is 10 percent or more, then however large the sample, there are no statistical principles which guarantee

METHODOLOGICAL ISSUES ~ 215 Mat Me results are correct to within 2 or 3 percent, . . . but any claim that this is Cue must be based on the undocumented opinion that the behavior of those who refuse to be interviewed is not very different from that of those who are interviewed. These comments, which are not a criticism of [Kinsey, Pomeroy, and Mariin's] research, emphasize the difficulty of answering the question: "How accurate are Me results?", which is naturally of great interest to any user of the results of a sex study. (Cochran, Mosteller, and Pokey, 1953:675) The rationale for using response rates as a "yardstick" to assess the accuracy of survey estimates is twofold: (~) high response rates reduce the influence of selective participation in surveys and hence the potential for bias in the estimates, and (2) for a given target sample size and sample design, the higher the response rate, Me larger the actual sample and the smaller the standard error of estimate. In other words, high response rates are better than low rates, provided the procedures used to achieve high response rates do not increase the degree of selectivity or inaccuracy of the responses. Few contemporary surveys on any topic achieve response rates higher than the 90 percent figure cited in Cochran, Mosteller, and Tukey's review of Me Kinsey report; indeed, response rates in most surveys are considerably below that mark. In principle, then, questions about selective participation (i.e., about differences between respondents and nonrespondents) are of concern in judging the accuracy of most survey estimates, not only those that derive from surveys of sexual behavior. Such concerns have generated a substantial literature on the character of nonresponse in surveys and what to do about possibly biased estimation resulting from nonresponse (see, for example, Goyder's 1987 synthesis of nonresponse research and Me series of volumes on incomplete data in sample surveys edited by Madow, Nisselson, and OLkin [198311. Surveys with response rates that are much lower than 90 percent may still provide useful estimates of population charactenshcs if it can be established that participation or nonparticipation is unrelated to Me characteristic for which an estimate is sought. Furtherrhore, response rates higher Man 90 percent do not guarantee accurate estimation if survey participation is highly selective. Thus, in most cases, Me value of the response rate by itself is insufficient justification for claims of accuracy or "representativeness" of survey estimates or for counterclaims that estimates fail in this respect. Such claims should be based on careful study, documentation, and possibly adjustment for bias as a result of refusals and other sources of nonresponse. In Me following review, the committee examines recent efforts to survey sexual behavior and related HIV risk factors Mat use probability

216 ~ APPENDIX C samples from general populations. The review focuses on participation in such surveys and is motivated by the same concerns about nonresponse in probability samples that were expressed in the review of the Kinsey report. It attempts to answer three main questions: (1) What response rates have been achieved In recent surveys of sexual behavior? (2) What survey designs and procedures appear to be associated with higher versus lower levels of participation? and (3) What can be said, at present, about differences between sample persons who participate in sex surveys and sample persons who refuse to participate or do not participate for other reasons? (There is as yet too little information to hazard general statements about differences between participants and nonparticipants in seroprevalence surveys.) Questions about the validity and reliability of survey responses about sexual behavior, which were also noted in reviews of the Kinsey report, will be discussed in the later sections in this appendix. Scope of the Review The committee chose 15 surveys for its review, including some that are national in scope and some that target local populations. Most of these studies were initiated after We AIDS epidemic began in response to the need for population-based arsenates of sexual behaviors known to be associated with HIV transmission. Both telephone and face-to-face interviewing me~ods are represented, along with data collection through self-a~Tninistered questionnaires. There are wide vanations among the surveys In the proportion of questions they devote to measuring sexual practices and other risk behaviors. Four surveys were included because of their potential importance for monitoring the prevalence of HIV infection; these surveys attempted to collect a blood specimen for REV serologic testing from each sample person. The committee used four criteria for including studies in its review: (~) there was at least a minimal attempt to collect data on personal sexual behavior and, In some cases, other HIV risk factors as well; (2) some form of probability sampling was employed; (3) a response rate of He form (number of survey participants)/(number of sample persons) could be calculated; and (4) enough documentation was available to identify the pnncipal characteristics of the survey design and sampling procedures. Information about the designs, sampling procedures, and participation rates of these surveys appears in Table C-~. For the most part, the committee collected information about these surveys from published accounts In books, journal articles, and survey field reports. (The source documents are cited In Table C-~.) Occasion- ally, it was necessary to rely on conversations win survey field managers,

METHODOLOGICAL ISSUES | 217 especially for surveys that had been completed at the time of this writing. ~ several other cases, documentation is partial because of incomplete reporting or recor~keep~ng, or both. For these reasons, and because the total number of surveys is small, the committee has not attempted a statistical analysis of participation rates in relation to survey characteris- tics. Nevertheless, the review does identify differences in response rates In sex surveys that appear to be associated with procedural and design variations. It also reveals several opportunities for learning more about patterns of participation and nonparticipation. Participation in Sex Surveys Data Collection Procedures and Response Rates Each of the surveys listed in Table C-1 asked respondents to report on certain aspects of their past and present sexual behavior. For the most part, the questions used in recent surveys (i.e., those initiated after the AIDS epidemic began) attempt to measure the occurrence of sexual behaviors associated with HIV infection and transmission and fall into three general categories: sexual orientation (with a focus on homosexuality), selection of sexual partners (number and characteristics of partners, presence of same-sex partnerships), and manner of sexual intercourse (e.g., anal, vaginal, oral). Because of the sensitive and highly personal nature of these questions, virtually all of the surveys made some provision to permit respondents to reveal the details of their sexual behavior without undue embarrassment or fear of disclosure to third parties. Most of the surveys included one or more of the following: special guarantees that responses would be kept confidential; assurance of anonymity that is, that the person viewing the results would not know the identity of the respondent; privacy during the interview; and placement of the sensitive questions near the end of the interview. Apart from these similanties, the 15 surveys differ widely with respect to basic methods of data collection and number of questions about sexual behavior. Interviewing was conducted by telephone in four of the surveys (nos. 3, 4, 9, and 10 In Table C-1), by face-to-face interview in five cases (nos. 7, 8, 12, 13, and 14), and by a combination of face-to-face interview and self-administered questionnaire (SAQ) in six (nos. I, 2, 5, 6, Il. and 151. Virtually all of the surveys that contained long, detailed inventories of sexual questions were conducted through face-to-face or telephone interviews. When SAQs were used, the length of the self-a~ninistered forms vaned considerably (some did not exceed 1 or 2 pages whereas others [e.g., survey no. 1] were more than 10 pages long).

218 a' o 0 = Cal :t Cal to ;> m - C~ x a o 3 I: a: ._ CO ._ C) ._ Cal U. o a: o ._ A_ C) V m EM Cal ~ Ct O . z ~ C) Up D ~ A o c,> C~ ~ => ·~- ~ S ~ C '_ z~ 8 . ~ ,. ~ C oo C~ -cL ~o V) ~ U3 C ,8+ 3~" .C G>0 ~, U. ~ O o ._ :O P. C~O ._ c - y .c c os ~ 4_ ~: V2 __ ~ C ~ ~ C~ o4 o ' ~ OO o C) ~ ~o -C~ £ D 0 ~ = . _ t~ O ~ O S =0 ~,to ,~0~ ~ ~ C) S . - ~ _ D ~ ~ D o a~ C ct. ~ ° ~ ° ·3 ~ Dt X ll ~ = 8 ~ R ~ ,E 0 = ~ a ° = aS a ~, ~ ~ ~ y° ~= ,~, t j y C) C~ o C> s ~ <: ~ V' ~ ~,c ° .t ° C) ~ ~ ~.Cs ~ ~ ~ o~ 3 cq ~ E ~ = C) ~ a ~ ~ , ~ 00 ~ ~ _ ~ 8 : E ~ ~ c = o~c.O =0 a 3 ~ ° ~ ~ o ~ ~ ' ~ s o 3 s ·— C) C ._ - o o`, c Z, ~ ~ £ A a ~ , A C l,,, £, s — U3 ·~ o C7\ oo oo ~ ~ ~ o ~ — o ~ ~ + ~ ~ 2 _ o: t, CC ~ s s 3 ~°.5 3 :^ _ oo ~ o _ `.) —o o V C o ° ~ ~ ·S ~.5 3 s ° {_ ~ — ~ s ._ ~ +, oo o ~ _ _ _ ~ ~ o o E oc ~ c~ j

219 U. t) ~ o ~ C 8 =^ .C, ° ~ .O ~ ° ~ £ o _ C c: ° C ~ o E ~ ' -~ E ~ ~ ~ t 11 11 o: t o' ~e t t~ o, ~ ~ o o s ;~ ~ ' _ ~ ._ ~ 3 4, 3 ~ ~ ~ U) ~ o C C 4~ ·_ c,- ~ 3 te ~ 3 4, ~ ~ ._ C C 3 .' ~ Z ~ X ~ ~ ~ ~ .— o ._ ~ >. CO ~ ~ ~ ° ~ ~ ._ ~. 3 =' o .~,c . ~ 4 _ C~ Ce C o ._ ._ ~ ._ ~. 3= o .O s . ~ ~ L' co ~ ,c s o ~ ' ~ °- E ~ C ~ ~ ~ ~ ~ o o 3 3 ~ c '- ~ ~ o o ~D ~o 11 Ct 4) ~D Z~ c. _ — C~ — Z C) o (~! Z C,> 4) ~— ~L~ 3 ~o o ~ Z o ~ C ._ _ ._ C),Z 4I,) o,~ o C) c 3 o ~ C — CL ~ ~ C.} ~ C ._ c ~ z 3 D ~ t 8 E ~ e y ~ ~—~ e ~ 9 5 5 E c ~ E ~ ~ ' + E i_ E , ~ C y e -oo~=E: ,~ yg: E=,- _ 9ee~}sc e=~,= 00 X _ ~ _ — ~ _ o _ ~ ~ 4, 0 ~ t Z t ~ 3 ~ y ~ V·° ~ o- ~ t E

220 · Cq o D e~ g, C C = e ._ ~ ~ ~ 3 C~ ~ .O ·.C ~ ~ C _ ~: . c C, CO c) ~ — c, o ~ ~ G) as a - 5e _4 m 5 ~ . ._ o ~ · C`: ~ c, 3 G) — C) _ ~ ._ CL. _ > ~o U) _ o V) _ .° 04 c5 ~ — (_ ~: o o ~ 3 C~ — a ;~. U. 11 ^ P~: 00 U~ 11 g) o ~ > .— C 3 ~ a~ ·s ~ U. ~ C.} _ ~. 3= o C) s — ;> ~ ~ U. C) C~ ~,c s o~ ~o . C, C~ q) ~ CL ~ ^~= ~—~ o cc Dm ~ S ~ D 4) ca e' ° - ~ O aC C~ O S O0 e~o O ~ C C) 60 ~ >, O ~ C ~ + O O D" X X X — C Z°= ~ ; -c v v ~ ~ - v ~ - u) c) c c) 'e o 4) c~ - ~L' 0 -0 4) ~ cd t ~ co c c c - ~ ,,4 o c) D 4_ ~D C) -O = C ~ C o C~ C~ = 3 = 0 3 E s ~ ~ ~ = z 3 ° e ° ° =, ~ e ~ ~ Y ~", O O a _ ~ ~ ~ ~ ~ > ~ ~ S c~,C C ~ c ,, — ~, 3 0 ~ _ ._ c, ~ e~ ~ ~ ,c c c, c~ ~, 3 0 ~ _ ._ c c'~ .£ . 3~° ° ~ s ~ > 4, U3 ~.= s ~ c ~ ._ ~. 3= 0 c) ~ ~ ~ ~ u, ~ c s° c~ — ~:: 43 _ c _ .= ._ _ ~ ._ c~ ~w ~ ~ ~ c~ u: .3 ~ ~ ~ 0 ~ ~s ~= ~s° ~ .= -O o ,~ ~ ~ O — D S — u' ~C) S — u3 ~ o °D 3 o oo :1 ~ ~ ~ '— O ~ ~ '— O .C & O C) S ~ O ~ S ~ — 4) C O ~ q) ~ c~ 5 ~ V ~ os ~ c5 c~ x _ o~ .c c <5 x _ s 4~) c~ ~ ~ u' v:~ ~ c ~ 4) ~ _ ~ u, == OC4 C C' > o - ._ .D — o - s ~ ~ ~em 4) X ~ ~ _ C 0 ~o x 0 ~ _ _ ~ - ~8 .c 4~ ~ - =^ . _ C) C~q

221 ~ c ~ ~ E ~ ~ -c ~ ~ E 5 o ° E to · _ — · ~ ~ ~ .= C~ o o o ~ o Z 3 ~ ~ ~ ~ ~ ~0 ~ ;` ^ v, D C e ~ e a ^ e ~ E 3 0 8 E c a E c V~3a~v~= ~s~sOa=~=a ~ ~ C) ~ ~L) C.) ~ L J _ ~ L _ ~_ . _ O ~ ~ ~ 1 <5 3 D ~ ~ C: dD ~ D ~' ~ '~ e ~ c E = ,~, ° ~4 =. cc E o.S ~ °.t s '~ 0 .= ~ ~ ~ t: ~ ._ c,) ._ ~ 3— ~ >~ o~~° O ~L) O o .~ ~ O _ t s c.~ 4) — ~ oo ~ — .S ~° ~ ~ ~ s ~ ~ ~v ~ .° E ~,,- ~ = E ° ° .~ ~ ~ ~ ~ ~ ~ s s S ~ D" ~ ._ C O ~ O C: ~ ~ ~ C~ U. ~ ·- ~ 3 a aW-~ S ., <: V a x X 00 0\ C X :3 _ o0 ~ O _ ~ _ ;^ - ~ ~ .= u: c~ ~ ° 8 0 ~ ~ ~ ~ ~ ~ ~ C _ ~5 Cd _ . _ C~ /m W ~ _ ~ ~m <~ .= ~ _ _ C ~ O ._ — e~o — 00 _ — ._ u, O aS — _ c—_ ~ ~ - ^_ a ~ ~ ~ ,_ _ . 0 ~ ~ _ 4) C) _ ~ ._ _ tC C.) C _ C: ~ ~ '~ ~ O, ~ O O _ c ^ X O ~ . ~ =~ . _ _ ~ ~ — m_ _ ~ _ :~: ~ t: C _ _ ~ 3 C~ O _ _ .,, O ~ C Z E: ¢ E ,8 ~ ~ c ~ ~ ~ ~ ~ _ c ~ . _ 0 ~ ^x _ u~ o~ _ 0 5,C) —w ~ ~ ~ _ C~ _ C,) _ C~ ~ ¢5 C — C _ ~ ~ E ~ ~1- 5 ~ cq _= c~ c~ _ w ~^ 0' ;>, ^— q, o~ u' t; w ,_ ~ °\ c~ v x ~ _ ~t c.;~ o~ L> ~ s, _ ¢: — c _ c O ~ ~ ~ , c ~ ,_ ~, _ 0 ~ {_ ~ _ _ — ~ c,~ 00 ~ _ Z ~ ' ~ X m- O ~ .= ~ .' _ u' ~ c c cs, 3 O ~ ~ s ~ ~

222 ~ APPENDIX C The most frequently used data collection procedure was a face-to- face interview followed by a relatively brief SAQ that focused on sexual behavior. In all instances, after the respondent completed the SAQ, it was placed in an envelope (stepped of identifying information except for a serial number, which permitted the questionnaire responses to be linked with interview data), sealed in the presence of the respondent, and collected by the interviewer. In one study (no. 15), the interviewer invited the respondent to accompany him or her to the nearest mailbox from which the questionnaire could be mailed directly to the field office for data entry. In theory, the SAQ sealed envelope procedure should induce greater cooperation (and perhaps more candid answers) by eliminating the need to verbalize responses to sensitive questions. A potential disadvantage is loss of control over the administration of the questions and recording of responses. Literacy or other language problems may prevent the respondent from completing the questionnaire without assistance. In addition, it is easy for a respondent to avoid filling out the questionnaire or to skip questions without refusing directly; on-the-spot checks for item or form nonresponse cannot be conducted without destroying the quasi-anonymous character of the procedure. Figure C-2 displays response rates for 13 of the surveys4 classified by type of data collection: face-to-face interview with SAQ, face-to-face interview only, and telephone interview. The response rates for face-to- face and telephone interviewing were calculated by dividing the number of completed interviews by the total number of sample persons. For surveys that used an SAQ, the response rate is the number of completed SAQs divided by the number of sample persons. In three of the surveys (nos. 4, 10, and 13), the reported response rates should be regarded as upper-bound estimates. Item nonresponse rates were generally low or not reported in the available survey documentation. As noted earlier, the response rate in a study is conventionally used as one yardstick for measuring the potential accuracy of survey estimates. The response rates in Figure C-2 span a wide range, from a low of 33 to a high of 81 percent, win most falling between 50 and 80 percent. Several of the surveys reported levels of participation that compared favorably with rates achieved in surveys dealing with less sensitive issues. The majority of them, however, failed to attain response rates that exceed the usual survey standard of 70 to 80 percent. Given such rates,-it is difficult to rule out the possibility of substantial bias from selective participation in 4Response rates could not be computed for surveys no. 1 and 9.

METHODOLOGICAL ISSUES ~ 223 90 so 70 - a' c' a) — 60 111 G O 50 co IL 30 · 15 · 7 · 6 · 13* · 8 · 11 · 2 · 5 ~ A, Face-to-face/ Face-to-face SAC only TYPE OF DATA COLLECTION · 4* 14 12 · 10* · 3 Phone Interview FIGURE C-2 Response rates in selected surveys (from Table C-l) by type of data collection used. surveys of sexual behavior. It is therefore crucial to document any claims of representativeness by a careful study of patterns of nonresponse. No obvious association between response rate and the three methods of data collection was found, although response rates in the telephone surveys appear to be somewhat lower than those in surveys using other procedures. (The reader should note that the response rates reported for surveys no. 4 and 10 are probably overestimates; for details, see the section "Use of Telephone Surveys" below.) Similarly, additional comparisons (not shown in Figure C-2) indicate Mat there is no correlation between response rates and We scope of Me sampling (local versus national), the size of the sexual behavior component of the interview, or (surprisingly) whether sample persons were asked to donate blood

224 ~ APPENDIX C specimens for serologic testing. Although the variation in response rates appears to be largely unrelated to easily documented characteristics of survey sampling and design in this group of surveys, more specific and detailed comparisons yield some plausible explanations for some of the observed differences. These explanations are discussed below. Survey Configurations Associated with High Response Rates Four of Me surveys listed in Table C-l report response rates higher Wan 70 percent. The two that are local seroprevalence surveys will be discussed later with other surveys Nat include a blood component. The focus here is on two large-scale surveys based on national samples of U.S. residents: Me 1988 General Social Survey (GSS) and the 1988 National Survey of Family Growth (NSFG). These two sun veys differ in substantive focus, definition of the tar- get population, sampling methodology, and details of survey procedure. Nonetheless, they share certain characteristics that may help to explain why they achieved relatively high rates of participation: they are each part of a series of periodic surveys Mat have consistently had good re- sponse rates; the sexual items consist of a relatively small set of "add-on" questions; Me surveys are conducted by the same survey organ~zations5 and to some extent by the same supervisory and field staff that have been responsible for previous surveys in the senes, and Me field procedures include provisions for large numbers of callbacks and special attention to Me conversion of revisers. The 1988 GSS is Me fourteenth in a series of annual opinion surveys conducted by the National Opinion Research Center (NORC) under a grant from the National Science Foundation. Since 1977 the GSS sample has been selected using full probability sampling of adults, IS years and oldster, living in U.S. households. The GSS interview, which is approximately 90 minutes long, is conducted face to face in the home and is sometimes followed by SAQs dealing with special topics. In 198S, GSS interviewers gave each respondent a one-page, self- a~n~stered questionnaire Mat included the following questions: · How many sex pamers have you had In He last 12 monks? · Was one of He pamers your husband or wife or regular sexual partner? NORM for Me General Social Survey and Cycle 1 of the NSFG; Westat for Me NSFG cycles 2, 3, 4.

METHODOLOGICAL ISSUES ~ 225 · If you had over partners, please indicate all categories that apply to them . · close personal friend; · neighbor, coworker, or long-term acquaintance; · casual date or pick-up; · person you paid or person who paid you for sex; · other (specify). Have your sexual partners in the last 12 months been: ex- clusively male, bow male and female, exclusively female? The marked questionnaire was placed In an envelope, seated, and col- lected by the interviewer. The conduct of fieldwork was similar to that of previous GSS efforts. One hundred Dirty interviewers, most of whom had worked on earlier rounds of the survey, were recruited from NORC's interviewer pool and trained in 1988 GSS procedures by supervisory staff. The target response rate for the survey was 77 percent, win a minimum acceptable response rate per sampled areas of 65 percent. NORC employed special procedures to minimize nonresponse. Two hundred preselected respondents living in areas In which recruitment difficulties had been encountered In previous rounds of He GSS were offered $10 to complete the interview and SAQ. addition, NORC assembled a mobile squad of expert interviewers who were assigned to do refusal conversion interviews. In He later stages of fieldwork, field management authorized interviewers to pay respondent fees from $10 to $50 for completion of interviews after an Initial refusal. The field staff completed 1,481 interviews during the 10-week period for a total survey response rate of 77.3 percent. Of those who completed the interview portion of the suney, 93.9 percent returned the SAQ, yielding an SAQ response rate of 72.6 percent. The response rate for fully completed SAQs was somewhat lower (66.7 percent), owing to nonresponse on venous items. The 1988 National Survey of Family Growth (Cycle IV of the series that began in 1973) was conducted for He National Center for Health Statistics by Westat Corporation survey staff from January to August 1988. The target population for this survey consisted of noninstitudon- alized civilian women, aged 15 to 44, living in the United States. The target sample for Cycle IV consisted of more than 10,000 eligible and 6That is, per prumary sampling unit (PSU).

226 ~ APPENDIX C randomly selected women who were listed on household rosters7 from NCHS's National Health Interview Surveys (NHIS)8 conducted between October 1985 and April 1987. About 80 percent of the sample persons were located through address correction requests and telephone contacts before the beginning of fieldwork. After a series of screening questions to verify eligibility, interviewers conducted a one-hour face-to-face home interview that dealt with family planning practices, family size expectations and preferences, sources of family planning services, and aspects of maternal and child health. The interview included several questions about personal sexual behavior (e.g., age at first intercourse, frequency of intercourse, lifetime number of male partners, a series of items on beliefs and attitudes about AIDS, and 19 questions about sexually transmitted diseases. A field staff of 240 interviewers, the majority of whom had at least one year of prior interviewing experience, completed a total of 8,450 interviews by He end of the data collection phase of the survey. This number of completed interviews yielded a total survey response rate of approximately 79 percent (figured by multiplying He 82.1 percent re- sponse rate among NSFG sample persons by a 96 percent enumeration response rates in the NHIS). This rate was achieved after considerable investment by reaching sample persons who were not at home during the initial calls and by converting refusals. The response rate (among NSFG sample persons) after the initial interviewer assignments was ap- proximately 69 percent. It increased to 78 percent after nonrespondents were reassigned and contacted by local interviewers; it increased still further (to 82 percent) in a subsequent foBow-up of a random half of the remaining nonrespondents who were assigned to a special core of conversion interviewers. Nonresponse to specific questions about sexual behavior averaged about 4 percent. Judging by the response rates in these surveys, "piggybacking" a small number of questions about sexual behavior onto established surveys appears to be a feasible strategy for obtaining estimates of He prevalence of certain risk factors for sexual transmission of HIV in general popula- tions. Adding sex-related items to He protocols for the 1988 GSS and 7 The NILS attempted to enumerate all persons living in each household included in its target sample. This was possible for 96 percent of the target households. The names entered on the resulting NHIS "household rosters" provided the basis for the NSFG sampling. 8The NIBS is an annual survey of approximately 50,000 households conducted by the Bureau of the Census. The response rate for dais survey is more than 95 percent. 9The NHIS household rosters were used to sample respondents in the NSFG. Household rosters were missing for approximately 4 percent of the households included in the NHIS target sample.

METHODOLOGICAL ISSUES I 227 NSFG~° had little effect on the response rates normally obtained in those surveys, although item nonresponse may have been slightly higher for the sexual questions than for the nonsexual material. As noted earlier, relatively high rates of participation are typical of these surveys, and the ingredients for such success are well known to survey practitioners: prior experience with similar surveys, continuity of staff, a high "target response rate" combined with a field operation that promotes diligent follow-up of nonresponding sample persons, and an ample budget. Un- der these conditions, response rates for small subsets of sex-related items can be similar to rates achieved in well-conducted surveys that do not inquire about sensitive personal behaviors. The above remarks, however, do not necessanly apply to surveys with a greater number of questions about sexual behavior or with ques- tions of a more intimate nature than those in the GSS and NSFG. In both of these surveys, questions about sexual behavior constituted a relatively minor part of the interview and were not emphasized In pre~nteniew contacts win potential respondents. Furthermore, neither survey asked about the details of sexual encounters, such as the full range of types of sexual contact. Surveys dominated by sexual questions, especially those that include long Kinsey-~e inventories and sexual histones, may encounter more serious problems of nonresponse than are encountered in piggyback surveys. For example, the demands of prior informed consent In true sex surveys win require more complete advance warrung about the nature of the questions to be asked, perhaps increasing respondent con- cerns about embarrassment and disclosure. Whether the relatively high levels of participation that charactenzed the 1988 GSS and NSFG can be achieved in more extensive piggyback surveys or in true sex surveys remains an open question. Use of Telephone Surveys The collection of survey data through telephone interviews has become an increasingly popular alternative to face-to-face interviewing as a result of the generally lower costs of telephone surveys and developments in sampling and interviewing technology (notably, sampling through random-digit dialing fRDD] and computer-assisted telephone inteNiew- }ng CATS. Compansons between face-to-face and telephone interview- ing in the context of national surveys indicate that the overall response rates for telephone surveys are on the order of 5 percent lower than those expected in personal interviews (Groves and Kahn, 19791. Although Lithe NSFG traditionally had some items on sexual behaviors (e.g., age at first intercourse); in 1988 the number of items was increased.

228 ~ APPENDIX C response rates for the two methodologies tend to converge when unan- swered telephone calls are ignored, acceptable standards (e.g., Council of American Survey Research Organizations, 1982) for reporting response rates mandate that a fraction of the numbers not contacted be included in the denominator when calculating response rates for telephone surveys. The generally lower response rates of telephone surveys, along with stud- ies of respondent reactions to the two founts of interviewing, suggest that the rapport and trust between interviewer and respondent attained in face- to-face surveys are harder to achieve in the telephone interview. Thus, it is reasonable to expect somewhat lower rates of participation in tele- phone surveys that contain explicit sexual content, compared with similar surveys using face-to-face interviewing or a combination of face-to-face interviewing and an SAQ. Unfortunately, there is only a limited empirical basis for judging the feasibility of conducting interviews about sexual behavior by telephone. Few such surveys have been conducted (four are listed in Table Cap, and only one of the four reviewed by the committee gives an indication of the utility of telephone interviewing for measuring sexual behavior in general populations. Nevertheless, a brief review of response rates in these surveys is warranted, if only to indicate the character of the attempts Bus far. The two local area telephone surveys (nos. 9 and 10 - one of which was conducted In San Francisco, He other In Seattle are examples of the use of targeted sampling of phone numbers and telephone interviewing to find and interview persons who may be considered to be at higher than average risk of HIV infection. Yet such surveys are not without their drawbacks. Although probability sampling was ostensibly employed, peculiarities in the definition of He target populations and in He execution of He surveys make generalizations about participation In these surveys hazardous. The 1987 San Francisco survey (conducted by Communication Tech- ", . . . 1 i The ratio of eligible to noneligible contacts found among telephone numbers that db answer may be used to deflate the total number included in the denominator. This practice assumes that the proportion of telephone numbers that have no eligible respondents (e.g., business numbers, households without persons meeting screening cntena, etc.) is the same among nonanswers as among those for which responses are obtained. i2As part of their comparison of parallel telephone and personal interview surveys, Groves and Kahn (1979:97-99) asked respondents whether they "felt uneasy" discussing various topics that had been included in parallel telephone and personal interview surveys that were conducted in tandem. Survey topics included finances, health, voting, and political attitudes. Groves and Kahn reported finding that larger proportions of respondents interviewed by telephone said they '~felt uneasy" about each of these topics. The greatest differences were found for questions on income (17.9 percent "uneasy,' in telephone survey versus 15.3 percent in personal interview) and political attitudes (12.1 versus 8.5).

METHODOLOGICAL ISSUES ~ 229 nologies, Inc.) was undertaken to determine levels of high-risk sexual behavior among gay and bisexual men living In San Francisco. It was conceived as a baseline survey, the first in a series that would document changes in risk behavior that might be related to educational campaigns directed at gay and bisexual men. The sampling frame, which con- sisted of male names and linked telephone numbers, was constructed from commercial lists of households that were stratified by census tract. By oversampling telephone numbers from tracts with a large proportion of unmarried males, the researchers composed 24 replicate samples of 500 names and numbers to be released In sequence to interviewers until 500 interviews had been completed with eligible respondents. Eligibil- ity in this case consisted of being male, 18 years of age or older, and self-identifying in the screening interview as a gay or bisexual man. Interviewers made as many as four attempts to contact a household and, after initial questioning to detennine the presence of a male aged 18 or older, attempted to screen for gay or bisexual behavior. Persons who identified themselves as gay or bisexual were asked a series of questions about their sexual behavior. The report of this survey cited a refusal rate of 19.4 percent. It is not clear, however, whether this rate referred to the initial screening for the presence of an adult male or to the subsequent screening for gay or bisexual behavior. Although there is no mention of the number of households that were not contacted or of the eligibility rate per contacted household, the authors hazard the opinion that "Ewlith a sample size of 500, results were protectable to the universe of self- identifying gay and bisexual men in San Francisco within +4.5% at the 95% level of confidence" (Communication Technologies, Inc., 1987:25~. The 1985-1986 Seattle survey was a pilot study of the use of tele- phone interviewing to collect data on sexual behavior In subgroups of the general population who were likely to be at risk of acquiring HIV infection. The target population consisted of persons aged IS to 45 years living in selected localities of Seattle. Based on census data and "local knowledge," the investigators sampled households in census tracts that were considered likely to include large numbers of homosexual and bi- sexual men and heterosexually active persons. Households were selected from reverse telephone directories~3 so that names and phone numbers could be recorded prior to the telephone contact with the household. This procedure was necessary because the investigators required that an advance letter be sent to all persons to be contacted indicating the kinds of questions to be asked and He voluntary nature of participation. Of the 13 Reverse telephone directories are arranged in numerical order by telephone number and give the name and address associated with each listed number.

230 ~ APPENDIX C nearly 3,000 advance letters sent, approximately one-third were returned (as undelivered mail, wrong addressees, or business addresses). Con- tact was made with seventy-two percent of the remaining households; of these, 46 percent contained no eligible respondents, and 16 percent refused to be interviewed prior to the household enumeration. At the end of fieldwork, 389 interviews had been completed. No interviews were attempted with persons who said they had not received the advance letter. The response rate was estimated to be 55.7 percent, based on the assumption that about one In three of the nonresponding households contained a person who was eligible to be interviewed. The two telephone surveys targeted to a broader population include a national and a statewide survey. The national survey, conducted by the Los Angeles Times in July 1987 (no. 3 in Table C-~), included a short series of questions about sexual behavior at the end of a series of opinion questions. The response rate (reported in a secondary analysis of the poll data) was quite low—about 33 percent no doubt because fieldwork was completed in five days and only three callbacks were permitted. Low response rates are not unusual In short-term commercial surveys; nevertheless, they provide little indication of what can be achieved in more rigorously executed telephone surveys. Surprisingly, the analysis of reports of one aspect of sexual behavior in this survey appear quite consistent with those obtained in another survey that achieved a much higher response rate (see He discussion below). The 1987 Califomia survey (no. 4 In Table C-~) was commis- sioned by the California State Depamnent of Health Services to pro- duce statewide estimates of the distribution of HIV risk factors among adults IS years of age and older. The investigators generated a sample of telephone numbers by random-digit dialing with deliberate oversampling of prefixes associated with areas containing high proportions of minori- ties. At first contact, the interviewers attempted to collect information on household composition and to select at random one eligible person to be interviewed. Potential respondents were told in advance about the nature of the questions to be asked. The interviewer then administered a IS-minute series of questions that included screening items pertaining to HIV risk status (e.g., same-gender sexual contact among men, heterosex- ual contacts with multiple partners or partners in a known HIV risk group, use of recreational drugs). Any respondent who reported one or more of these risk factors was asked additional questions about specific risk behaviors. The interviewers were instructed to make up to 12 attempts to contact each sampled telephone number. Initial refusals were reassigned to other interviewers who made further attempts to complete the ~nter- view. At the end of fieldwork, 2,012 persons had been interviewed from

METHODOLOGICAL ISSUES | 231 a total of 2,834 persons known to be eligible a completion rate of 71 percent; Me 29 percent of incomplete interviews included sample person refusals and interview break-offs. Because noncontacts and enumeration nonresponses were ignored, however, 71 percent must be regarded as a rather generous upper bound on the response rate as conventionally calculated. The response rates in these few telephone studies appear to be somewhat lower Wan those obtained in face-to-face interviews of sexual behavior. This conclusion holds true as well for the statewide California survey because the true response rate is undoubtedly lower than the reported completion rate. Experience with sex surveys conducted by telephone is too Innited, however, to determine the levels of participation that can be achieved in such surveys and whether the lower response rates in the available cases are a generic feature of telephone surveys or simply the result of early and somewhat idiosyncratic first attempts. In view of Me substantially lower cost of telephone versus face-to- face surveys, as well as the limited scope of current experience, carefully designed experiments should be encouraged to test the feasibility of this methodology for surveys of sexual behavior in general populations. Be- cause noncontacts are a major component of nonresponse in existing surveys, such experiments should include provision for large numbers of callbacks and extended interview periods. Additional increases in response rates may be achieved through research on the best ways to introduce questions about sexual behavior over the phone, improvements In questionnaire design, and more diligent attempts to complete inter- views after initial refusals. Another area of possible investigation is to compare—at least at the aggregate and possibly at the individual levels— the results of telephone and personal interview surveys of the same target population. Participation in Seroprevalence Surveys If sex surveys are properly executed, they can provide important clues about the potential for sexual transmission of REV infection in general populations. Survey estimates of the distribution of behaviors associated with H[V transmission, combined with epidem~ological findings on the transmission efficiency of REV in sexual contacts and biological aspects of the natural history of HIV infection, are the raw matenal for model- based inferences about the future spread of AIDS In known risk groups and in populations not yet considered to be at risk (May and Anderson, 1987; Anderson and May, 1988; May, Anderson, and Blower, 1989; Miner, Miner, and Moses, l989:Ch. 21. Yet even if the data on sexual behavior in general populations were far more complete than they are at

232 ~ APPENDIX C present, He validity of model-based inferences and predictions would be suspect owing to uncertainties about a variety of other forces that govern the epidemic growth of infection. At best, sex surveys of representative samples can indicate degrees of vulnerability to sexual transmission of HIV in individuals and in population subgroups. In this way they can make a significant contribution to prevention and control. Sex surveys cannot, however, substitute for direct assessment of the prevalence of infection. The most attractive design for direct measurement of the prevalence of infection is the seroprevalence survey.~4 A seroprevaTence survey applies well-established principles of probability sampling and survey methodology to the problem of collecting sample blood specimens with the aim of estimating the prevalence of infection ~ a population. In the- ory, such surveys could emanate most of the outstanding uncertainties regarding the size of the AIDS epidemic, the prevalence of infection In major risk groups, and the degree to which HIV has entered populations that are not presently considered to be at significant risk for infection. Yet, the practical difficulties of mounting a seroprevalence survey on a local or national basis are formidable. Not the least of these are the prob- lems of potentially high levels of noncooperation among sample persons and possible correlations between participation and HIV serostatus. Currently, a national seroprevalence sun vey is in He planning and de- velopment stage under a contract between the National Center for Health Statistics (now a part of CDC) and Research Triangle Institute (RTI). The goal of the first phase of this work, which consists of pilot and pretest sur- veys, is to select a design that meets several objectives simultaneously: protection of respondent anonymity; attainment of participation rates high enough to justify confidence in prevalence estimates derived from the survey; development of procedures to assess nonresponse bias; and identification of cost-effective fieldwork strategies. Choosing an optimum design requires extensive experimentation win alternative combinations Of design characteristics. Among the challenges are determining tile best ways: to introduce the study to sample persons and the general public, to maintain anonymity, to collect blood specimens, to ask questions about risk factors, to compensate respondents for the time required for panici- pation, and to minimize fears of disclosure. Sun vey developers hope Hat preliminary testing will result in a feasible strategy for a national survey of approximately 50,000 households. 14Leng~y discussions of Me meets of seroprevalence surveys and of We problems of implementing Gem on a national scale can be found in loner and Fay (1987/1989) and De Gruttola and Fineberg (1989).

METHODOLOGICAL ISSUES ~ 233 To date, knowledge of what can be accomplished In seroprevalence surveys rests on the results of four local efforts. Only one of these stud- ies, RTI's small-scale pilot seroprevalence survey in Allegheny County, Pennsylvania (which includes the city of Pittsburgh), is part of a program of research to explore alternative designs for a national seroprevalence survey. The other efforts were designed to provide local estimates of the extent of HIV infection In connection with intensive study of the epidemi- ological factors associated with transmission. In all four surveys, sample persons were selected from local, residentially defined target populations through multistaged probability sampling procedures. It will be obvious from the descriptions given below that these sur- veys occupy vastly different positions in the "design space" of potential options for seroprevalence surveys. The RTI Allegheny County (Pitts- burgh) pilot survey (no. 15 in Table C-~) is one extreme a one-time anonymous survey in which blood is collected In the home by ven~punc- ture or finger-stick, accompanied by an SAQ about risk factors. In contrast, the two San Francisco seroprevalence surveys (nos. 12 and 14 in Table Cal) were the first cycles of longitudinal studies that involved periodic collection of blood specimens, extensive personal interviews to obtain risk factor information, and, In one case, routine physical exami- nations in a clinic setting. Not surprisingly, there is a gradient of response rates In these four studies that corresponds roughly to He intensity and duration of the participation required from each respondent: the response rates for blood samples range from 46 to 81 percent. The highest rates were obtained in the least demanding surveys. The San Francisco Men's Health Study (no. 12 in Table C-~) is believed to be the first seroprevalence survey of HIV infection conducted in the United States. The baseline survey, which was designed as the recruitment phase of a longitudinal study, began In the spring of 1984 and continued until April 1985. The target population was defined as currently unmarried men, aged 25 to 54 years, living In the 19 census tracts of central San Francisco that in 1984 had the highest cumulative incidence of AIDS for the city. The investigators anticipated that a majority of the eligible men living in the target area, known locally as the Castro District and considered to be a "gay" area, would have had recent homosexual contact. Within the sample strata (census tracts), the sample was drawn by strict probability sampling at the level of households, and aD eligible men within each household (with no advance screening for type of sexual activity or sexual orientation) were invited to participate In the study. Interviewers made the initial contact in a visit to each selected household during which they attempted to complete a household enumeration, to identify persons eligible for participation, and to schedule

234 ~ APPENDIX C appointments for a visit to a local citric where participants were to be interviewed, given a physical examination, and asked to donate blood and other specimens for laboratory assay. Throughout the 12 months of fieldwork for the baseline survey, there were numerous callbacks for hard-to-reach sample persons, frequent rescheduling of missed clinic appointments, and sustained efforts to convert those who initially refused to participate. At the end of the recruitment phase, 1,034 sample persons had completed their first clinic visit, representing a response rate of 56.2 percent (of approximately 1,839 eligible sample persons, including an estimated 157 eligible men in sample households In which the initial household enumeration could not be completed3. In 1986, CDC conducted a seroprevalence survey in Belle Glade, Florida, as part of an investigation of the causes of an AIDS outbreak in that area (see survey no. 13 in Table C-~. The target population com- prised persons 18 years of age and older and, with parental consent, chil- dren aged 2 to 10 years. Households were selected by stratified random sampling from comprehensive lists for 12 locally defined neighborhoods. Approximately 70 percent of the selected households were located in neighborhoods with the largest numbers of reported AIDS cases in the city. After signing a written informed consent document, sample persons were interviewed at home using a standardized questionnaire; they were then asked for a blood sample (obtained by venipuncture) and examined for signs of HIV infection. The report describing this survey indicates that 557 of the selected households were visited while someone was home, and of these households, 73 percent (N = 407 households) agreed to participate In the study, yielding a total of 877 persons who completed the full study protocol. The 557 households do not include those In which no contact was made; thus, 73 percent should be considered an upper bound of the household response rate. The study report did not indicate the response rate for individual sample persons. A second seroprevalence survey of San Francisco neighborhoods was initiated in 1988-1989, this time to monitor REV infection in areas thought to be likely candidates for transmission through IV drug use or heterosexual contact, or both. The target population was defined as currently unmarried men and women, aged 20 to 44 years, living in three areas (16 census ~acts) characterized by high rates of STDs among women, high rates of admission to drug detoxification programs, and AIDS cases among their residents that were not attributable to male homosexual contacts. Full probability sampling was employed at the is three attempts to contact the household were made at different times of We day (K. G. Castro, Centers for Disease Control, personal communication, May 2, 1990).

METHODOLOGICAL ISSUES ~ 235 household level, and all eligible persons within the selected households were invited to participate In the study. The field protocol included an advance letter to each sample household, signed informed consent, a lengthy personal interview in the home covering HIV risk factors in great detail, and collection of blood (venipuncture In the home or at a local clinic) by interviewers who were certified phIebotom~sts. Participants were paid $20 as compensation for completing the protocol. Repeated callbacks, rescheduling of home visits, and refusal conversion strategies were employed throughout the 10 months of fieldwork. Based on an estimated 2,983 sample persons living In the selected households, Me 1,781 sample persons who completed the interview portion of the protocol constituted an interview response rate of 59.7 percent. The response rate for the blood component of the survey was 46.3 percent because blood specimens were not obtained from 401 of the 1,781 persons interviewed (owing to refusal and, In a few cases, the inability to complete the blood draw). The initial pilot study to test procedures for a national seroprevalence survey was conducted in January 1989 In Allegheny Count, Pennsylva- n~a. As is planned for the national study, the target population for this small-scale pilot consisted of the civilian, non~nstitutionalized population aged IS to 54 years old at the time of the survey. A sample of 539 households was selected by area probability sampling methods, and one eligible adult in each household was randomly chosen as the sample per- son. The fieldwork protocol caned for an enumeration interview (which included questions about the age, sex, race, and marital status of each person in the household), selection of the sample person, signed informed consent (which was left with the respondent at the end of the interview to ensure anonymity), collection of a blood specimen (venipuncture or finger-stick by a phIebotomist who accompanied the interviewers, a $50 incentive payment that was promised before the blood draw, and, finally, completion of a brief SAQ Covering demographic characteristics and ba- sic HIV risk factors) Hat was subsequently placed in a sealed envelope. Videotapes were shown during the contact with the respondent to explain the survey and motivate participation. Earlier, the existence of the survey and its importance had been stressed on all local television news stations. No names were taken at any time during the visit to the household. At the end of fieldwork, 95. ~ percent of He 450 occupied households were successfully screened, and 263 (85.4 percent) of 308 enumerated sample persons completed the survey protocol. The response rate thus was (.951 x .854) 81.2 percent. The Allegheny County pilot study indicates Hat relatively high re- sponse rates can be obtained In seroprevalence surveys Hat involve public

236 ~ APPENDIX C (e.g., TV news) appeals to participate, that make limited demands on par- ticipants' time and that offer substantial monetary incentives. The design of the pilot study stands in sharp contrast to the more epidemiologi- cally oriented surveys that employ lengthy personal interviews covering a much wider range of behavioral risk factors, repeated blood draws (as in the San Francisco longitudinal studies), and invitations to learn antibody status through disclosure counseling. In the Allegheny survey, participation was anonymous: no names were recorded, the risk factor information was obtained by the SAQ/sealed envelope procedure, and there was no feedback of HIV antibody test results to respondents. It is not surprising that response rates in the pilot study were substantially higher than those in more demanding epidemiologically onented surveys. These comparisons indicate the trade-off that must often be made between maximizing response rates in a streamlined design and intensive epidemiological investigation with lower response rates. The discussions Of the design options for a national seroprevalence survey anticipated the necessity of such a trade-off (Turner and Pay, 1987/1989~. Thus, the Allegheny experience suggests that the selection of a survey design that Innits the demands on respondents by making participation relatively easy, anonymous, and nonthreatening—may have been a wise choice. Further testing and refinement of this approach on a larger scale wiB establish whether it constitutes a feasible design for a national survey. Nonresponse Bias in Sex and Seroprevalence Surveys Nonresponse biases occurs when participation in a survey is selective with respect to a characteristic whose distribution is to be estimated Tom survey responses. A high response rate tends to minimize the effects of such selectivity on survey estimates as long as the procedures used to attain it do not in fact increase the correlation between the characteristic of interest and the act of participation. Response rates in most surveys, however, usually are not high enough to justify ignoring problems of selective participation, and there is in fact, a literature that indicates that certain kinds of persons (notably those of low socioeconomic status) are likely to be underrepresented in most samples In ways that may compromise survey estimates (Turner and Martin, 1984:Vol. I, Fig. 3- 1; Goyder, 1987). The committee's review likewise indicates Nat tile response rates achieved In contemporary sex and seroprevalence surveys i6Nonresponse bias is the deviation between the distribution of responses obtained from persons who participated in the survey and who responded to the survey question, and the response that would have been obtained if all persons in the target sample had participated in the survey and answered tile question.

METHODOLOGICAL ISSUES 237 leave ample opportunity for selective participation to affect the validity of survey estimates of sexual behavior and HIV seroprevalence. (The reader should note that response bias that is, misleading or inaccurate survey responses—can have similar effects. This issue is addressed later In this appendix; this section deals only with biases that result from selective nonresponse.) There are three kinds of selective participation that have somewhat different effects on estimates of sexual behavior or H[V infection: (~) selective participation with respect to characteristics that are independent of sexual behavior and HIV serostatus; (2) selective participation related to attributes (e.g., mantal status) Hat may be correlated win sexual behavior or HIV infection; and (3) selective participation Hat is directly related to sexual behavior or serostatus. The first kind of selection can be ignored in the construction of estimates of He sort being considered here because (by definition) this type of nonresponse is unrelated to sexual behavior or infection status. Selection of the second kind does result in biased estimates, but the bias might be remedied if, given the selection factors, the (conditional) distribution of sexual behavior (or serostatus) is known to be the same for respondents and nonrespondents and the distribution of the selection factors among nonrespondents can be ascertained. Let us suppose, for example, that participation in a sex survey is correlated with mantal status in such a way that single men are underrepresented. If the mantal status of male r~onrespondents is known (from, for example, a household enumeration interview) and if there were a good bossism for the belief that, for any particular marital status, the sexual behaviors of respondents and nonrespondents were similar, sample estimates of the distribution of sexual behavior might be adjusted, using imputation or maximum likelihood procedures, to adjust for this nonresponse bias. Pay and colleagues (1989) provided a rather sophisticated example of this form of adjustment In connection with estimates of the frequency of male same-gender sexual behaviors from the Kinsey/NORC national survey conducted in 1970. The most troublesome kind of selectivity is participation that de- pends directly on sexual behavior or serostatus- for example, when He decision to participate is made In relation to fears about reporting socially proscnbed sexual practices or the disclosure of a positive antibody result. ~ this case, simple imputation from observed data will rarely lead to un- biased estimates of prevalence, although in some cases it may be possible to anticipate He direction of bias. Concerns about this form of selection 17Such a basis might, for example, have been provided by a methodological study that did more in- tensive follow-up of a subsample of nonrespondents.

238 ~ APPENDIX C have motivated the development of strategies that make participation In sex and seroprevalence surveys less threatening. Nevertheless, there is likely to be some degree of selection bias attributable to fears about disclosing sexual practices (particularly those considered "deviant") or about revealing one's HIV serostatus In the best executed surveys. Current knowledge about the structure of nonresponse bias In sex and seroprevalence surveys comes from two kinds of compansons: com- pansons of survey estimates with census data and internal analysis of the correlates of different levels of nonresponse. Roughly half of the 15 surveys reviewed here attempted some form of comparison of survey estimates with census data. There is an apparent positive correlation between years of schooling and participation in several of the surveys but few over regularities in the deviations between survey estimates and census figures could be detected. (See Table C-1 for a summary of the comparisons reported in the source documents.) In any case, a good match between census and sample survey distnbutions, although encour- aging in some respects, does not guarantee that estimates of prevalence rates for sexual behavior or HIV infection are unbiased. The other major source of information about nonresponse bias comes from analyzing the characteristics associated with nonresponse at differ- ent stages of a survey interview. Although such stages vary from study to study, a seroprevalence survey might typically include the following: (~) completing a household enumeration form, which includes basic demo- graphic information about each member of the household; (2) completing the personal interview, which might include an SAQ dealing with sexual behavior and over HIV risk factors; (3) completing every item on the SAQ form; and (4) allowing blood to be drawn for serologic testing. Most of the 15 surveys reviewed here involved some degree of staging of this kind, although not always In this order. In many cases, it is possible to study nonresponse at a given stage in terms of information collected at a previous stage- for example, by comparing (in terms of the responses given by both groups In the personal interview) the characteristics of persons who agreed to give a blood specimen with those who refused. There were many opportunities for such comparisons in the surveys examined by the committee, but few of those opportunities had been seized. Smith's (1988) study of nonresponse bias in He 1988 GSS is an important exception. Because the GSS consisted of a lengthy personal interview followed by an SAQ covering sexual risk factors, it was possible to examine correlations between the interview responses and two types of nonresponse: nonresponse to the entire SAQ form and nonresponse to specific items. On He basis of this analysis, Smooth draws

METHODOLOGICAL ISSUES ~ 239 an encouraging conclusion: "In general, the non-response does not appear to be related to differences In sexual behavior. Non-response differentials appear to be absent among those vanables most closely related to sexual behavior. Non-response instead is related to general factors such as low education, Tow political interest, and general uncooperativeness that are not highly related to sexual behavior. As a result, non-response bias to the supplement (i.e., the sex SAQ) appears to be negligible." Smith's conclusion raises an interesting question about the basis for ~nfernng the presence of nonresponse bias. Namely, on what basis can the absence of selection with respect to known or assumed correlates of sexual behavior be taken as evidence of the absence of direct selection (i.e., the third type of selection distinguished above)? There is clearly a need to encourage further analyses of existing sex and seroprevalence surveys to learn more about He structure of nonre- sponse. When surveys are organized in a series of stages, which is the case in many instances, analysis can probe more deeply into the poten- tial cause of nonresponse bias than is possible using simple comparisons of survey estimates with external data. Such analyses should provide a firmer basis for estimating population characteristics In the presence of nonresponse, and it would thereby engender increased confidence in the prevalence estimates derived from sex and seroprevalence surveys and surveys of sexual behavior. NONSAMPLING ISSUES IN AIDS SURVEYS This section focuses on He survey measurement procedures that yield data for basic and applied studies of the behaviors that transmit HIV. The emphasis here is on nonsampling factors that affect the quality of these data. In this regard, it is important to remember that behind every e-way tabulation, logistic regression, or other analytical mode} using such data lies a human encounter between two individuals, an interviewer and a respondent. The situational, cognitive, social, and psychological factors that arise within that interpersonal exchange affect He answers that are given and the data that are thereby generated. To understand the sexual and drug-using behaviors that are at issue in survey research on HIV transmission, one must ultimately confront the uncertainties introduced by this question-and-answer process. Terms and Concepts In discussing the complex array of factors that can distort survey (and i8A mayor aim of the NCHS pilot studies for the national HIV seroprevalence survey is the investiga- tion of He nonresponse bias in surveys that seek to estimate HIV prevalence.

240 ~ APPENDIX C other) measurements, various interrelated terms are used with a reason- ably standardized meaning by statisticians and researchers in the behav- ioral and social sciences. 'rhe constellation of sampling and nonsampling factors that influence a particular measurement define the error struc- ture of that measurement. Sampling factors can include incompleteness or other errors in the sampling frame used to draw the sample, failure to obtain data from all sampled persons, and so forth. (This type of nonresponse is discussed in the preceding section.) Nonsampling factors that affect measurements include misunderstanding of questions, respon- dent unwillingness to reveal sensitive information, interviewer mistakes in reading questions, clerical and other errors made during coding and processing of data, and so forth. Although the factors that affect measurements are diverse, the effects they produce can be divided into two classes: systematic and random. To appreciate this distinction, let us restrict the discussion to phenomena, such as chronological age, for which it seems conceptually simple to sustain the notion of a true value. 19 Although any particular measurement procedure might produce inaccurate readings, most people would agree that there does exist a "true" chronological age for all people. Random errors made in ascertaining age affect the reproducibility (widely called "reliability" in this context) of those measurements. Let us imagine, for example, a situation in which interviewers checked one of two boxes: 0-39 years or 40+ years. Assume further that 50 percent of the population were, in fact, 0-39 years old, and 50 percent were 40+. If interviewers accidentally checked the wrong box for ~ in 100 random persons whose Due age was less than 40, and there was an equal error rate in coding persons whose true age was 40+, the aggregate distribution of ages in the resultant survey data would be identical to the true age distribution for Me population. Nevertheless, 1 percent of all individuals in the sample would have been assigned erroneous ages. Random mea- surement errors of this type produce measurement unreliability, which can be detected in a survey by reinterviewing respondents and asking We relevant questions again. Responses that are inconsistent between the two interviews provide evidence of measurement unreliability. Systematic errors directly affect the validity of a measurement, with- out altering its reliability. Continuing with the example of chronological age, let us assume that, in some survey, interviewers made neither random 19In fact, when the argu~Tlene is pushed far enough, there will always be cases for which the notion of true value becomes difficult to sustain (see Turner and Martin, 1984:97-106). Indeed, even chrono- logical age cannot be Mown with arbitrarily high levels of precision without careful specification of the precise starting point for life. (The time of birthing, for example, does not readily accommodate specification in milliseconds.)

METHODOLOGTCAL ISSUES ~ 241 nor systematic errors but that 10 percent of respondents who were actually 40+ told the interviewer that they were 39. In this case, there would be a systematic error20 in the aggregate data. The estimates generated from the survey would yield 55 percent who were under 40 and 45 percent who were 40+.2i This discrepancy between the true distnbution and the observed distnbution is known as the measurement bias. It should be understood that such biases may occur with perfectly reliable measure- ments. So, for example, the 10 percent of respondents who reported a lower age might well do so again on a reinterview, which would result in perfectly consistent responses in the two interviews: the results would be reproducible, and therefore reliable, but biased nonetheless. To assess bias in measurements, evidence can often be sought from independent sources. In the case of age, for example, public birth records might be compared with the survey reports to detect bias. (This record check might be performed for a small subsample selected for a validation study rather than for the entire sample.) Survey Measurement of Sexual Behaviors Overview this section the committee reviews what is currently known about mea- surement problems that beset surveys of sexual behaviors. Although the survey literature contains much methodological research on nonsampling issues (for reviews, see Bradburn et al., 1979; Rossi, Wnght, and Ander- son, 1983; Turner and Martin, 1984), there is good reason to suspect that the problems encountered In studying sexual behavior are ur~que In some respects.22 There appears to be little that is theoretically unique, however, about the problems of random error and the resultant unreliability In sur- vey measurements of sexual behavior, and there is ample evidence in the literature (see below) that respondents do provide reasonably consistent responses to survey questions on sexual behaviors. When it comes to validation of responses, however, the problems are much more numerous, and they introduce considerable uncertainty Into the interpretation of almost all survey data derived from self-reports of sexual behavior and drug use. Moreover, these problems do not appear to be amenable to quick or easy solutions. Indeed, although the committee 20Systematic error (bias) affects measures of central tendency such as averages and medians, as well as We variability in the distribution of measurements. 21 This again assumes Mat the true distribution was 50 percent under age 40 and 50 percent over 40. 22For a review see Catania and colleagues (199Oa).

242 ~ APPENDIX C later notes a number of reasonably secure inferences that can be drawn from available data and offers several recommendations for improvements In methodology, researchers and other readers of this appendix will find many more questions than answers In the following pages. In addition, these questions will involve the most fundamental matters, such as the issues raised In a question of the sort: Given the observation in a survey that X percent of a (probability) sample reported that they had engaged in unprotected anal intercourse during the previous week, and given the possibility that some respondents may know- ingly or unknowingly distort their responses, what can be inferred about the actual proportion of the population that engaged in this practice? A scrupulous answer to this question might well be that it is uncertain how many persons engaged in this behavior. There are many approaches Mat might help reduce (but could not eliminate) this uncertainty. Careful investigations of respondents' under- standing of a question, for example, could allow an assessment of biases introduced by subjects who did not comprehend We concepts. (There are anecdotal reports, for example, that researchers occasionally find a few heterosexual respondents who confilse rear-ent~y vaginal intercourse win anal intercourse.) Experiments with alternative question wordings might also be conducted to evaluate the extent to which the wording of a question influenced He responses given. Sun~larly, investigators could test alternative questioning procedures Hat afforded the respondent more privacy (e.g., by using self- or computer-a~n~stered questionnaires). Simple, probing questions might be used to gauge the extent of misre- porting.23 In addition, the accuracy of summary responses involving, for example, numbers of partners or rates of sexual contact, could be checked by asking a subsample of respondents to keep dianes. Although each of these tactics provides valuable ancillary informa- tion Hat would reduce some of the uncertainties, none of them provides a completely satisfactory validation of these measurements. There is, for example, good reason (see Table C4 below) to expect that the sensitivity of the topic is more of an obstacle to accurate measurement than the inherent cognitive complexity of the questions (when the time period being recalled is brief) or the unfamiliarity of the terminology. That is to say, questions such as, "Have you had sexual contact to orgasm win another man in the last week?" may be more prone to bias induced by the sensitivity of the topic than by the respondents' inability to understand He question or remember the event. 23Newcomer and Udry (1988), for example, had success with Me straightforward strategy of asking respondents whether they had given false responses.

METHODOLOGICAL ISSUES ~ 243 Furthermore, although there is ample reason and sufficient emp~ncal evidence to justify experimentation with alternative methods of ensuring respondents' privacy, there is also reason to believe that some respondents may conceal behaviors even in the most private reporting setting and that subsequently they will not admit that they did so. This possibility must be given considerable weight in the face of statutes in many states that classify some sexual behaviors (including oral sex, male-female and male-male anal intercourse) as crimes.24 Funny, there is Me possibility that behaviors in which the respondent engages while under the influence of excessive Mugs or alcohol may be poorly recalled, if at an. Given these considerations, lingering concern about the trustworthi- ness of such survey estimates is virtually inevitable. Nonetheless, some inferences may be made with relative certainty if one is prepared to make assumptions about the direction of reporting biases. Inference in the Presence of Bias It could reasonably be argued that the preponderant source of bias In reports of sexual behaviors or Mug use for many (but not all) populations and measurement procedures will be caused by respondents underreport- ~ng their behaviors. It might be hypothesized further that this underreport- ~ng will follow the norms and taboos of the respondents' society. If this hypothesis were true, one would expect the most extreme underreporting for behaviors that are most disparaged and that carry the most severe penalties for discovery (e.g., adult-child sex, rape, etc.~. Similarly, one would expect the least reporting bias for behaviors that are sanctioned or encouraged (e.g., vaginal sex between spouses). (Indeed, In this latter case, one might anticipate an exception to the rule of underreporting. For married couples, it might be reasonable to anticipate some pressure to conceal a low rate or absence of sexual contact between spouses.) For cases In which underreporting is a reasonable assumption, a safe inference from weR-collected survey measurements is that the resultant estimates represent a reasonable lower bound on the true prevalence of Me behavior In the population. Such a tack has been explicitly taken In methodological studies of sexual and other sensitive behaviors. For example, investigators who seek to improve the reporting of sexual be- haviors such as masturbation (Bradburn et al., 1979) and Me reporting of 24~ 1986, the Supreme Court mled that states could enforce Criminal sanctions against consensual homosexual behaviors, even when practiced by adults in the privacy of their own home (Bowers v. lIardwick, No. 85-140, June 30, 1986). The statute at issue (Georgia Code Annotated at 1642, 1984) held that "(a) A person commits the offense of sodomy when he performs or submits to any sexual act involving the sex organs of one person and the mouth or anus of another ...." The argument and opinion in that case implied, however, that although the application of this statute was upheld in the case of male-male sex, it would not be held constitutional if applied to a married male-female couple.

244 ~ APPENDIX C alcohol and drug use (e.g., Waterton and Duffy, 1984) typically assume that the net reporting bias for those behaviors is negative, that is, more persons conceal behaviors In which they have engaged than report behav- iors in which they have not engaged. Consequently, these investigators attempt to identify survey procedures (e.g., the use of SAQs) that would increase reporting of these behaviors fin the aggregate). Similarly, a re- cent attempt to estimate the prevalence of same-gender sex among men in the United States (Fay et al., 1989) argued that the resultant estimates should be treated as lower-bound estimates, on the assumption that the net reporting bias would be negative. Although this strategy does allow lower bound inferences to be made with relative confidence, it is not an entirely satisfying solution because point estimates (e.g., actual frequencies or means) are, after all, of con- siderable interest. Indeed, these estimates figure centrally in two of the most important research challenges of the epidemic's second decade: de- tenrun~ng whether there are declines over time in the incidence of sexual behaviors that risk transmitting HIV, and assessing the effectiveness of AIDS education and prevention programs by comparing the behaviors of persons who participate In those programs with the behaviors of Dose who do not. Both of these tasks require comparison of the distribution of behaviors reported by different samples or by the same sample at differ- ent points in time to detect differences. Changes In behavior over time or the superior effectiveness of particular interventions In changing behavior could then be inferred. If the only statements about these issues that can be made with certainty must be phrased in terms of lower bounds, these important questions cannot be answered. Assumption of Constant Bias in Measurements To answer the questions referred to above, it is neither theoretically necessary to make measurements without bias, nor always required to have accurate assessments of the magnitude and direction of He bias. One can detect change (or intergroup differences) without ever knowing the "true" value, providing it is possible to assume Hat the reporting bias is equivalent at the two points In time (or across the two groups). If the study makes no attempt to validate the self-reports, this assumption is considered to be implicit in studies of time trends or intergroup differences in behavior. This assumption, however, is often problematic. It is quite possible, for example, that the reporting bias itself will vary over time. Respondents (and the population at large) may become more accustomed to questions about sexual behavior aIld may be less likely to conceal sensitive behaviors. (For example, condoms, anal intercourse, and other topics Hat were rarely discussed in the media prior to the AIDS

METHODOLOGICAL ISSUES ~ 245 epidemic now appear with greater frequency.) On the other hand, as the result of an educational program or over intervention effort, respondents may become less likely to report unsafe behaviors. Indeed, it might plausibly be argued that the same social and psychological pressures used by intervention programs to encourage behavioral change also make it less likely that respondents will report these same behaviors.25 Thus, the group receiving an intervention may also have its reporting bias modified In the direction of less complete reporting of risky behaviors.126 Approaches to Validation Measures of sexual behaviors may have much In common with measure- ments of subjective phenomena (e.g., attitudes, options, intentions).27 Although sexual behavior can, in theory, be observed, there are only a few special circumstances in which the testimony of independent observers could be used to validate respondents' self-reports.28 Outside these spe- cial circumstances, Were are no obvious independent measurements that could provide a basis for directly assessing the extent of bias in self- reports of sexual behavior. As discussed below, however, there may be indirect approaches that could be used for such assessments. 25 There is also evidence, however, that in some interventions directed toward IV drug users, increased rapport between clients and program staff over tune results in more honest reporting of risly behaviors. D. Worth, Department of Epidemiology and Social Medicine, Montefiore Medical Center, personal communication, June 19, 1989. 26Similar arguments apply to efforts to study the '`association" between reports of sexual behavior and other variables. Bias in measurements may covary with variables of substantive interest and thus produce spurious correlations. For example, studies of same-gender sexual contact indicate that re- spondents with college educations report more contacts in childhood and as adults than are reported by those with less than a college education. Although this variation may reflect a true difference be- tween groups with different levels of education, it is also possible that college-educated respondents may be less likely than those with no college education to conceal same-gender sexual behaviors in the context of a survey. To the extent that such differential concealment occurred, a spurious correlation would arise between education and the prevalence of same-gender sexual contacts as a result of the nonequivalent measurement biases in the groups. 27 In 1984, another NRC panel published a review of concepts and methods for assessing the error structure of survey measurements of subjective phenomena (loner and Martin, 1984). The report concluded that validation for such measurements might best be sought in demonstrations at an aggre- gate level that iime-series of these measurements were related in a theoretically reasonable fashion to independent time-series that measured objective phenomena to which the subjective phenomena ought to be related. Sufficiently long time-series are not available for most survey measurements of subjective phenomena, but the available data indicate that self-reported attitudes toward the safety of different contraceptives (Beniger, 1984), public perceptions of national "problems" (MacKuen, 1981, 1984), and presidential popularity (MacKuen and Turner, 1984) do show the expected relationships over time. 280ne such circumstance involves the report of sexual contact by a couple in which each respondent is reporting on the same behavior for example, the frequency of coitus or the use of a condom; see A.L. Clark and Wallin [19641 and Levinger [1966] for examples.

246 ~ APPENDIX C Validation Using STD Rates. To assess the validity of reports of sexual behavior, the committee's first report recommended consideration of a strategy parallel to that used with subjective measurements. The committee argued: It may be possible, for example, to construct a convincing validation by demonstrating that an independent series of measurements of change in the incidence of gonorrhea in a population over time could be predicted from a concurrent time-series monitoring the self-reported incidence of unprotected sexual contacts with new partners in the same population. Although there are many potential pitfalls to executing a successful validation in this way, the committee believes that the feasibility of using such indirect procedures should be given further, careful consideration. (Turner, Miller, and Moses, 1989:150) The tentativeness of the committee's advice on this validation strat- egy reflected the practical and theoretical difficulties that attend the use of sexually transmitted disease (STD) statistics for this purpose. The committee noted, for example, that national statistics on STDs may be in need of improvement. It recommended a careful review of these data sys- tems and, if necessary, an increase in resources to improve them (Turner, Miller, and Moses, 1989:167~. Furthermore, for any group, the trends in STD rates over time will reflect phenomena such as the changes, if any, in the behaviors of group members that expose them to infection, trends in the rates of STDs among the population from which the group selects its new sexual contacts, and changes In reporting practices. It is Bus pos- sible to observe in a particular group a rise in STD rates over time that occurs concurrently with a true decline in the rates of "risky" behaviors. This seemingly paradoxical outcome can result from rising STD rates in the population at large, which mask the protective effect of behavioral change in the smaller subgroup. (That is, although "risky behaviors" are less frequent In number, they are practiced with parmers who are more likely to be infected.~29 Nevertheless, despite the attractiveness of the strategy of conducting validation studies of sexual behavior using STD rates, no assurance currently exists that such a strategy could be made to work. Psychometric Approaches to Validity. The approaches discussed above follow from measurement traditions In the physical sciences. Yet there is also a robust tradition of measurement in psychology, which has added to the literature bow a substantial body of empirical work 29 Numerous other artifacts bedevil this strategy as well. For example, it has been observed that the institution of aggressive STD control programs may inflate the number of cases of Shads that are actually reported. Contact tracing may also serve to bring more STD cases to the attention of public health workers.

METHODOLOGICAL ISSUES ~ 247 and, more important, alternative ways of viewing the error structure of measurements for variables, like intelligence, mat are never subject to direct observation. The psychometric literature is vast, and it would be impossible to summarize it here. Excellent texts and reviews are available elsewhere (Gulliksen, 1950; Cronbach and Gleser, 1965; Anastasi, 1976; Wigdor and Gamer, 19821. For the purposes of this report, it may suffice to note the approaches to validation offered in this literature. Most important among these are criterion validity, content validity, and construct validity.30 · Criterion validity for an intelligence test refers to the extent to which a given measurement or test is predictive of the properties it purports to measure. Consider, for example, the Scholastic Aptitude Test, or SAT, which is supposed to measure the test taker's aptitude for college-level course- work. Its criterion validity could be assessed by measuring the correlation between the scores of individuals on the test and some direct indicator of the performance of these individuals In college.3i · Content validity (sometimes called face validity) is not as- sessed empirically. Rather, it involves a judgment that the questions being asked are appropriate (or Unappropriated to the characteristic being measured. Thus, it might be judged that questions that tested reading comprehension and written composition had high content validity (i.e., were "appropn- ate") for a measure that purported to assess the likelihood an individual would succeed in college. In contrast, a measure of the person's athletic prowess might be judged to have low face validity for success in most college programs. · Construct validity refers to the extent to which a set of measurements relate to one another in a coherent way as specified by some (formal or informal) theory. Unlike crite- rion validity, construct validity cannot be reduced to a single correlation coefficient. Construct validity instead involves a judgment about the extent to which a given measurement fits together with other measurements in a manner that is theo- retically meaningful. (~ assessing a questionnaire measure 30The discussion in this paragraph draws heavily on that in Turner and Martin (1984:Vol. 1, 12~125). 3lThis example simplifies matters somewhat. The criterion validity could be directly assessed by admitting all students to the same college program and then measuring their performance. In real- ity, practical considerations make such an uncomplicated assessment impossible (e.g., low-scoring students are not admitted to many colleges and universities). These complications require various adjustments and simplifying assumptions.

248 ~ APPENDIX C of psychological depression, for example, psychologists ex- pect to find a coherent pattern of associations between feel- ings of helplessness and despair, self-destructive fantasy or behavior, somatic complaints such as sleep disturbances, and negative emotions "Hamilton, 1960; Beck et al., 1961; Himme~weit and Turner, 19821.) While these three psychometric notions of validity can be helpful in approaching the problems involved in validating survey measures of sexual behaviors, their past use has sometimes been less Man optimal. One of the most important failings results from overreliance on correlation coefficients In reporting on the validity of measurements. Although ideal self-report measurements of a phenomenon ought to be perfectly correlated with (error-free) direct observations of We same phenomenon, the presence of a very high or even perfect correlation does not, in itself, guarantee that the survey measurement is not contaminated by significant biases. Consider, for example, hypothetical reporting biases that caused the frequency of anal sex to be underreported by 20, 40, and 60 percent from its "true" incidence In Free measurements. Such biases would yield three measurements that differed substantially, but that were, nonetheless, perfectly correlated (r = 1.0) with one another. These correlations would occur even though the mean frequency of anal sex observed using these three measuring instruments would be quite different. If one imagines these different reporting biases to be characteristics of different time periods (or different intervention conditions in an evaluation study), one could conceivably report equivalent "validity coefficients" over time (or experimental conditions), although there were, In fact, substantially different and undetected biases in the measurements.32 It is unfortunately the case that Me current literature contains numerous instances In which researchers report only Me bivanate correlations as evidence of validity (or reliability). Indeed, readers will note In the following pages numerous instances in which the committee has had to rely exclusively on such coefficients because the published reports of studies do not provide other needed information. In this regard, the committee also notes that, when examining the validity and reliability of their measurements, researchers would be well advised to avoid a premature rush to rely on statistical procedures that assume (multivariate) normality. Many measurements of epidemiologi- cal interest have distributions that are not normal. Indeed, there may be strong reasons to focus special concern on biases that disproportionately 32These differences could, of course, be detected by exaniining the means or the intercept of Me re- gression of one measurement on the other.

METHODOLOGICAL ISSUES ~ 249 affect extreme segments of the response distribution. For example, per- sons with very large numbers of sexual and drug-using partners play a disproportionate role in the spread of HIV. It is for this reason that the committee previously recommended that reports of "averages" or tabular displays that hide the long "tails" of the distributions of these variables be avoided whenever possible (Turner, Miner, and Moses, l989:Ch. 2~. For the same reason, researchers who analyze and report on the error structure of they survey measurements should attend not only to the overall performance of their instruments but also to biases and errors that may disproportionately affect the reports of persons who are in the high-activity end of the response distribution. EMPIRICAL STUDIES OF SEXUAL BEHAVIORS Validation There is only a very limited range of evidence that can be collected to provide independent corroboration of the validity of self-reported sexual behaviors. In the past, three broad types of research evidence have been collected: · partner reports, in which regular sexual partners respond to the same questions as the study respondent; · "invalidation evidence," which is derived from longitudinal studies in which it is possible to obtain some measure of reporting accuracy by examining the temporal patterns for impossible temporal sequences (e.g., persons who report having engaged In sexual intercourse when interviewed at age 15 but who report at a later age Hat Hey have never had intercourse); and · clinical evidence of sexual activity that can be used to verify the fact that a particular sexual activity has occurred (the committee is aware of only one study in which this sort of evidence has been gathered). Partner Reports Kinsey and his colleagues reported the first empirical study of which the committee is aware in which the self-reports of two sexual parmers were compared to explore He reports' accuracy. The Kinsey team used 231 pairs of spouses who appear to have been individually ~nteniewed33 33The description of this research (Kinsey et al., 12~128) does not indicate whether the same inters viewer questioned both spouses. The text does indicate, however, that "in many instances there were intervals of two to six years or more between the interviews with the two spouses" (p. 127)

250 ~ APPENDIX C TABLE C-2 Findings of Kinsey et al.'s (1948) validation study of sexual behaviors reported by spouses. cows UNIT ITEMS 0E INVOLVED MEASURF MENT Vital Statistics Years married 1 year Pre-marital acquaint. 12 mon Engagement 4 man Age, 0~ at marr. 1 year AL;e, Q at marr. 1 year No. children 1 child No. abortions 1 event Lapse, first coitu~marr. 6 man. Lapse, marr. —first birth 6 man. Educ. Ievel, ~ 2 years Educ. Ievel, Q Dyers Occup. class, ~ I of 9 Coital Freq. Max. freq., merit. coitus 21wk. Av. freq early marr. 1/wk. Av. freq., now 1/wk. % with orgasm, Q 10% Techniques in coital foreplay Kiss Yes, No Deep kiss Yes, No Hand—~ breast Yes, No Mouth—~ breast Yes, No Hand—Q genitalia Yes, No Hand—~ genitalia Yes, No Mouth—~ genitalia Yes, No Mouth—~ genitalia Yes, No Coital technlq. Male above Yes, No Female above Yes, No On side Yes, No Sitting Yes, kilo Standing Yes, No Rear entrance Yes, No Coitus nude Yes, No Multiple orgasm, ~ Yes, blo _ _ COMPARING DATA FROM 231 PAIRS OF SPOUSES Rs,/~S w~- O IDENT O. /o 88.6 96. 1 57.9 86.9 57.1 78.2 68.6 97.3 61.8 92.5 99.6 100.0 90.3 98.4 74.4 89.4 66.7 89.7 84.1 99.1 79.4 97.8 91.8 98.6 33.2 68.2 34.7 73.3 56.6 88.1 55.0 71.1 97 85 95 89 90 85 82 85 93 76 76 81 ~9 83 90 95 _ COFFFIC OF CORREL. DIFF. OF MEANS: ~ _ ~ MEAN OF MEAN OF HUSBAND S WIFE S REPORTS REPORTS 6.35 ~ 0.42 6.40 ~ 0.42 42.11 ~ 2.83 40.88 ~ 2.74 12.64~ 1.03 12.85 ~ 1.07 27.27 ~ 0.37 27. 17 ~ 0.37 24.88 ~ 0.32 24.75 ~ 0.32 0.90 ~ 0.09 0.90 ~ 0.09 0.32 ~ 0.06 0.41 ~ 0.08 5.09 ~ 0.88 4.72 ~ 0.86 28.05 ~ 1 .99 28. 19 ~ 2.01 16.23 ~ 0.26 16.16 ~ 0.25 4.41 ~ 0.21 14.67 ~ 0.21 5.32 ~ 0. 14 5.27 ~ 0.14 0.54 6.72 ~ 0.31 6.74 ~ 0.31 2.73 ~ 0. 13 3.00 ~ 0.14 l .91 ~ 0.11 2.21 ~ 0.13 69.82 ~ 2.16 66.83 ~ 2.26 % Husbands % Wives ReporiingYes ReportingYcs 95.6 ~ 1.35 99.1 ~ 0.62 85.1 ~ 2.36 82.0 ~ 2.54 95.6 ~ 1.36 96.S ~ 1.22 90.4 ~ 2.04 86.0 ~ 2.43 92.6+ 1.73 92.2= 1.77 +0.4 83.6 ~ 2.46 85.8 ~ 2.32 35.9 ~ 3.23 37.3 ~ 3.26 - 1.4 33.6 ~ 3.14 35.4 ~ 3.18 94.3 ~ 1.54 93.S~ 1.63 54.8 ~ 3.30 49.1 ~ 3.31 39.3 ~ 3.26 37.1 ~ 3.23 19.0 ~ 2.64 17.6 ~ 2.56 10.1 ~ 2.00 7.9 ~ 1.79 24.6 ~ 2.88 17.4 ~ 2.53 87.1 ~ 2.46 89.2 ~ 2.28 4.9~ 1.45 4.1 ~ 1.33 229 214 156 226 228 231 185 227 87 220 223 219 223 225 226 218 229 228 228 228 229 226 290 226 228 228 224 221 227 224 186 2~3 Pears. r 0.99 0.88 0.83 o.ss 0.63 0.99 0.76 0.85 0.96 0.97 0.92 0.98 0.50 0.60 0.75 Tetra- chonc r 0.92 0.72 0.78 0.79 0.61 0.70 0.84 0.93 0.75 0.74 0.68 0.63 0.64 0.17 0.82 0.74 —0.05 +1 .23 _0.21 +0. 10 +0.13 o.oo -o.os +0.37 —0.14 +0.07 -0.26 ~o.os —0.02 —0.27 -0.30 +2.99 —3.S +3.1 -0.9 +4.4 —1.8 +0.8 +5.7 +2.2 +1.4 +2.2 +7.2 —2. 1 +0.8

METHODOLOGICAL ISSUES ~ 251 following the normal procedures used in the Kinsey studies (see Gebhard and Johnson, 1979; Turner, Miller, Moses, 1989:~83~. The results obtained from comparing the reports of each spouse regarding coital frequency and technique are summarized in Table C-2. The table shows- as Kinsey, Pomeroy, and Martin (1948:125) themselves remark "tarn amazing agreement between the statements of the husbands and of the wives In each marriage ..." Given the relatively small sample size, few instances of bias in this study would reach conventional standards of significance. There is, however, a modest trend for females to report less frequently than their male parmers the practice of female superior and rear-entry coitus; females also tended to report a higher average frequency of coitus both in their early marriage and at the time of the interview. In assessing the rather high degree of congruence between the reports of spouses In Kinsey's study, one is also well advised to note the interpretive caveat that Kinsey, Pomeroy, and Martin (1948:125) offered, namely, that "allowance must be made for the possibility that there may have been collusion between some of the parmers, and a conscious or unconscious agreement to distort the factEs]." Following Me example set by Kinsey, subsequent sexual behavior researchers have on occasion gathered data from spouses and other part- ners to gauge the accuracy of self-reported data. With some exceptions, these studies do not cover as comprehensive a range of sexual behaviors as Kinsey explored. They are, nonetheless, perhaps of more relevance for understanding the qualities of the behavioral data that are routinely used In AIDS research. This is so not only because the social norms re- garding sexual activity have changed since the 1940s but also because the methodologies employed in recent studies tend to involve standardized interviews or self-adn~nistered questionnaires rather than Me Kinsey-Wpe interviews that last many hours. Table C-3 summarizes the results of selected recent studies that use spousal reports to gauge the accuracy of reporting of sexual behaviors. One of the first Wings regarding this tabulation Mat will be noted—and lamented is the relative dearth of information on response congruence Mat is typically supplied. Kinsey and colleagues presented the propor- tions of spousal pairs who were in perfect agreement and who agreed within +1 unit; in addition, they supplied correlation coefficients and means and standard deviations for reports by husbands and wives. In contrast, recent studies typically present only correlation coefficients. As noted previously, these coefficients do not provide some of the infor~na- tion that is most crucial in understanding the error structure of measure- ments. They do not, for example, allow any statement about bias for example, for heterosexual couples, the extent to which husbands' reports

252 et x o 0 of .= v 0 o .~ ~5 s m a' x U. o C) 0` PA 0 .= UP o ._ a: .~ - =0 3 U. V m Cat - D ~ ~ ~ S = D o — O o ~ ~ G 3 E X he = lo, o o _ e . ~ e ~ ~ 2 .0 Unto == US O . S ~C~ ~0 ._ O Cal O On ~ ~ ~ ._ 3 ^.o 3 t t ~ c:~.,. ~ ca ~ U. o ~ o .~ E ~ E E E t E oo o ~ ~ o ~ ~ ~ _ ~ _ _ ~ S ~ i ~ i ' ~ | 2 - ~ ~ a ~ . ~ ~ ~ ~ ~ E · ~ 5, E , E E 0 5 e e _ 0 E E,: E— e ~ E ° 3§ ' CC o C~ o~, C~ ~ U. o . . ~fi s~ ~ _ , ·— X C~ C,) o~ ° ~ 4 e ~ E E 0 0 ~ ~ ,~, ,~,, c=> ~ ;; w ~-E 3 ~ ~: ~ ~ _ ~ ~ ~ t~ =, - 5\ o~ ~ ~ ~ ~ 0 ~ _ E ~ E 4 - ~ U Y vl E =~ U

253 Cl: u 'A ° ~ _ 2' ° _ C I -3 e _ ~ g ~ i, ~ ~ ', I' ~^ _ ' ~ ~ ~ L ' Y ~ , 2 I, I _ · ' hi, ~ ~ ~ ·, ~ _ ~ ~ _ ~4~3i ·~1l is A; 3 ~ ~ ~ ~ ~ ~ ~ ~ ~ , D ~ -, _ ~ _ ~ ~ C C-> a: o Cat C) o - 04 .= S o - a: o Cat ._ - . _ CO ·- ~ U. C) Cat a: ~ I3~;~ U _ _ ;> ~ 3 ~ ~ o ~ ~ C) C) ,_ $,, _ C~ ~ oo C~ m- . o ~ U. — ~ ·> ^= -= ~ ~ ~ L. C) $~ 55 4D X · t— U' . c) ~ c> =5 .- ;~ o s 04 ~c~ ~ ~ t: c.) c) . c, ~> 3 — c, c~ X X .O .= ~ ,4 _ C~ . _ . _ _ ~ ~ ~ .= .= ~ ~o ~ ~ =, 'z ~ ~ ~ o

254 ~ APPENDIX C systematically differ from those provided by wives. Overall, the results in Table C-3 together with the Kinsey results suggest that there is substantial but not complete agreement between sexual partners in describing the behaviors In which a couple has engaged. Indeed, In some of the instances reported in Table C-3, the nature of the questions makes the levels of agreement that were obtained somewhat surprising. Coates and coworkers (1988), for example, requested that the gay men in their study report Me total number of sexual encounters they had with a particular partner and the total number of times (or percentage of encounters) during which they engaged in specific sexual behaviors. Given that these men reported a mean of more than 250 encounters with Weir pamers occurring over an average of more than two years, accurate reporting from memory was clearly beyond the ability of most respondents. Nonetheless, the correlational evidence suggests that sexual partners provided quite similar reports of their behaviors. Similarly, Levinger (1966) asked heterosexual married couples to estimate the mean frequency of intercourse during their maniage. (The couples were mamed an average of approximately 14 years.) Although one would prefer some report of the joint distribution of spouses' re- ports, it is impressive nonetheless that the mean monthly frequency of intercourse reported by husbands and wives in this study varied by less than one act per month. B]umstein aIld Schwartz (1983) report parallel evidence from ma~e-ma~e, maie-femaie, and femaie-fema~e couples ~- dicating that reports of the frequency of intercourse over the precedillg year were quite consistent between partners (r = .78 to .86), although somewhat lower levels of consistency were found for questions about other sexual behaviors. Other Validation Techniques Although partner validation provides the most obvious source of in(iepen- (ient information on sexual behavior, it is not the Only validation methocl that has been use`~.34 ~ 1966 I.P. Clark and Tifft reported the use of a polygraph (die detectors to motivate respondents to correct misreports . . _ 340ccasionally, the literature reports studies that use expert3udgments to validate behavioral measure- ments. For example, Koss and Gidycz (1985) reported the use of a self-adniinistered questionnaire to identify "hidden" rapes (i.e., those not reported to police) in samples of college students (242 female and 144 male). Validity was subsequently assessed by comparing questionnaire responses with the results of interviews with a "post-master's level psychologist." Twenty-five percent of survey respon- dents consented to the interview, and the author reported correlations of .73 (female) and .61 (male) when questionnaire data and judgments made by psychologists were aggregated in categories (no vic- timmaiion, sexual coercion, sexual abuse, and sexual assault). Ibe interpretation of these results is problematic, however, because Were is no assurance that the judgments made by the "psychologist" were more accurate than those obtained from the survey questionnaire.

METHODOLOGICAL ISSUES ~ 255 they may have made In a previously completed questionnaire. By mon- itonng the changes made by the respondents, Clark and TifEt obtained a measure of the magnitude and direction of reporting biases that may occur in surveys of sexual behavior. Clark and Tifft's subjects were 45 college males who reported on (lifetime) experiences with prostitutes, sexual coercion, homosexual contacts, masturbation, and nonmantal sex. (It should be emphasized that the criterion measure was not the polygraph judgment but rather the subjects' behavior in "correcting" their answers to the questionnaire when confronted with the polygraph.135 Table C4 presents selected results from this study. It win be seen that for every measurement a substantial fraction of the sexual behaviors that were finally reported were reported only after the possibility of lie detection was introduced. Thus, although virtually every male (95 percent), when confronted with the polygraph, indicated that he had masturbated, 30 percent of the men in the sample denied masturbating on the initial questionnaire and subsequently changed their response. Similarly, although 22.5 percent of these college men ultimately reported some male-male sexual contact, IS percent of the men initially denied such contact and subsequently changed their answers. As these results indicate, the response bias in questionnaire measures can be substantial. Furthermore, the net bias (i.e., the difference between the percentages ovelTeporting and underreporting) is Epically weighted in We direction of underreporting. Emp~ncal evidence such as this underlies the common assumption that survey measurements of sexual and drug use behaviors may set lower bounds on the actual frequency of these behaviors in the population. Udry and Morns (1967) used a different strategy to study a behavior that may have been less sensitive to report—that is, intercourse for married women. These investigators collected daily urine specimens for a period of 90 days and tested the specimens for the presence of 35 Subjects were given multiple opportunities lo change their initial answers. In postsurvey interviews, researchers would ask a respondent for his "utmost cooperation in making his response accurate and [the researchers] strongly alluded to the likelihood of inaccurate responses being given on question- naires administered in group situations" (J.P. Clark and Deft, 1966:519). The respondent was then asked to retrieve his questionnaire using the identification number that had been assigned and "to make whatever modifications (in private) that were necessary to bring it to 100 percent accuracy." Subsequently, subjects were told that the researchers would like them to submit voluntarily to a polygraph examination. (No subjects refused the polygraph.) During the polygraph examination, the researchers reported that "when an indication of deception occurred with one of our respondents, the examiner asked the respondent if he wanted to make a change in his response." The responses supplied by respondents after all "corrections" had been made are taken to be more accurate than those originally supplied in the survey. Although this assumption seems fair, there is, nonetheless, some possibility that respondents may have altered their answers for reasons that would not result in increased accuracy (e.g., to comply with covert social pressure).

256 ~ APPENDIX C TABLE Cal Selected Results Reported in Clark and Tifft's Study of Bias in Question- naire Measurements of "Deviant" Behavior: Percentage of college men (N = 45) who: (a) report selected sexual behaviors where confronted with lie detector, (b) deny behavior in questionnaire but report it when confronted with lie detector (underrepordug), and (c) report behavior in questionnaire but deny it when confronted win lie detector (overreporiing) Estimate of Bias Percent Reporting Behavior With Under- Over- Lie Detector reporting reporting Behavior (a) (b) (c) Had sex relations with a person of the 22.5 15.0 5.0 same sex Masturbated 95.0 30.0 5.0 Attempted to force or forced a female 15.0 7.5 2.5 to have sexual intercourse win me Had sex relations with a person of the 55.0 17.5 15.0 opposite sex (other than my wife) Gotten a female other than my wife 7.5 2.5 2.5 pregnant Visited a house of prostitution 17.5 2.5 2.5 Had in my possession pictures, 50.0 12.5 7.5 books, or other materials which were obviously obscene and prepared to arouse someone sexually SOURCE: J.P. Clark and Ant (1986). sperm to vali~te the women's self-reports of intercourse (which were also obtained daily). Although the absence of sperm does not provide definitive evidence that intercourse had not occurred, Me presence of sperm can be taken as evidence that intercourse had occurred at some point in the recent past.36 Fifteen (of 58) women in this study showed evidence of sperm in their urine on one or more occasions. For 12 of We women, all 32 reports of the presence of sperm in their Wily urines corresponded with prior reports of coitus. Among the other 3 women, there were 5 (of 9) instances in which sperm were present but the subject had not repotted intercourse dunng the preceding 48 hours. Independent evidence validating self-reports is precious. Yet for 36There is some uncertainty about how long spells may be found after intercourse. Ud~y and Morris adopted the standard that the sighting of three or more intact sperm was consistent win a report of coitus during the previous 48-hour period. (In making these judgments, the authors centrifuged the urine specimens and analyzed one drop of the sediment for 10 minutes.)

METHODOLOGICAL ISSUES ~ 257 many studies, obtaining such evidence is impossible (owing to the "pr~- vate" nature of the behavior) or so difficult as to become infeasible. Consequently, other evidence is required to sustain the claim that a par- ticular data collection method is useful for research. Replication of Surveys on Samples of the Same Population In lodging a claim that measurement in a particular domain is "scien- tific," one may legitimately pose the question: Do independent attempts to measure the same phenomenon produce consistent results? Given Me litany of possible contaminants that can corrupt the measurement pro- cess, there is substantial interest in knowing whether similar answers about characteristics of the same population can be obtained by different survey organizations using roughly comparable techniques. Because of the dearth of basic research on sexual behavior in particular and the lack of investment In methodological research in general, there is no ample stock of matenal from which to make such compansons. There are, however, two examples that demonstrate We important point that survey measurements of sexual behaviors can produce reliable measures of behavior In well-defined populations. Proportion of Teenagers Who Are Sexually Active The most thorough demonstration is offered in the literature on teenage sexual activity (an area Hat fortunately received some Investment prior to AIDS to pursue studies to understand patterns of sexual behavior). Kahn, Kalsbeek, and Holders (1988) recently published a comparison of rates of teenage sexual activity reported by the 1959-1963 birth cohort in three independent surveys: · the 1979 national survey of young women undertaken by Zeloik and Kantner (1980) and the Institute for Survey Research of Temple University (KZ79 In Figure Cog; the 1982 National Survey of Family Grown conducted by Westat for the National Center for Health Statistics (NSFG in Figure Cog; and . · the 1983 wave of the National Longitudinal Survey of Youth conducted by the National Opinion Research Cen- ter (NORC, University of Chicago) for a consortium of federal agencies (NLSY in Figure C-3~. The comparison undertaken by Kahn, Kalsbeek, and Hofferth in- volved surveys that had several systematic differences in methodology. The three studies differed, for example, in the precise wording of He

258 ~ APPENDIX C loo LL c, By c: G llJ 50 =) C' KZ79 · NSFG · NLSY _ 12 13 14 15 16 17 AGE 1 1 ,,1 19 FIGURE C-3 Estimates derived from Tree independent surveys of the cumulative percentage of teenagers sexually active, by exact ages. NOTE: Survey estimates have been adjusted to make surveys-as comparable as possible by, for example, eliminating population segments that are not included }n all three surveys. SOURCE: J. R. Kahn, Kalsbeek, and Holders (1988). question, the age of the respondents at He tune they were asked to re- call Weir early sexual experiences, and study design. Adjustments were made (e.g., eliminating population segments that were not included in all surveys) to make Be studies as comparable as possible; the adjusted proportions of women reporting sexual activity by each age are shown in Figure C 3.37 This plot shows that there is a very close correspondence between two of Me three surveys and that He third survey (~e National Longitudinal Survey of Young produces estimates Hat are quite simi- lar for ages 12 to 15 and IS to 19; however, it provides substantially lower estimates of sexual activity for this cohort of women at ages 16 and 17. Although these observed discrepancies are not tnvial and are 37Respondents in the 1979 survey by Zelnik and Kantner (1980) were asked: When was the first time you had intercourse? That is, in what month and year? How old were you at the time? Respondents in the National Survey of Family Growth were asked: Whining back after your first menstrual period, when did you have sexual intercourse for the first time what month and year was that? (If date was not known): How old were you at that time? Was it before your n-th birthday, or after? Respondents in the National Longitudinal Survey were aslced: At what age did you first have sexual intercourse? (Soume: Kahn, Kalsbeek, and Holders, 1988:193, Table 3).

METHODOLOGICAL ISSUES ~ 259 / much larger than would be expected to occur as a result of sampling error alone, one cannot lose sight of He very substantial similarity of the findings obtained by the three surveys. Furthermore, Kahn, Kalsbeek, and Holders (1988) go on to demonstrate that the three surveys produce statistically equivalent estunates of the net associations between sexual activity (i.e., having engaged in intercourse at a given age) and various background vanables. The background variables used in this analysis included the exact year of birth, mother's education, whether the respon- dent was Catholic, and whether the respondent lived with both natural parents at age 14. Of 93 tests for survey-specific differences in the net association between the background variable and sexual activity, in only one case was the estimated discrepancy judged to be different from zero with a p less than 0.05.39 The discrepancies observed among young women aged 16 to 17 are large enough to suggest that naive comparisons of results obtained in different surveys can be misleading if the basis for inference is sampling error alone. It has come to be a well-accepted notion in the literature on survey measurements (as it long has been In analytical measurements In other disciplines) that variability in measurements among procedures, investigators, and laboratories cannot be determined by a simple ex- trapolation of the variances observed In measurements from a single experiment. Rather, different procedures, investigators, and laboratories typically produce measurements that converge on values over repeated measurements that are nevertheless not the grand mean of all measure- ments. Thus it is not unusual to find Hat He effects of minor vanabilities in the particular procedures used by different investigators and research organizations may contaminate individual measurements to some degree. In the case of the three surveys discussed above, however, the overall impact of this effect, although large enough to be statistically detectable, is not so large as to lead to markedly different inferences about the pat- terns of teenage sexual behavior In this cohort of young women. Indeed, it is somewhat comforting to note that the two cross-sectional surveys (by Zebu and Kantner and the NSFG) produced very similar results at all ages. As for the discrepant results of the longitudinal survey, surveys that perform repeated measurements of the same respondent may have many 38The net association was tested by entering as dummy variables the product of the dichotomous variable reflecting the characteristic of interest (e.g., Catholic or not) and dummy variables for the surveys. 39Note that with 100 independent comparisons, one expects random fluctuations to produce five dis- crepancies of this magnitude when the true discrepancy across all tests is zero (and the significance level is set at .05).

260 ~ APPENDIX C characteristics that can affect typical survey measurements. At a mini- mum, the design of the longitudinal survey must not only accommodate nonresponse at the initial interview stage but is also subject to attrition of respondents from the sample over time. Furthermore, the experience of repeated interviewing can create a different interview climate which may influence the results that are obta~ned.40 Number of Sexual Partners Reported by Adults A second example is provided by surveys performed in 1987 by the Los Angeles Times and by the 1988 NORC General Social Survey (GSS). Both surveys asked respondents about the total number of sexual partners they had had dunng the past year.4i The responses to this question yielded a number of interesting results. Of most interest for methodological purposes was the substantial sunilanty of results obtained in the two surveys. This similarity is surprising owing to the following: The quality of survey execution vaned markedly. As noted in Table C-1 Me GSS obtained one of the highest response rates of the surveys reviewed by the committee and We Los Angeles Times survey obtained the lowest. · The surveys differed In the wording of questions, survey content, and mode of a~n~nistration (phone survey versus self-adm~stered questionnaire in a face-to-face survey). · The 12-month time penod did not cover the same dates. (The time periods were approximately July 1986 to June ~987 for the Los Angeles Times survey versus March 1987 to February 1988 for the NORC survey.) Nonetheless, analysis42 of a five-way crosstabulation of the two surveys by using five age groups, gender (male, female), mantal status (married, not married), and number of partners (0, I, 2, 3+) provided surprising evidence of consistency in the survey findings. (This table is presented in the Appendix.) In this five-way tabulation, once the demographic marginals of the two surveys were made comparable,43 it - 40Careful studies of d~e responses of persons inte~viewed in the Bureau of ~e Census Current Popu- lation Survey indicated that employment experience reporting was subject to ariifacts resulting from repeated interviewing (Bailer, 1975; Brooks and Bailar, 1978). 41The following questions were posed to respondents: About how many sexual parmers would you say you have had in the last year? (L as Arlgeles Times); and, How many sex parmers have you had in We last 12 monks? (NORC). 42See liner, Miller, and Moses, 1989:104 108. 43The four-way demographic marginals (age by marital status by gender by survey) differed owing to divergences in the populations included in telephone versus personal interview surveys and to die substantial differences in the quality of the execution of Me sampling plans.

METHODOLOGICAL ISSUES ~ 261 was virtually impossible to discnminate statistically among the results obtained In the two surveys with regard to Me number of sexual parmers in the last year or In the patterns of association found between this variable and marital status, gender, or age.44 Replication of Measurements Using Same Respondents A parallel approach to the replication of entire surveys on new samples from a population is the repeated measurement of a (stable) characteristic of the same respondent, a method that effectively reduces some of Me sources of"noise" In comparisons (e.g., differences that arise from vana- tions In the mesons of survey measurement). Indeed, for characteristics that are unchangeable during the interval between repeated measurements, such as age of first intercourse (for nonvirgin respondents), a survey mea- surement that was perfectly reliable might be expected to produce 100 percent consistency of response between two separate interviews. In re- ality, however, errors of recall, misunderstanding of questions, miscoding of responses, and a host of other factors may cause two survey measure- ments made at two different times to be less than perfectly consistent. This lack of consistency is true not only for questions that might be "sen- sitive" (e.g., those about sexuality), but also for seemingly innocuous inquiries about demographic characteristics. As a stanch component of careful survey practice, it is common for surveyors to recontact a (random) subsample of respondents and to repeat questions that provide crucial measurements. The tabulation of these repeated measurements against the original survey responses provides an index of Me reliability of the survey measurements. That is, it provides a way of assessing the consistency of survey response. As 44Using a stratified jackknife procedure to fit a hierarchical series of log-linear models to the five- way table (Abner, Stiller, and Moses, 1989:10~108), it was found that a model that fit the {PAGM} and {AGMS} marginalS (where P = partners, A = age, M = marital status, G = gender, and S = survey) could be improved slightly by also constraining the model to fit the {PS} matinal (jackknifed likelihood-ratio chi-square for comparison of two alternate models: J2 = lA9, d.f. = 3, p = .051). This improvement is of borderline statistical "significance." An examination of estimates of the A{PS} parameters for this "improved', model indicates that the observed intersurvey discrepancy was largely attributable to minor variations between the surveys in the numbers of persons with no sexual partners and with one paddler. Estimates of the A{PS} parameters (and standard errors) for a log-linear model constrained to fit {PAGM} {SAGM} {PS} marginals were .127 (s.e. =.070) for zero partners; -.123 (.063) for one partner, .035 (.073) for two palavers; and -.040 (.072) for the category three or more partners. (Model parameters are [arbitrarily coded so that positive values indicate an "excess" of NORC cases in the specified category.) Way the effects of weighting and the complex sample design in the range of day= 1.6, significant effects were not folmd for other multivariate parameters involving P and S (e.g., {PAS}, {PGS}, {PMS}, etc.). It should be noted that all model comparisons fit the {SAGM} marginals, which allows for intersurvey differences in the demographic composition of the samples drawn by the two surveys.

262 ~ APPENDIX C noted earlier, consistency alone is no guarantee that the measurements are mearungfu] or valuable. However, survey measures that have zero (or very low consistency) between repeated measurements are inherently suspect (if He trait being measured is stable over timed. Because repeated measurements can be performed within the context of a single survey, there is opportunity for amassing these measurements, and the published literature (as well as unpublished technical documents) provides greater detail Can is available on replicated surveys. Table C-S summarizes some of the relevant results, which indicate substantial levels of consistency between answers to questions about sexual behavior obtained at two different points in time. The observed consistency, however, is not as great as the consistency obtained for some other topics. Saltzman and colleagues (1987), for example, report test-retest reliabilities for two a~n~rustrations of a ques- tionnaire asking about sexual orientation, sexual behaviors, and change in sexual behaviors. Respondents were Il6 gay men participating in an HIV study being conducted at He Fenway Community Health Center In Boston.4s Compared with the test-retest reliabilities of .94 obtained for questions about smoking behavior and .79 for dietary habits, the reliabil- ities for reports on sexual behaviors were in the range of .40 to .99. The lowest reliabilities were found for number of nonsteady partners during last six months (.53) and number of partners for insertive anal intercourse without a condom (.611. Other studies summarized in Table C-5 found test-retest reliabilities in He range of approximately .5 to .9. The retests in several studies assessed behaviors in a different time period than the first measurements (e.g., this past week). As a result, temporal instability in actual behaviors masquerades as "unreliability" in some of these studies. Although the reliability analyses reported in Table C-5 encourage the belief that respondents can report consistently on their sexual behaviors, they do not, in themselves, tell anything directly about the systematic distortions that may occur In reporting. In an enlightening analysis of similar test-retest data, Rodgers, Billy, and Udry (1982) used data from successive waves of a longitudinal survey of adolescents (ages Il to 16) to deduce evidence of invalidity. In this study, 408 adolescents completed two parallel sexual behavior questionnaires at an interval of 12 monks. Rodgers and colleagues reported reliability analyses using items for which different responses at the two points in time are logically Impossible. For example, teens who said they had had intercourse one or more times In their life in the first survey should give the same answer in 45 Respondents with AIDS or AIDS-related complex were excluded.

METHODOLOGICAL ISSUES ~ 263 all subsequent surveys. However, 43 percent of adolescents who reported "ever masturbating" In the first survey reported no such experience In the second survey; for intercourse the comparable figure was 19 percent. As a comparison, a similar analysis was conducted for reports of using alcohol; this effort yielded an inconsistency rate of 6 percent. The authors also reported results for an extremely vague question ("Have you ever groan several inches In height very quickly?"), which produced a 34 percent inconsistent response rate. Across a series of I! behavior measurements, the authors reported that adolescent respondents were most likely to rescind reports (between times ~ and 2) of touching of sex organs and, to a lesser extent, sex- ual intercourse. Less intimate behaviors (hand-holding, kissing) were less likely to be rescinded. For masturbation (in wave I) and sexual intercourse (in waves ~ and 2), reliability was also assessed within each interview by asking questions at the beginning and end of the session. Inconsistent responses within interviews averaged 7.8 percent for inter- course and 8.3 percent for reports of masturbation in He first survey,46 but there was substantial variation by race and sex. For both intercourse and masturbation, the greatest inconsistency in reporting was observed among black males.47 EMPIRICAL STUDIES OF DRUG-USING BEHAVIORS Many of the methodological issues investigators face in studying drug use behaviors are identical to those found in studying sexual or other sensitive or illegal behaviors (see, for example, Siegel and Ballman, 1986; Kaplan et al., 1987; Zich and Temoshok, 1987; Kaplan, 19891. Yet some of the difficulties encountered In studying drug use behaviors are unique to this area of inquiry. Some of the most difficult research challenges arise from He fact that patterns of drug availability and injection behaviors may change rapidly. In the last several years, for example, considerable evi- dence has accumulated that drug-using populations In some communities have changed their needle use practices to avoid HIV infection (Chaisson et al., 1987; Friedman et al., 1987b; Des JarIais, Friedman, and Stoneb- urner, 1988; Des lariats et al., 1988; Robertson, Skidmore, and Roberts, 1988~. These changes have included bow decreased nee~e-shanng and increased use of disinfection regimens for injecting equipment. 46Results for the second survey are reported for intercourse alone, for which 4.3 percent of respondents gave inconsistent responses within the interview. 47 In the first survey, 23.7 percent of black male respondents gave inconsistent responses within the interview to the question on sexual intercourse, and 26 percent gave inconsistent responses to the ques- tions on masturbation. In the second survey, 13 percent of black male respondents gave inconsistent responses to the question on sexual intercourse; no results were reported for masturbation.

264 cO -~- - ca o ~ - m - x U. P¢= ~ - Cat o ._ - :^ - D U-, U) a ~1 ~ ~2 ~ ~ ~ ~E ~ ~ ~ x at,, O ,, c ;O fE _1~g ~ ~ O en, ~ e i,, ~ ~ e _ Y ~ e ~ ~ ~ _ ° s ~ ~ ~ F ~ — ~ 5 'I, 2 2~ ~'1~ i. 't oo X o' on _ _ ~ ~t ~ i o' ~ ~ — ~ ~ ~ E _ C ~ o ~ ~ on cq ~ ~ ~ _ ~

265 ~ ~ . ~ ~ 3 By: ~ ~ ~ = = c ~ O 3 0 ~ ~ TO = ]3 TO too ~ E E ~ O O ~ e ,, ~= 0 = ~ O `,. , . ~ 3 c ~ x E 0 e ~ `~ 0 g ° ~ - E a i, 2 = 0 ~ e -= ~ ~ = ~ ~ ~ e to e Y '~ e ' e O ~ ~ g g ° g g g v a ,- e ~ o E ~- ° ~ , ~ E ~ ° °, ° y ¢, e oo .8 C, ~ ~ . ~ ~ g, u ~ ~ s == ~ ~ _ ~ ~ ~ >~ ° o . - 3 E ~ ~ -3 = -= 3 ~ ~ ~ 3 _ ~ ~ 0 = _ = ~ u~ ~ E— ~v E~ {,, _ C~ ~ ~ _ ~ E~ .— ~ E~ u~ ~ ~ ~ o ~ c) ,_ ~ ~ c¢ c' u, _ o~ ~ 0 os - - sci - - ) y c, ~i e~ o ~o ~ L~ c) e5 ~ - - - ~ - _ g ~' ~ ~ g.g ~ ~ g e, E c~~ ~ c~ 04 c~ ~ ~ u, c, - ce - cq ~ o o ~ _ c., .= c¢ x c`, &s - - 3 x - - u, o v, o c~ u~ ed C~ ~ &~ & c5 3 ~ ~ _ s _ '~0,o C~ ~ ~ ~ _ ~ ~ _ o ~o - - ~5 :~: —o ~o C ~: ~ ~o C) ~ ~ ~ oo C) C~ C~ o ~ C _ o t5 §- ~ .= oo C o _ ~ — U, o o . - ~ _ — X ._ _ c5 _ 5 oO . - ~ ~ o0 ^ C) I_ r~ _ ~ 0\ 0 a-_ V

266 U' . C o C) CL CQ .= U' v m Y ~ _ I ~ y ° 3 y n 3 ~ 3 c ~ -§ a 3 ~ 4 - ~ ~ § ~ E >' c ,, ~ 8 ' ~ c E ~ _ ~, 3 ~ ~ e ~ ~ ~ ~ T ~ ~ ~ ~ ~, 9 ~ E ~ ~q _ ~E c _ ;~ > ~, c . e , ~ y · S ~ · _ _ r C: 4.) X O '~ 3 v c, c~ c. c, ~= oo ~ ~ 4~ t— .~ ~ c.~ c~ v c, Cd ~ 2 ~ ~ ~ ~ ~ o c: — U~ t~ o ~ _ V C) ~ ~ =_ ~ ~ o ~ ~ ~_ x o~ - ._ C) 1"- ~ tC ._ ._ ._ ~ E ~ ~ ~ 3 ~ _ . - ~ — , o _ ~ ~ ~ 6,> ee ~ C.) c5 — — ~ ~ - ~ .> a cr x o~ - - ~ ~> 44 - - ~ - o z ~ C = = E ~ ~o :'s s ~ 0 3 _ cc E 3 ~ °° E E c) a, O ~ ~ ~ 3 D C ~ = X cy = 3 C o ~ ·' ~ o ~ C (~` ~ — C; ~ j ~ ~ E ~,, 0 ~,, e e~ ,5? E c ~ O c c ~ 3 0 x ^=, = ~ ='~ c = E ~ E ~ ~= ~ ~ . e 9 c = c 9 ~ ° ~ v ._ ~ c ~ O c, c, 8 0 0 e ~ 3, y . _ c E C' v Y E Y c 2 Y E—O ~ E " . E y E ~ ° O ~ oo '; ~ ss .~' ·S == =- s 5 8 E c ~ ~ Z r wY ~ 9 .5 8 sc °

METHODOLOGICAL ISSUES ~ 267 As noted in Chapter ~ of the committee's most recent report (Miller, Turner, Moses, 1990), changing patterns of cocaine use complicate this picture. The 1980s saw a dramatic increase in the use of cocaine (Drug Enforcement Administration EDEAl, 1988~; in a related development, trade in "crack" (smokable) cocaine burgeoned from initial reports in 1981 of its availability In 3 cities, to its availability in 47 states and the Distnct of Columbia during 1987 (DEA, 19881. Similarly, some observers48 believe that "ice," a synthetic amphetamine that is smoked, may play a major role In drug use in the United States in coming years. Changes in the illicit drugs in most common use can have important implications for HrV transmission. Studies presented In 1988 and 1989, for example, indicated that the injection of cocaine was associated with elevated rates of HIV infection (Cha~sson et al., 198B, 1989; Des lariats and Friedman, 198Sb; Friedman et al., 1989; Novick et al., 1989~. This association is thought to be the result of high frequencies of injection and decreased needle hygiene over the course of an injection session (Friedman et al., 1989~. Further studies In New York (Friedman et al., 1988) and San Francisco (Fullilove et al., 1989, 1990) have found that smoking crack cocaine, which In itself presents no direct risk of REV transmission, is nonetheless associated with risky sexual behavior. Indeed, cocaine use has been associated with a higher Incidence of syphilis (Minkoff et al., 19891. Such findings suggest that an outbreak of HIV infection among cocaine-using adolescents and adults is a possibility. Accuracy of Self-Reports of Drug Use Behaviors Huang, Watters, and Case (1988) point out Hat the lack of standardized measures, combined win the heterogeneity of the population of drug users and He impaired cognitive functioning that may result from exten- sive drug use, tends to make survey research In this population a difficult task. A drug user who is ~nebnated, for example, cannot be expected to recall needle-sharing episodes accurately. Fur~ennore, the wording of key questions In past AIDS research has sometimes been ambigu- ous. Consider, for example, He following hypothetical experience of an injection drug user:49 Bill selected He syringe that looked least-used from among four in a half- filled water glass. Its action was a little stiff, so he took it apart. He swished the plunger around in He foil packet of a lubricated condom and then dipped die metal needle into a small vial of bleach. Putting the `'works" 48L. Thompson, "Ice: New smokable form of speed," Washington Post, November 21, 1989, A-11. 49The term injection user is used to include bow Pose who inject drugs, like heroin, intravenously and Nose who insert drugs intramuscularly.

268 ~ APPENDIX C back together, he tested it by drawing up water from the glass. Very smooth now. Then he, his wife, and their friend cooked up the heroin. Bill used the syringe first, then his wife used it, but they wouldn't share with the friend, who had his own set. The next morning at the detoxification clinic Bill agreed to be interviewed for a research project. The AIDS researcher asked how often does he use condoms, clean his needles with bleach, what percent of the tune does he share needles? The drug user ~ this vignette could report that he uses condoms frequently (to lubricate his synnges), that he "never" shares needles (only doing so with his wife), and that he cleans his needles "every time" (even though he may subsequently contaminate a clean needle with water shared with other syringes). Furthermore, a survey question about how often BiB shared needles could be answered accurately only if it cIanfied that "sharing" included using needles with his wife. Similarly, most questionnaires about the use of nee~e-cleaning regi- mens are not sensitive enough to distinguish between disinfection that is thought to provide some protection and disinfection that is clearly ~nef- fective. (Wermuth and Ham t1989l, for example, report that some drug users disinfect their syringes by using bleach before and after sessions of sharing them with others but not between use by different persons.) Clarifying some of the ambiguities In survey measurements of these be- haviors may eventually explain why longitudinal studies have failed to demonstrate ~ relationship between needle cleaning and rates of sero- conversion among injection drug users (see, for example, Moss et al., 1989). Measurement Bias As with sexual behaviors, a major problem In measuring the risk behav- iors of Injection drug users is that researchers usually cannot observe the behaviors of interest directly. Occasionally, outreach workers may unob- trusively observe behaviors in such sites as "shooting galleries" (places in which drug users inject drugs together), but in general, drug use is not open to observation by research workers. Consequently, most AIDS research on drug use behaviors has relied heavily on the self-reports of persons who use IV drugs. Response bias, however, is a serious problem in such research. Because of such potential biases, there has been ongoing con- cern among drug use researchers about the accuracy of injection drug users' self-reports. The most frequently asked validity question has been whether subjects underreport their drug use. Several studies have ~nvesti- gated the accuracy of injection drug users' self-reports of their drug use. The results of these studies are quite vaned. Amse] and colleagues (1976),

METHODOLOGICAL ISSUES ~ 269 for example, followed 865 criminally involved drug users and employed various methods to assess the reliability and validity of self-reports of cymbal and ~ug-taking behavior. Overall, their results indicated that, although the self-reports tended to have a downward bias, 74 percent of respondents accurately reported their drug use during the previous four weeks. That is, their self-reports were consistent with urinalysis findings. On another 9 percent of cases, a validity check could not be made ow- ing to incomplete data.) In 14 percent of cases, however, respondents reported no drug use, whereas tests of their urine indicated the presence of drugs. Although these findings are encouraging, it should be noted that urine specimens were provided by only 267 (of a total of 865) respondents. The rates of drug use reported by those who did and did not provide specimens showed little difference,50 but it is possible that, in many cases a respondent's prior inaccurate reporting on the questionnaire may have prompted him or her to refuse to provide a urine sample. If this were, indeed, true, the high rates of consistency observed in this study would not be a good indicator of the overall validity of the self-report data. In addition to using urinalysis as a validity check, Amse] and cowork- ers obtained evidence of reliability in two ways. First, they examined questionnaire responses to detect inconsistent reporting of patterns of Mug use. In only 7 percent of cases were inconsistencies detected in reports of the frequency of drug use and the reported cost of the habit. Inconsistency in the reporting of illegal activities was found in somewhat more cases (13 percent). The investigators also assessed the reliability of their measurements by rea~ninistering a small number of questions in a separate questionnaire two to six weeks later. They found that 97 percent of respondents gave consistent responses over time to the two questions asked about Mug use in this second measurement. In a similarly designed study, Bale and colleagues (1981) interviewed 272 male veterans who were heroin users about their heroin use and compared their statements with a urine sample. The investigators found self-reports to be reasonably accurate In that 84 percent of those denying heroin use in Me previous three months and 78 percent of those claiming no use in Me previous week had urine samples that were negative for opiates. Table C-6 summarizes Me results of 17 over studies reviewed by Magura and colleagues (1987) In which self-reports of drug use were validated by urinalysis. Several studies obtained reports of drug use Mat were as consistent with urinalysis findings as the studies of Me Amsel 500f those providing urine specimens, 41 percent reported no drug use, versus 45 percent of those who did not provide specimens for testing.

270 ~ APPENDIX C (1976) and Bale (1981) research teams. Yet the findings summarized in Table C-6 also show great vanability In the results achieved by different investigators and in the results achieved by the same investigator when obtaining reports of the use of different drugs. Use of heroin and other opiates, for example, is consistently reported with greater accuracy than is the use of other illicit Mugs. Yet even for heroin use, the range of validity results is quite broad (.26 ~ K < .78),5~ and the median level of agreement between uIinalysis results and self-reports is relatively modest (K ~ .51. The range of results improves, however, if the comparison is restncted to studies that assessed heroin or opiate use within seven days of the urinalysis (.S9 _ K < .781. Although the direction of the reporting bias in most studies is toward underreporting of Mug use, the results are not entirely consistent with that interpretation. Focusing on those surveys (marked with an asterisk in Table C-6) that asked about drug use within the period (seven days) for which urinalysis can reliably detect it, one finds instances of major studies In which underreporting biases of up to 13 percentage points were found (McGIothlin, Anglin, and Wilson, 1977; N = 4971. On the other hand, one major study (W. F. Page et al., 1977; N = 896) and one small-scale study (Wish et al., 1983, N= 26) found small biases in the opposite direction (i.e., more respondents reported drug use than were detected In urinalysis). (Such divergences might be attributable to the fallibility of Me urinalysis itself.152 Although the results shown In Table C-6 are "usually interpreted as supporting the validity of addicts' self-reports,"53 they also provide clear evidence of the errors that can affect these measurements. Furthermore, although Me studies are not unanimous In Weir findings, it does appear 51 Coefficient reported is kappa (see Bishop et al., 1975:395). 52It should be recognized that the validity criterion in these studies is also subject to error. In blinded testing of the proficiencies of 50 laboratories performing urinalysis for opiates, cocaine, and other illicit drugs, Davis, Hawks, and Blanke (1988) found that, of 389 urine specimens known to be negative for particular drugs, the laboratories reported five false-positive readings (a false-positive error rate of 1.3 percent). The majority of these false-positive results involved identification of the wrong drug in urine specimens that were positive for some drug; however, two false-positives were given for cannabis and for methadone in samples that were actually drug free. In 350 tests involving urine specimens known to be positive for particular drugs, the laboratory tests had a 31 percent false-negative rate (109 of 350). The substances most often missed were phenylcyclidine (51 percent), morphine (47 percent), cocaine (38 percent), and methamphet~mine (28 percent). The false-positive and false-negative rates obtained in such testing are, of course, a function of analytic method and the cutoff values that are used. Although the above findings provide some indication of the error rates obtained by standard practices in contemporary drug testing by commercial laboratories, it is possible that the procedures used in the validity studies reported in Table C-6 may be different. 53This characterization is offered by Magura and colleagues (1987). who cite the writings of Aiken and LoSciuto (1985) and Harrell (1985) as examples of such interpretations.

271 At: I_ o oo Cat - Cal of Cal so - U. o ._ - Ct - ._ ._ ._ Cal Cal ~o ~Q o - ._ - ~: o C~ C) ._ - C~ o Ct E U. V m ~ 0: Ct & & C5 ; - ~D 04 ._ C~ - .0 _ CD U. CO - C) V, X ~ ~ X ~ _ Ch ~ ~ C~ ~ ~ ~ ~ . . . . . . . ~ ~ ~ ~ O . t_ _ 00 ~ O ~ ~ t_ ~ ~ _ _ _ ~) · · S ~: .~ ~: o U~ _ - ) _ _ O ~ ON ~ ~ ~ ~ x ~ cr~ _ _ C~ ~ ~ U~ ~ _ \0 ~ \0 0N 0\ ~ 0\ 00 00 X _ ~ ~ X ~ 0\ X \0 _ _ _ ~ ~ l ~ ~ u' .S _ .= .= ~ .= .— s o& &-~ == & ~ 2 ~ ° ~ ' 2 ~ E ~o ~ .~ C~2 >t -O _ C) ~ ~ ~ ' ~ ~ E E E E E E t o r~ _ = C O~ X X _ ~ ~ ~ _ ~ ~ o C~ ~ 0 3 4, 0 = ,,, _ ~~; C) 0, ,c E

272 lo ~ C: ~ .c . _ CQ 2 U3 on cut - .E' us ~ so oo oo 0 0 0 ~ us ~ ~ ~ cut 0 — _ 0 0 . . . . . . . . . . . . . . . . . cat 0 0 0 x x ~ 0 ~ in, t_ ~ en us us ~ is ~ cot ~ — ~ _ —) ~ ~ c~ ~ ~ ~ ~ ~ _ ~t _ ~ _ ~ _ ~ u~ x oo oo X oo ~ ~ cr. ~ v~ ~ ~ ~ ~o — c~ d 2 a 0 e .- _ ~ u 2 2 ~ ~ E <~ 2 <~ c ~ 2 = = _ c, a O ~ E E c~ ~ ·O c c x c>- c _ tI5 ~ c ~ c E ° C E E .°, ~ E 2 E ~ w ~ 1 ~ 5 ~ o ~_ o ~ oo _ ~:s ~ _ ~ ~ 0 D ._ 3 <' a: :t

273 _ Us ~ ~ ~ _ . . . . . . Cat — X o _ t— ~ — d of X Go X Cat Cal C> U) ~ o in,, _ ~ · ~ 3 _ S · ~ 3 o ~ ~ ~ ~ ~ or A: Cal o . ~ o ~ ~ o ~ C) o . . Vat Cal ~ C, .~ ~ ~ ~ u hi, He en . ~ ~ ~ ~ j.y ~ ' ~ ' e e ° , .= co 3 .' ~ ~ ~ ~ ° _ ' ~ ~ ^~ ~ ~ D . -= 9 ,~, ~ ~ ~ ~ y8 ~ ~ e s— E Y ~ 5 Y E E ^ ', e ,~,,, 3, _ ~, e ~ y ~, ~ O e . 'E i ,E .~-.~.~9.~-~,.~3!~ ,, ., {.f, ., ~ .~ . 9 . j ~ ~ ~ . ~ ~ 9 . j ~ 9 . . ~ . ~ 9- ~ Z ~ ~ ~ I! ~ ~ — 3 ~ ~ O ~ ~ x D D ~ o ~ ~

274 ~ APPENDIX C that a bias toward underreporting of drug use is quite common. In considering the implications of these results for AIDS research, there are furler reasons to constrain one's optimism about the overall validity of drug use measurements. First, the committee notes that one statistic of considerable interest is the prevalence of drug use in broadly defined populations. Chapter 3 of the committees's most recent report (Miller, Turner, and Moses, 1990), for example, discussed surveys Mat attempted to assess the prevalence of drug use among high school students using the students' self-reports. It is quite possible that surveys of such general populations may be subject to a rather different pattern of bias and errors than that afflicting the surveys included in Table C-6.54 The populations sampled In Table C-6 were all drawn from groups who were publicly identified as Mug users. Because these respondents were already labeled as users of illicit drugs, not only by researchers but by treatment personnel and the criminal justice system, they may have been less motivated to conceal mug use. Furthermore, the routine use of urinalysis In research and in monitoring of treatment may further discourage attempts by this population to hide drug use. Generalizations from these findings to AIDS research are further limited by the subject matter covered in the survey questions. As Roun- saville and coworkers pointed out in 1981, investigators in the drug use field traditionally examined a relatively restncted range of factual data. In AIDS research, however, the knowledge required is somewhat more ex- tensive. Questions of particular interest include not just whether subjects use Mugs but how they use them. How often do they use needles? share needles? clean needles? Little is known about the accuracy of responses to more fine-gra~ned questions such as these, although the accuracy of self-reports of drug use has been studied extensively. A recent study by McLaws and coworkers (1988) found rates of infection with hepatitis B virus that suggested that respondents were actually sharing needles at a rate higher than that reported to researchers. As with self-reports of sexual behaviors, differential reporting bias is a danger in the use of self-reports of drug use behaviors. In evaluation studies, for example, subjects assigned to an ~ntenention (versus a con- trol) group might tend to underreport such behaviors as needle sharung after having gone through an intervention that stressed the importance of ~4 Some surveys of adolescents have attempted to test for ove~reporiing of drug use by asking teenage respondents how often they had used a fictitious drug (e.g., bindro, adrenochromes/wagon wheels, etc.). Negligible percentages (O to 4 percent) of respondents in these surveys reported using these fictitious drugs (see Habennan et al., 1972; Whitehead and Smart, 1972; Petzel, Johnson, and McKillip, 1973; Single, Kandel, and Johnson, 1975; Needle et al., 1983). Yet although this result is encouraging, it does not speak to the more likely bias, that is, underreporting of drug use.

METHODOLOGICAL ISSUES ~ 275 stopping these risky behaviors.55 There is some evidence that reporting biases may sometimes work in the opposite direction. Worth and col- leagues (1989), for example, found that women drug users tned to be "good survey takers." After participating in an AIDS prevention group they tended to give responses indicating more frequent risk taking. The authors attribute this finding to the subjects' greater degree of honesty, re- flecting increased comfort win the questions and the interviewers. Such a differential reporting bias would tend to mask the effects of interventions. SUMMARY OF FINDINGS Although there is ample evidence of error and bias in extant surveys of AIDS risk behaviors, and such evidence should be of concern to investigators, some important conclusions can nevertheless be Dawn from this body of work. Feasibility First, based on the empirical evidence presented In the first section of this appendix, there appears to be little question that such surveys can enlist the cooperation of the vast majority of the American public. Carefully designed surveys inquiring about sexual matters, for example, appear to be capable of obtaining response rates that rival those of commercial and academic surveys on less sensitive topics. Of course, special efforts may be needed to ensure high levels of cooperation, and, as always, careful collection of evidence of response bias is necessary. There appears to be little question, however, that such surveys are feasible as scientific ente~pnses. RepZicability Second, the recent literature contains two demonstrations that indepen- dently conducted national surveys of aspects of sexual behavior (age at first intercourse and number of sexual partners in past year) produced reassuringly similar estimates. This sunilarity was achieved despite van- ations in survey methodology. These results, if repeated across a wider range of measurements, provide salutary evidence that surveys of AIDS risk behaviors can, indeed, provide replicable measurements—Mat is, different investigators using roughly similar methods to survey Me same population cart obtain equivalent results. 55Some of the potential for bias might be eliminated by using research interviewers who are indepen- dent of the staff who deliver the intervention. This tactic, however, should not be expected to eliminate completely the potential for bias.

276 ~ APPENDIX C Validity In most sex surveys, it will be difficult (if not impossible) to obtain con- v~ncing evidence of measurement validity. The committee finds nonethe- less that the research literature contains several important demonstrations of the validity of sexual and UTUg use behavior measures. When couples report independently on factual aspects of their own sexual interactions, for example, there is considerable agreement between the survey reports that are obtained. Similarly, in one instance in which physical evidence could be obtained, it was found that, when questioned In an interview, a very high proportion of married women for whom there was physical evidence of intercourse reported that they had, indeed, had intercourse. Furthermore, studies that have validated self-reported drug use with un- nalysis have found moderate levels of agreement. These results are certainly encouraging, but there is also a variety of other evidence that suggests that some sexual and drug-using behaviors may be considerably underreported in surveys. Although the evidence is limited, it appears that male-male sexual contacts may be massively underreported (at least by college student populations). Reliability There is a fairly large body of research reporting on the consistency of responses over short periods of time in survey reports on venous AIDS risk behaviors. These studies have generally demonstrated moderate levels of response consistency over tone. It is possible, however, that some of the inconsistencies observed in these studies reflect true changes in behavior over time rather than errors in reporting on behavior. IMPROVING VALIDITY AND RELIABILITY The above evidence leads naturally to questions concerning how He reliability and validity of self-report data on these behaviors might be improved. In this regard there are a number of tactics that should be considered. [,iteracy Literacy is an obvious concern when self-a~n~n~stered forms or other written matenals are to be used to collect data. In designing surveys, researchers should be sure that their survey questionnaires (as well as Weir consent forms, Information sheets, etc.) are readable and, In particular, that the difficulty of the materials does not exceed Be reading level of their subjects. ~ cases In which there is a possibility that some of Be

METHODOLOGICAL ISSUES ~ 277 respondents recruited for a study may be illiterate, provisions must be made for detecting this problem and providing an alternate data collection method. In this regard the committee notes that Hochhauser analyzed the readability of AIDS educational materials in 1987 and found that, on average, they required a 14th grade (i.e., college) reading level.56 Such a mismatch in the reading level of educational materials intended for the general public suggests that sensitivity to literacy problems is not widespread. The literacy problem may be even more complex when it is extended to include the research consent forms and information sheets required by institutional review boards. It may In fact be very difficult to make these quasi-legal documents truly readable. One study of the reading level of patient information at a psychiatric institute revealed that all brochures were written below the educational level of 85 percent of their readers but that the patient's consent form was written above the educational level of 77 percent of the patients (Sorensen and Leder, 1978~. Alternatives to Self-Reports In addition to adopting procedures that ensure that respondents can un- derstand the questions they are being asked, it is desirable to supplement self-reports with alternative measures whenever possible. Such mea- sures, which include ethnographic observations, physical evidence, skills demonstrations, and reports of "significant others," can provide ~mpor- tant data on the biases that may affect key measurements. Ethnographic observations, which are discussed at length in the next section, can be a particularly valuable too] In understanding responses to quantitative self-report measures. Lange and colleagues (1988), for example, recently reported that, according to self-reports collected in survey interviews, only 70 percent of IV drug users In New York City had shared needles at some time In their life. Ethnographic studies provided a quick corrective to this inaccurate conclusion. These studies indicated that essentially all IV Mug users In New York had used someone else's Injection equipment at least one time In their life when they first began to inject drugs (Des JarIais and Fne~nan, 198Sc). In this case, ethnographic observations helped cIanfy a limitation of data gathered Dough interview me~ods. Over alternative measures are discussed briefly below. 56Wells and coworkers (1989) found somewhat lower reading levels in a random sample of brochures, cards, inserts (N = 104), and pamphlets, books, and monographs (N = 41 ) selected from Me 1988 AIDS Information Resources Directory Wells et al. reported that a preliminary analysis of 57 brochures found that 72 percent were written at the equivalent of tenth grade or above and 10 percent at a grade level beyond high school.

278 ~ APPENDIX C Physical Evidence A variety of other data can supplement self-reports of drug use. Skin examinations for puncture marks and urine drug screens can be used as cross-checks against self-reports (Sorensen et al., l989c), and low- frequency physiological surrogate markers can sometimes be useful if the population is large enough. For example, Bardoux and colleagues (1989) interpreted a declining incidence of hepatitis B among injection Mug users in Amsterdam as a reflection of decreases in nee~e-shaling and related risk behaviors among that population. Similarly, as noted previously, some researchers have monitored prevalence rates of hepatitis B virus and used these rates as a cross-check on self-reports of nee~e-shaling among their respondents. Skills Demonstrations A further supplement to self-reports are skins demonstrations in which subjects show their ability to practice preventive behaviors. For exam- ple, a test of a prevention program with injection drug users has used demonstrations of the ability to clean needles and use condoms properly as outcome measures (Heitzman et al., 1989~. Such demonstrations can provide useful cross-checks against self-reports. For example, even if a subject reported cleaning needles 100 percent of the time, the preventive value of this "clearing" would be questionable if the subject's demon- stration of needle cleaning skins revealed unfamiliarly with We basic procedures needed to prevent transmission of ~V. Other Safeguards for Surveys Although firm guarantees cannot be made as to the beneficial effects of any particular tactic noted here, the committee believes that Mere is strong presumptive evidence to indicate that a considerably larger invest- ment of resources needs to be made in exploratory work prior to the fielding of major survey investigations.57 (This problem is not confined 57 Also needed are careful methodological investigations of the relative meets of different methods of survey administration, including use of self-administered questionnaires versus interviewer question- ing in surveys of sexual behavior and telephone versus face-t~face surveys. The effects of different data collection methods have not been extensively studied by drug researchers. Magura and colleagues (1987) found with their 248 methadone patients that the age of clients and the type of interviewer di- rectly affected the rate of underreporting. On the other hand, Skinner and Allen (1983) randomly assigned 150 drug treatment clients either to a computerized interview, face-to-face interview, or self- report format and found no important differences in reliability, level of problems, or consumption patterns of alcohol, drug, and tobacco use. Similar results were found by Needle, Jou, and Su (1989), who compared adolescents' reports of dmg use when randomly assigned to report with mailed ques- tionna~res or in-person survey interviews.

METHODOLOGICAL ISSUES ~ 279 to studies of sexual or drug use behaviors; inadequate investment in such exploratory work is common to surveys of other topics according to knowledgeable observers. See, for example, comments by Cannell and coworkers [1989:31.) For surveys of behaviors that risk HIV transmis- sion, this lack of exploratory work is particularly troubling, given the underdeveloped state of research in this field. In this regard, Me com- m~ttee notes that some of the questionnaires it reviewed made impossible demands on the cognitive capacities of respondents, an unfortunate error that would have been detected if the questionnaires had received more thorough pilot testing. For example, one previously cited survey asked respondents to report the total number of sexual encounters they had had with particular partners during the entire length of relationships whose median duration was 24 months. The survey further asked respondents to report the total number of times (or percentage of encounters) dunng which they engaged in one or more of 13 different sexual practices. As is discussed below, there may be good reasons for collecting such data despite the frailties of respondents' recall. Emp~r~cal studies of memory for other events, however, suggest that a respondent's ability to recall such events is limited, and random and systematic errors intrude on the responses. Such studies argue for careful research conducted as ancillary efforts to the epideniiological undertaking that would charac- tenze these errors. Indeed, it might be argued that fin some instances) there would be good cause for restricting recall to time periods in which accurate recall might be assumed with greater certainty. So, for exam- ple, reports on sexual encounters during the preceding one to three days place more manageable demands on respondents' memory (although, even here, perfect recall should not be assumed). Although the arguments for use of different time frames are compli- cated (see below), there is little doubt that researchers' appreciation of the problems that attend key measurements can be sharpened by greater use of exploratory studies prior to the launching of major surveys. Among the techniques that can be profitably employed are ethnographic studies and focus groups whose aim is to explore the frames of reference and language that respondents use in approaching a given topic area; pretests and pilot studies that explore the respondents' understanding of prel~m- inary versions of questionnaires; and cognitive research strategies Mat detail the limits of recall and the strategies respondents use in answering questions that demand recall of events that are not Erectly accessible (e.g., how many sexual encounters have you had with John in the past S8For example, masturbation, inser~ve and receptive oral sex, insertive and receptive anal sex, and so form.

280 ~ APPENDIX C two Hearse. Furthermore, major surveys can embed experimental studies in their designs to assess the effects of key aspects of the research process (e.g., the nature and perceptions of confidentiality guarantees, question wording and context, and the measurement variance and bias introduced by interviewers themseIves). Examples of such techniques are provided below. Randomized Response Techniques A widely known set of tactics for increasing cooperation and accurate reporting of sensitive information In surveys are the randomized response techniques. These techniques are intended to provide an estimate of the distribution of a sensitive characteristic In the population without requiring that individuals reveal sensitive information about themselves. They introduce a random element into the response process so that no individual respondent is definitely identified as admitting to Me sensitive trait. One variant of this technique instructs a random half of the sample to answer "yes" regardless of the question while the other half of the sample is asked to give an accurate answer to the sensitive question. (The randomization might be performed by having the respondent flip a com without letting the interviewer know the result of the coin toss.) Because one-half of the sample would be expected to answer "yes" as a result of the coin toss, the proportion who answered "yes" to the sensitive question can be estimated. (Variations on this basic strategy include one that requests respondents to answer one of two questions—one sensitive and one not—based upon a coin toss or other randomizing device.) Although these techniques are attractive in theory, they do have some drawbacks (Campbell, 1987~. Accordingly, researchers applying randomized response methods should recognize We need for careful pilot testing prior to embarking on a major research effort using them. One drawback is that larger sample sizes are required to obtain estimates of equal precision because the randomization procedure substantially ~n- creases the sampling error.59 Second, although these techniques permit estimation of the univanate distribution of discrete population character- istics (e.g., proportion having same-gender sex In last year); they are not easily adapted for use in estimating continuous vanables (e.g., number of sexual partners).60 Binary, respondents may not understand or follow S9Even where the total sample size is large, use of randomized response techniques may yield estimates for subpopulations (e.g., unmanied males, ages 21-30) that have unacceptably large standard errors. 60See Fox and Tracy (1986:44 '18) for a discussion of the randomizing devices that have been used and the estimation procedures. See Tracy and Fox (1981) for a successful application to estinadng arrest rates.

METHODOLOGICAL ISSUES ~ 28} the instructions used in these techniques, or they may distort their re- sponses despite the theoretical safeguards afforded by the randomization technique. Empirical evidence on the success of these techniques is mixed. Although there have been convincing demonstrations of their ability to increase response rates and to reduce bias in some instances (Goodstadt and Gruson, 1975), the evidence from other studies has been equivocal (e.g., Bradburn et al., 1979:~-13~. Boruch (1989), In reviewing the results of 23 studies using randomized response techniques, concluded that these methods appeared to increase cooperation and decrease bias in about one-half of the cases In which they had been used. Unfortunately, the available evidence does not provide clear guidance as to the conditions under which these techniques will work. Pilot Studies Survey researchers typically distinguish between two types of exploratory studies. Pilot studies are commonly semistructured inqumes conducted prior to the design of the final (or penultimate) version of a survey ques- tionna~re. These studies are used to gather information that is helpful in drafting the instrument. Pretests, on the other hand, are structured tests of a survey questionnaire, one or more of which may be conducted prior to the fielding of a typical survey. Pretests provide information on the range of difficulties Mat may be encountered in a~n~stenug a questionnaire-for example, respondent difficulty in understanding the questions, respondent resistance to providing sensitive information, in- terviewer difficulty in following "skip patterns," and so forth. Although examination of survey questionnaires used in past research on AIDS reveals some deficiencies resulting from We rush to gather data, there have also been some laudable examples of careful preparatory work. The committee notes, for example, that Brian's Social and Community Planning Research unity has included provisions for a three-stage devel- opment effort prior to fielding the British survey of sexual attitudes and behavior. This program (see Lo. Spencer, FauLkner, and Keegan, 1988:3) includes a pilot study63 (consisting of a series of unstructured in-dep~ 61 Skip patterns are instructions to skip blocks of questions that depend on responses to prior questions. For example, a series of questions on first sexual experiences would be skipped for respondents who indicated Hat they were inexperienced. 62Under contract to the British Health Education Authority. 63The work by L. Spencer, Faulkner, and Keegan (1988) testifies to the inconsistent use of the terms pilot study and pretest among survey researchers. In their publication, pilot sly includes testing of structured questionnaires, and the large-scale test of the final questionnaire is called a large-scale pilot study or feasibility test. The preliminary unstructured research is termed an "investigative study" (p. 3).

282 ~ APPENDIX C interviews), the development of structured questionnaires and small-scale pretests, and a large-scale pretest of the final questionnaire. The pilot phase of the research included among its aims exploration of the dimen- sions of the topics that were to be included in the survey, clarification of appropriate language to be used in posing questions, identification of sensitive issues and ways of gathering data on these topics, and pre- lim~nary examination of the impact of interviewer characteristics (e.g., gender, sexual onentation) on the willingness of respondents to discuss their sexual behaviors. These British investigators learned several valuable lessons from their pilot work, which used semistructured interviews with 40 respon- dents. Three of these lessons are reported here, not because they win necessarily generalize to other samples but rather to indicate the impor- tant design considerations that can be missed without careful preparatory work. As a result of debriefing respondents about their reactions to the interviewers to whom they were assigned, the investigators learned the following: Women expressed a strong preference for women interview- ers;64 some women indicated that they would not consent to an interview with a create interviewer, whereas other women said that "they would not have been as open and might have refused to answer some of the more personal questions" Do. Spencer, Faulkner, and Keegan, 1988:10~. · Masturbation, surprisingly, seemed to cause respondents more embarrassment than oral or anal sex. Furthermore, there was considerable variability in the use of this term de- pending on the respondent's sexual orientation. Among re- spondents with predominantly heterosexual histories, mas- turbation was usually understood to refer to self-stimulai~on to orgasm in the absence of a parmer. Among gay-identified men, mutual masturbation was a readily acknowledged be- havior. For women who identified themselves as lesbians, masturbation referred exclusively to self-stimulation; stim- ulation by a partner's hands or fingers was "making love" or"having sex." · There was a strong preference among respondents for use of formal rawer than street language in discussing sexual behaviors. Furthermore, although terms such as penis and vagina were well understood, some of the terminology that 64The evidence reported in this study was based entirely on respondents' statements to female inter- viewers. An the pilot study, male interviewers were never assigned to female respondents.

METHODOLOGICAL ISSUES ~ 283 has become standard In the epidemiological literature was quite foreign to respondents (e.g., vaginal sex), and there was considerable variation in what was inferred from the terms homosexual and bisexual.65 Other terms (e.g., anal sex, oral sex) were widely understood. One cannot, of course, assume that such findings with 40 British adults will generalize to other samples—or, indeed, that one such study fully canvasses the problems involved in surveying the British population. The study does, however, alert researchers to important aspects of the way in which these individuals perceive and taDc about their sexual lives and Me factors that may impede collection of accurate survey data. Pretests Usually, pretests are dress rehearsals for the final survey. As such, Hey can have great value In allowing field staff and investigators to identify any procedural problems inherent in a questionnaire. Similarly, selective debnefing of interviewers and a subset of survey respondents can help identify aspects of the survey that were difficult to am ster or understand. Indeed, simple tactics, such as asking respondents to restate In their own words what was meant by an individual survey question, can be tremendously useful in identifying survey questions Hat do not have equivalent meanings for all respondents (or, equally important, do not have the same meaning for respondents and the investigatory. Although much can be done using these procedures, some ~nvestiga- tors have ~n~oduced more systematic data gathering into the pretest pro- cess. Cannell and colleagues (1989), for example, implemented ancillary data-ga~ering activities awing pretests to provide additional information on interviewer problems with asking He questions as worded, respondent problems of comprehension, and respondent problems with knowing or proving the required information. In brief, Free types of supplemental data were collected during He pretests: 65Thus, the authors (L. Spencer, Faulkner, and Keegan, 1988:24-25) write: "Some people are clear that Homosexual] refers to both men and women who are attracted to and have sex with their own sex. There are, however, a number of people of different ages and social classes who associate the word exclusively with men: 'Homo equals men.' (Woman, 37 years, married) For these people, women who have sex with each other are lesbians, not homosexuals." The Spencer group also reported that although most respondents identified bisexuals as persons who had sex with bow sexes and heterosexuals as persons who had sex with persons of the opposite sex, there were a few notable confusions. One 31-year-old single man, for example, thought bisexual meant "kinly sex, 3 in bed, that sort of thing" (p. 25); some respondents seemed confused by "ordinary people" being labeled at all with regard to sexual orientation and defined heterosexuals as persons who "liked either sex" or their own sex.

284 ~ APPENDIX C · Coding of behaviors of interviewer and respondent: survey interviews were recorded, and relevant aspects of the survey interview were coded to indicate whether the interviewer asked the question as written, whether the respondent asked for clarification, and so forth. Systematic coding of the pretest interviews provides useful data for identifying ques- tions that pose special difficulties for the interviewer or Me respondent, or both, and that may need to be redesigned. · Probes: follow-up questions are used to determine whether the respondents understood the question in the manner that the investigator intended. . Rating of questions by interviewers: after completing their interviews (and after completing trading on the nature of questionnaire problems), interviewers were asked to rate individual questions and to identify the different types of problems they encountered in asking the question. These data, which go beyond those usually collected in pretests, can provide important ancillary infonnation on the extent to which the survey ~nteniew is "standardized": that is, whether questions are asked in identical fashion and have the same meaning across respondents. Cognitive Research Strategies A further preliminary strategy to be considered involves the use of Me findings and techniques of cognitive research. These strategies are par- ticularly appropriate when the task at hand is likely to make substantial demands on a respondent's memory. In that regard the committee notes that two examples used previously may be atypical of the types of information in which AIDS researchers are most interested. Questions regarding chronological age and sexual activity during the past day (or week) solicit information Hat is easily remembered. Most individuals can readily recall their sexual experiences of the previous day, and chronological age is such an important vanable in this culture that the inability to recall it is considered an almost certain sign of dementia. In contrast, many survey researchers ask for information involving matters to which respondents do not have direct and immediate access- for example, the average number of times per week they have had intercourse over the past six months or year. A mistaken estimate in this case may not be the result of "distortion" either conscious or unconscious- but may simply be due to the relative unavailability to immediate memory of the information that has been requested.

t METHODOLOGICAL ISSUES ~ 285 When such information cannot be directly and easily recalled, two factors must be considered. The first factor is the recall strategy used by the respondent. For example, in estimating frequency of sexual activity over a prolonged period, the respondent may concentrate on a particular short time interval (e.g., the last week or two) and then multiply. Al- ternatively, the subject may rely on a recent conversation with someone else about the issue or even on some external benchmark (e.g., "I have always had intercourse at a rate twice the averages I've heard reported so therefore my rate must be . . . "I. There are a number of such strategies, and each results in different types of systematic biases. A "free scan" may yield events or a period that is particularly memorable—for reasons unrelated to the purposes of the question. For example, researchers work- ing on other topics have found that events (e.g., victimizations that have led to some subsequent actions on the part of the respondent (reporting to the police, collecting insurance) are recalled more clearly than those that have not (Wagennar, 19861. (See Bradburn, Rips, and Shevell tI987] for a more detmied discussion of such strategies.) The second factor, as noted by Bartlett In 1932, is that memory is basically a reconstructive process and hence is amenable to influences that are described in the memory literature by such terms as "schemata," "scripts," "good stories," and so on. The reason such effects occur is simply that respondents carry out recall on the basis of the information currently available to them, and just as current expectations and precon- ceptions can affect perception, they can also affect memory—only often more so (see Neisser, 1981; Pearson, Ross, end Dawes, 1989~. The effects of these recall factors may not always constitute a disad- vantage from the perspective of the researcher, but they certainly present problems often enough that they must be considered in designing research protocols and questionnaires. Two studies illustrate this conclusion. In We first, by Smith, Jobe, and Mingay (in press), subjects kept a diary of what they ate over a four-week period and were Ten asked to recall Weir food intake from the most recent two weeks and from the more remote two weeks. The recall for the most recent two-week period more closely matched the actual items eaten during those two weeks (i.e., those In the diary) than the recall for the items eaten during the more remote two weeks. In contrast, the recall for the more remote period matched the previous two weeks and the remote two weeks equally well. The authors concluded that recall for food intake prior to We previous two weeks produced a "generic" result (based on foods that were commonly eaten by the person recalling) whereas people had a more direct recall of the previous two weeks. Even though personal recalB may be subject to social scripts and schemata, such distortions may be unimportant to the

286 ~ APPENDIX C researcher. Indeed, for many purposes, the "generic" result may be of equal or greater interest.66 During the last decade, a number of promising examples of the use of cognitive research techniques In survey development have been pro- posed (Biderman and Moore, 1980; labine et al., 19851. The last year in particular has witnessed publication of several studies that used such tech- n~ques to evaluate and improve survey measurements of heal~-related and other events (Brewer, Dull, and Jobe, 1989; Lessler, Tourangeau, and Salter, 1989; Means et al., 1989; Tucker et al., 1989~. Yet, although these techniques offer researchers new possibilities to "take account" of the effect of the recall strategy and belief factors in assessing responses, such a task is difficult and may present problems for which there are no easy solutions. Clearly, if the phenomena of interest can be restricted to time frames that do not make impossible demands on the accuracy of the respondent's memory, then there is much to be gained by ensur- ing that the phrasing of questions appropnately restricts the time frame, although this kind of restriction may be impossible in some situations. In other situations, there may be no alternative. Some epidemiological research, for example, requires information on patterns of behavior that are sufficiently removed in time to make severe demands on memory. If the information is needed for distant time periods, then researchers have no alternative but to buttress their survey measurements with carefully conducted studies that probe the character of the memory processes that underlie subjects' reports (and misreports). A final strategy that can be used profitably in developing more accurate survey Instruments involves the use of ethnographic research techniques. Not only can these techniques be helpful in preliminary investigations to develop better questionnaires and research designs, but they also provide an important research strategy In their own right for studying questions and populations that may be inaccessible using over research techniques. This appendix concludes with a brief overview of current AIDS research using these techniques. ETHNOGRAPHIC STUDIES Anthropologists often deal win phenomena for which other scientific methods are unsuitable. For example, a descnption of the day's events In a drug shooting gallery or the attempt to understand the meaning of condom use for certain individuals and groups are problems of quite different 66Neisser (1981), for example, found (by comparing John Dean's tes~nony before He discovery of Nixon's White House tapes to the evidence of the tapes) that the "gist" of what happened was main- tained, even though the recall was wrong in almost every detail.

METHODOLOGICAL ISSUES ~ 287 dimensions than determining the HIV status of those same individuals and populations. The anthropologist's task is one of observation and interpretation rather than statistical evaluation or prediction and thus depends more on human powers to ream, understand, and communicate. Although anthropologists consider an understanding and interpretation of the ur~que aspects of social life to be an important part of their research, they also analyze their data for patterns and systematic relationships that can lead to generalizations and theory building. Ethnographic data gathered in the course of fieldwork requires ini~- mate participation in a community and observation of ways of life that often differ from one's own. Long-term participation in the everyday events of a community, neighborhood, or group provides access to de- ta~led information on behaviors, the contexts in which they are enacted, and the vocabulary used to describe them. Participant observation can thus help to identify contradictions between what people say they do and what they actually do. The recording and interpretation of another people's way of life (ethnography) is a process that reveals alternative conceptual frameworks, modes of being, forms of property, and ways of organizing domestic, religious, or political affairs. An appreciation of the diversity of social and cultural forms challenges the aura of naturalness that surrounds the institutions and conventions of life at home and allows one to rethink the basic categories and assumptions of one's own society. Ethnographic research me~ods provide particularly useful tools for gathering information about hard-to-reach populations, for acquainting investigators with the diversity of conceptual frameworks and social forms used to organize and interpret events, and for refining and assessing He appropriateness of questionnaires and over research instruments. Examples of Studies Related to HIV Transmission Male-Male Sexual Contacts Anthropological research among Mexicans and Mexican Americans in southern Califo~a67 provides a telling example of the benefits of an ethnographic approach to data gathenng. Studies of sexual behavior in Mexico indicate Hat Mexican men who engage In same-gender sex have 67By August 1989, 1,151 cases of AIDS had been recorded in Orange County, mainly among homo- sexual (67.8 percent) or bisexual (18.3 percent) men. Although most cases so far have occurred in the Anglo community, in He past two years the number of Latino cases has grown more rapidly than the number of cases in the Anglo community. Mexicans (sojourners who move back and forth across the border) and Mexican Americans (those born in the United States) constitute the largest proportion of tile Latino male population in Califomia (Carrier and Magana, n.d.).

288 ~ APPENDIX C a strong preference for anal intercourse over fellatio. Conceptually, the male playing the receptive role is considered homosexual (by societal standards), but the one playing the insertive role is not, a view that is not shared by men in the Anglo community. Because many Mexicans who engage in same-gender sex may not consider themselves to be homosexual or bisexual, AIDS education programs designed for Anglo gay men may not seem relevant to them (Ca~Tier, 1989; Carrier and Magana, In... Alonso and Koreck (1989) confirmed these findings among rural workers in northern Mexico, adding that similar patterns of sexual behavior may be found among other Latino groups, such as Cubans, in the United States. These researchers caution, however, that other important differences will become apparent once such variables as class and geographic location are considered. Carrier and Magana (not.) similarly note the variability of same- gender sexual behaviors among immigrant Mexican men. Although most immigrant Mexican men continue to engage in behaviors patterned on their prior sexual experiences In Mexico, some adopt mainstream An- gio behaviors. The major determinant of the change appears to be the extent to which their socialization in adolescence was with Mexican or Angio-Arnencan sex partners. The development of appropriate AIDS intervention strategies thus depends on an appreciation of the range, con- text, meaning, and distribution of sexual practices among different ethnic groups within the United States. Targeted educational interventions may also be required for people who are less educated or for those who are preliterate (Carrier, 1989; CalTier and Magana, no... Variation in Drug Use Patterns A second example of how ethnographic research may broaden and en- hance the knowledge base on hard-to-reach populations concerns stereo- typical views of IV Mug users. Anthropological descriptions of the diversity of behaviors, social networks, and self-dist~nctions that exist in different drug-us~ng communities discount the widely held image of the "dope fiend" as a person who devotes his life to acquiring and us- ing drugs. Although IV drug users in a black community in Baltimore (Mason, 1988) can certainly identify "dope fiends among their numbers," they also recognize and distinguish several other types of drug users: "addicts" who pursue drugs constantly but who often have families and jobs and thus participate to some extent in the "straight" world; "hope fiends" who are either unable or unwilling to pursue money to support their habit but who hang around drug

METHODOLOGICAL ISSUES ~ 289 areas waiting for an opportunity to get drugs by hustling or through cunning; · casual users or "chippers" who use IV drugs recreationally on the weekends; and · people who at venous points In their drug-using careers move between these categories. Seen In this light, IV drug users are a diverse and shifting population, and consequently, prevention and outreach services must be tailored to meet a variety of needs and objectives (Mason, 1988~.68 Ethnographic Methods Anthropologists thus investigate, interpret, and attempt to explain cul- tural difference and its changing nature. By emphasizing the diversity of beliefs, practices, and social conditions in different communities, anthro- pologists counter the tendency to see Hispanics, blacks, whites, Asians, Native Americans, men, women, or adolescents in the United States as undifferentiated monocultures. Such an appreciation, however, requires a foundation of quantitative data for better inteIpretanon of the illustra- tive case. Demographic data, usually gathered by the ethnographer using techniques of mapping and census-taking in the study of small-scale so- cieties, are supplemented by census or epidemiological survey data In larger, more complex settings. In recent years, anthropologists have become more sensitive to sam- pling issues as greater theoretical attention has been given to variations In the ideologies and beliefs of different individuals in tribal societies. The question of sampling becomes more acute in large populations in which cultural diversity, social stratification, and rapid change raise greater con- cern regarding methodological precision and die validity and reliability of ethnographic date See Pelto end Pelto, 1978; Bernard, 1988~. The task of ethnographic analysis is further complicated by an awareness that, like data in ah the sciences, anthropological data are not acquired through a pristine encounter with the world. Rawer, observation, recording, and measurement are directed by concepts and theones, and these concepts and ideas are subject to modification and change. Unless ethnographers take what they suppose to be a purely empirical approach to the world, they often have a "double sense" of ~e way Hey go about their work: they assume that their ideas are suitable for interpreting other peoples' 68Koester's (1989b) study of black IV drug users in Denver also challenges the common stereotype of the drug addict. In Denver, Koester found heroin habits that rarely required more than a $50 a day to sustain, and the petty thefts that supported the drug user's habits were often directed at over addicts.

290 ~ APPENDIX C beliefs but that these ideas, like the beliefs they are intended to interpret, are also the products of particular historical and social circumstances.69 The methods used to gather ethnographic data fall into five main categories: · direct observation of daily life on mundane and ceremonial occasions, description of We observed environment (some- t~mes with detailed accounts of commerce or household economies), and the recording of spontaneous conversations in which the ethnographer may or may not be a participant (careful records of speech events provide important matenal for linguistic analysis); · relatively open-ended interviews and discussions with key informants, as well as the recording of life histones; · gathering of information from existing records, "native" texts, songs, visual material, government records, and his- toncal archives; · surveys using structured interviews and involving large numbers of respondents; and . . ~ cat— · an Import type or tearing that receives little attention and involves knowledge not reported in notebooks or on filecards comes from long-term residence in me study com- munity. Memories of experiences, as well as perhaps un- conscious recollections and understandings (associated win certain sights, sounds, and smells) often form the backdrop for the patterns and connections that are made during inter- pretabons and analysis of field data. Much ethnographic data collection depends on developing relation- ships of trust with those whose lives are the subject of study. Such relationships may develop over an extended period of time, and to some degree they determine the accuracy, sensitivity, and complexity of the data. The researcher has an ethical responsibility to ensure that identities are protected and Mat the study causes no harm, an objective that the subjects of the study often monitor and probe. AIDS research presents a special methodological challenge as anthro- polog~sts investigate the worlds of men and women who have bisexual and same-gender sexual relations or who may be IV drug users and their sex parkers, prostitutes and their clients, male hustlers, prison inmates, - 69Analysis of anthropological data is tempered by an awareness that knowledge in various societies is distributed and controlled in different ways. Anthropologists flus see cultures not as fixed entities but rather as the products of continual processes of creation and contest.

METHODOLOGICAL ISSUES ~ 291 and undocumented laborers. Some ethnographic studies have looked at heady care workers, insurance companies, or students and staff In schools, but most research concerns populations that are seen as "marg~nalized"— outside the mainstream, often impoverished, and involved in activities that are illegal or that are seen as deviant. Methods of ethnographic or anthropological fieldwork developed In other contexts may not be well suited to studies of sexual behaviors and IV drug use, the behaviors that need to be understood and changed in order to stem the epidemic. Bow of these areas of study Bus present a challenge to accepted notions of participant observation. Ethnographic Methods in AIDS Research The difficulties encountered In AIDS research have elicited a variety of methodological responses from Be anthropological community. Some ethnographers continue to undertake something close to what is thought of as classical fieldwork. Working alone, they establish rapport and trust with the subjects of study and observe incidents in natural settings. Outside the United States, in Haiti and Brazil, for example, the researcher "lives" In the field, participating extensively in community affairs (Parker, 1987; Farmer and Kleinman, 1989; Farmer, 1990~. In the United States, on the other hand, investigators tend to follow an approach often used In urban fieldwork, that is, "visiting the field" more or less on a daily basis (Leonard, 1990; Connors, 1989; Sterk, 1989~. To overcome the difficulties of research win contemporary Mug users, for example, the ethnographer may belong to a team that includes former drug users as outreach workers. As a participant observer, the e~nographer documents the kinds of questions, attitudes, and theories the former drug users express, training them to do the same in their own daily reports. In turn, the ethnographer learns about drug use from these outreach workers, and their presence facilitates his or her acceptance as an outsider asking questions on sensitive issues (Mason, 1988; Weibel, 1988~. The use of a field station storefront appears to provide a context In which surprisingly sensitive information can be gathered. The pro- toco] of an AIDS intervention project for the sex partners of IV drug users In the predominantly black south side of Chicago, for example, began with a preliminary questionnaire (to gather sociodemographic and epidemiological data), followed by a longer, open-ended conversation that reconstructed Be routines of everyday life and some aspects of the respondent's life history. This latter session was designed to alert the sexual parmers to the dangers of their day-to-day behavior and offered the chance for a discussion of more subtle issues and personal concerns that were not elicited by the questionnaire (Kane, 1989a,b).

292 ~ APPENDIX C "Captive" populations are also the subject of anthropological study. Hospital clinics, county health departments, and bars provide settings for personal interviews, focus group discussions, or the distribution of questionnaires to be filled in and returned to the investigator (McCombie, 1986, 1990; Marshall et al., 1990; K. Kennedy, personal communica- tion).70 Although this approach is constrained by its inability to compare self-reported and observed behaviors, it can provide information on sex- ual behavior and drug use that is otherwise difficult to obtain. Adaptation of the traditional methods of kinship charting to construct a visual image of concrete social relations is an ethnographic method that has been used to elicit information from male and female black and Puerto Rican patients at a methadone clinic in the Bronx. Patients and an anthropologist worked together to create a visual chart of the patient's kin and friends, to which were added color-coded records of his or her ~ug- us~ng and non-drug-us~ng associates, the particular drugs chosen, sexual and social intimates, those who were aware of the patient's HIV status, those who were informed about their own HIV status, and the patient's household composition as well as that of his or her children (who often lived separately). From an anthropological point of view, the chart is only a stardug point for more extended enqu~nes about venous aspects of social life. For an epidemiologist, however, the chart provides an index of the number of people at risk for HIV infection because it records the frequency of sexual intercourse, He categories of sexual partners (those who live together, those who are lovers or more casual pawners, and those involved in prostitution), and whether contraceptives are used.7~ Regional comparisons, which are commonly used In ethnographic research, have been rare In the anthropology of HIV infection. A com- panson of two ethnically distinct neighborhoods In Baltimore, however, provides some insight into the different drug choices and the motivations for and meanings of multiple drug use among African Americans and European Amencans, as weD as the relative risks of HIV infection (Ma- son, 1989~. Similarly, an ethnographic study of the social contexts of injection equipment sharing conducted in New York City and San Fran- cisco revealed that, In the "shooting galleries" of New York, several sets 70K. Kennedy, Moneefiore Family Health Center, Bronx, N.Y., personal communication, August 1989. 71A. Pivaik, Montefiore Hospital, Bronx, N.Y., personal communication, October 1989. A different kind of creative mapping plots the distribution of circumcision practices in Affica win areas of high and low AIDS prevalence, an association that has provided significant results, as it suggests that un- circumcised men run a greater risk of becoming infected during sexual intercourse Han do those who are circumcised (Reining, 1989). In response to the uncritical use of ethnographic data that marked the early years of the AIDS epidemic, a computer-assisted data base is also being designed to collate cultural information from a variety of sources to complement local, national, and international data bases on HIV seroprevalence (Conant, 1989).

METHODOLOGICAL ISSUES ~ 293 of injection equipment might be shared by more than 100 individuals in a single day. In San Francisco, however, smaller circles of friends shared injection equipment. Given New York's much higher seroprevalence among IV drug users, documentation of these widely varying conditions of injection equipment use provides some contextual understanding of the spread of the virus and the diffenng character of the epidemic in each location (Watters, 1989~. Although most social research on AIDS has concerned the behav- iors that put people at risk for HTV infection, a recent review of the ethnographic literature on sexual behaviors (Cassidy and Porter, 1989) attempted to identify "safer" (nonpenetrative) sexual practices that might form a core of culturally sensitive interventions to control the spread of the virus (keeping In mund that the concept of "safer" sex is a peculiarly Western medical notion). As the review shows, low-risk nonpenetrative behaviors (interfemoral intercourse, masturbation, mutual grooming, sex- ual joking, and so on) occur commonly throughout the world. A mere listing of sexual behaviors that appear to be universal including coitus, other forms of intercourse, masturbation, same-sex relations is not very informative, however, and may even be misleading because the contexts in which the behaviors occur, the attitudes people express about these be- haviors, and the meanings of the behaviors vary enormously from locale to locale. Sexual behavior cannot be understood apart from its cultural context, which includes historical, economic, and political aspects. As the example of Mexican American and Anglo sexual behavior in Orange County illustrates, "homosexual" and "bisexual" relations may have a different meaning and expression for different ethnic groups living in the same community. Thus, to communicate effectively with people at risk, AIDS research and interventions must be sensitive to the va~iabil- ity of sexual meaning and experience within and among cultural groups (Cassidy and Porter, 1989). Findings of Ethnographic Research on AIDS Ethnographic research on the spread of HIV infection is still in the prelun~nary phases of data gathering. Nevertheless, some suggestive patterns appear to be emerging. First, prostitutes' use of condoms for professional but not personal sex has been widely observed in Europe and in some parts of Me United States (Kane, 1989b; Worth, 1989), the West Oldies (McCombie, 1990), and Africa (Bledsoe, 1990~. That a similar pattern does not currently exist among prostitutes in Orange County, southern California (Carner, 1989; Carrier and Magana, n.d.) or among church women in Zaire (Schoepf et al., in press) is a salutary

294 ~ APPENDIX C reminder that cultural data should always be examined for its internal variability and in appropriate historical context. Second, studies of the perceptions of some segments of the scien- tific community in the United States suggest that scientists like people everywhere—may hold local views of the world that are at variance with statistical or scientific understanding. Thus health care professionals sometimes fail to adopt precautions when they are at risk, although they show excessive caution in less risky situations. The perception of whether an individual is considered "safe" or "unsafe," for example, depends on a combination of social, economic, and visual criteria (McCombie, 1989~. Third, despite the hazards of attempting to compare heterogeneous data from several countries, some consistent cultural themes can be identified. A comparison of data from Central and East Africa (Zaire, Zambia, Tanzania, and Uganda) and West Africa (Nigeria and Sierra Leone), for example, indicates that condom use poses problems for populations that stress fertility in heterosexual relations; in addition, condoms are associated with promiscuity. Moreover, polygyny is an accepted cultural behavior among men in all three regions, and women may also have multiple sexual partners, often as a result of economic pressures (Bledsoe, 19901. Finally, several studies indicate that educational messages concerning the dangers of unprotected sex reach some audiences but have actually increased the dangers of HIV infection for other segments of the popu- lation. In response to public education about AIDS, men in some parts of Africa continue to pursue an active sexual life but have turned from high-risk groups (e.g., prostitutes) to low-risk pools (e.g., schoolgirls, who may be willing to exchange sex for money to finance their educa- lion) (Bledsoe, 19901. Similarly, as a result of educational campaigns in the United States, the clients of street prostitutes In New Jersey report choosing novices and apparently inexperienced young girls in hopes of avoiding long-term drug users who may be infected with HIV (Leonard, 1990~. Gaps and Deficiencies in Current Ethnographic Research The current flurry of anthropological research on sexual behavior and drug use suffers from the absence of a sustained scholarly tradition in both fields. Although there is more research on sexual behavior than on drug use, the study of sexuality has focused for several decades on sexual meanings and beliefs and has tended to ignore sexual behaviors. The usefulness of such data for HIV research thus is limited. When sexual behaviors were reported, normative behaviors were highlighted

METHODOLOGICAL ISSUES ~ 295 rather than the vaned ways in which people often choose to lead their sexual lives. Moreover, this earlier literature provides little information concerning the substantial changes that have now altered behaviors in once-isolated regions. Studies of drug use have provided information on patterns of behavior in homes and shooting galleries (Koester, 1989a,b; Watters, 1989; B.P. Page et al., 1990), on the meaning and practices of injection equipment use (Connors, 1989), and on different patterns of multiple drug use and high-nsk behaviors in different ethnic communities (Mason, 1989~. Attention is now beginning to turn to the broader determinants of high- risk Mug behaviors—such as the history and political economy of drug use and drug marketing (Hamid, 1990; Mason, 1989) and the way in which laws that make carrying a syringe a crime increase the probability that drug users will adopt risky behaviors (Koester, l989a,b). Much more information is needed, however, on such social determinants, as well as on individual perceptions of the risky behaviors associated with drug use. Anthropologists and epidemiologists have had some success in iden- tify~ng and investigating "risky" behaviors in many locations. Yet inter- vention strategies have sometimes been directed too narrowly at behav- iors rather than at people in context. Effective intervention may require a broader understanding of both the personal and social determinants of risk behaviors (cf. O'Reilly, 1989~. A further area in which much ethno- graphic work deserves to be conducted involves the relationship between belief and behavior. In this regard, there is a distinction to be drawn between constructs Mat constitute a public language and constructs that guide individual choices in specific situations. Recent studies point to the way in which individuals personalize the rules of behavior to fit their own wishes and Me limited options from which Key might choose (Eyre, 1989; Kane, 1989a,b; Cassidy and Porter, 19891. The ability of impov- erished women to practice "safer" sexual behaviors, for example, may be particularly circumscribed. Furthe~n~ore, very little is known about the interaction between private worlds of erotic behavior and the public domain of shared meanings recently explored by Parker (1989~. It has been suggested that rules regulating sexual behavior may be particularly prone to individual negotiation and improvisation (Cassidy and Porter, 1989~. All of these factors emphasize the continuing valuable role to be played by ethnographic research and the discipline of anthropology. RECOMMENDATIONS Given the evidence reviewed in the foregoing pages, the committee concludes that there is good reason to believe that accurate measurements

296 ~ APPENDIX C of AIDS risk behaviors can be obtained. The committee notes, however, that there is substantial room for improvement in current efforts. This potential is not surpns~ng given the immaturity of many of the relevant research fields. The committee believes that appropriate Investments to create a better foundation of relevant methoclological knowledge can lead to more certain scientific understanding about the behaviors Mae transmit HIV and Be factors that motivate and shape these behaviors. Toward this end the committee makes the following recommendations. The committee recommends that the Public Health Service and other organizations supporting AIDS research provide increased sup- port for methodological research on the measurement of behaviors that transmit HIV. Such research should consider inferential prob- lems introduced by nonresponse and by nonsampling factors, includ- ing (but not limited to) the effects of question wording and question context, the time periods and events that respondents are asked to recall, and the effects of anonymity guarantees on survey responses. The committee recommends that researchers who conduct be- havioral surveys on HIV transmission make increased use of ethno- graphic studies, pretests, pilot studies, cognitive laboratory investiga- tions, and other similar developmental strategies to aid in the design of large-scale surveys. The committee recommends that, where appropriate, researchers embed experimental studies within behavioral surveys on HIV trans- mission to assess the effects of key aspects of the survey measurement process. The committee recommends that, whenever feasible, researchers supplement self-reports in behavioral surveys on HIV transmission with other indicators of these behaviors that do not rely on respon- dent reports. REFERENCES Aiken, L. S. (1986) Retrospective self-reports by clients differ from original reports: Implications for the evaluation of drug treatment programs. International Journal of the Addictions 21:767-788. Aiken, L. S., and LoSciuto, L. A. (1985) Ex-addict versus nonaddict counselors' knowledge of clients' drug use. international Journal of the Ad/dictions 20:417- 433. Allen, R. M., and Haupt, T. D. (1966) The sex inventory: Test-retest reliability of scale scores and items. Journal of Clinical Psychology 22:375-378.

METHODOLOGICAL ISSUES ~ 297 Alonso, A. M., and Koreck, M. T. (1989) Silences: Hispanics, AIDS, and sexual practices. Differences 1:101-124. Amsel, Z., Mandell, W., Matthias, L., Mason, C., and Hocherman, I. (1976) Reliability and validity of self-reported illegal activities and drug use collected from narcotic addicts. International Journal of the Addictions 11:325-336. Anastasi, A. (1976) Psychological Testing. 4th ed. New York: Macmillan. Andersen, B. L., and Broffitt, B. (1988) Is there a reliable and valid self-report measure of sexual behavior? Archives of Sexual Behavior 17:509-525. Andersen, R., Kasper, J., Frankel, M. R., and Associates. (1979) Total Survey Error. San Francisco: Jossey-Bass. Anderson, R. M., and May, R. M. (1988) Epidemiological parameters of HIV transrnis- sion. Nature 333:51~519. Aral, S. O., Magder, L. S., and Bowen, G. S. (1989) HIV risk behavior screening: Con- cordance between assessments through interviews and questionnaires. Presented at the Fifth International Conference on AIDS, Montreal, June =9. Axon, W. G. (n.d.) The influence of interviewer gender on responses to sensitive questions in less developed settings: Evidence from Nepal. Population Studies Center, Institute for Social Research and Deparunent of Sociology, University of Michigan, Ann Arbor, Mich. Baddeley, A. D. (1979) The limitations of human memory: Implications for the design of retrospective surveys. In L. Moss and H. Goldstein, eds., The Recall Method in Social Surveys. London: University of London Institute of Education. Bailar, B. (1975) The effects of rotation group bias on estimates from panel surveys. Journal of the American Statistical Association 70:23-30. Bailar, B., Bailey, L., and Stevens, J. (1977) Measures of interviewer bias and variance. Journal of Marketing Research 14:337-343. Bailey, L., Moore, T., and Bailar, B. (1978) An interviewer variance study for the eight impact cities of the National Crime Survey cities sample. Journal of the American Statistical Association 73:1~23. Bale, R. N. (1979) The validity and reliability of self-reported data from heroin ad- dicts: Mailed questionnaires compared with face-to-face interviews. International Journal of the Addictions 14:99~1000. Bale, R. N., Van Stone, W. W., Engelsing, T. M. J., Zarcone, V. P., and Kuldau, J. M. (1981) The validity of self-reported heroin use. International Journal of the Addictions 16:1387-1398. Ball, J. C. (1967) The reliability and validity of interview data obtained from 59 narcotic drug addicts. American Journal of Sociology 72:65~654. Ball, J. C., Lange, W. R., Myers, C. P., and Friedman, S. R. (1988) Reducing the risk of AIDS through methadone maintenance treatment. Journal of Health and Social Behavior 29:21~226. Bardoux, C., Buning, E., Leentvaar-Kuijpers, A., Verster, A., and Coutinho, R. A. (1989) Declining incidence of acute hepatitis B among drug users in Amsterdam may indicate a change in risk behavior. Presented at the Fifth International Conference on AIDS, Montreal, June ~9. Bartlett, F. C. (1932) Remembering: A Study in Experimental Social Psychology. Cambridge University Press: Cambridge, Great Britain.

298 ~ APPENDIX C Batki, S. L., Sorensen, J. L., Coates, C., and Gibson, D. R. (1989) Methadone maintenance for AIDS-affected IV drug users: Psychiatric factors and outcome three months into treatment (abstract). In L. S. Hams, ea., Problems of Drug Dependence 1988: Proceedings of the Committee on the Problems of Drug Dependence. Washington, D.C.: U.S. Government Printing Office. Bauman, L. J., and Adair, E. G. (1989) Use of ethnographic interviewing to informal questionnaire construction. Presented at the Annual Meeting of the American Association for Public Opinion Research, St. Petersburg, Fla., May. Beck, C., Ward, C., Mendelson, M., Mock, J., and Erbaugh, J. (1961) An inventory measuring depression. Archives of General Psychiatry 4:561-571. Beniger, J. R. (1984) Mass media, contraceptive behavior, and attitudes on abortion: Toward a comprehensive model of subjective social change. In C. F. Turner and E. Martin, eds., Surveying Subjective Phenomena. Vol. 2. New York: Russell Sage. Benney, M., Riesman, D., and Star, S. A. (1956) Age and sex In the interview. American Journal of Sociology 62:143-152. gentler, P. M. (1968a) Heterosexual behaviour assessment. I. Males. Beh~iour Research and Therapy 6:21-25. gentler, P. M. (1968b) Heterosexual behaviour assessment. II. Females. Beh~viour Research and Therapy 6:27-30. Benus, J., and Ackennan, J. C. (1971) The problem of nonresponse in sample surveys. In J. B. Lansing, S. B. ~they, A. C. Wolfe et al., eds., Working Papers on Survey Research in Poverty Areas, Ann Arbor: Survey Research Center, Institute for Social Research, University of Michigan. Ben-Yehuda, N. (1980) Are addicts' self-reports to be trusted? international Journal of the Addictions 15:1265-1270. Bernard, H. R. (1988) Research Methods in Cultural Anthropology. Newbury Park, Calif.: Sage. Biderman, A. D., and Lynch, J. P. (1981) Recency bias in data on self-reported victimization. Proceedings of the American Statistical Association (Social Statistics Section) 1981:31~0. Biderman, A. D., and Moore, J. C. (1980) Report of the workshop on applying cog- nitive psychology to recall problems of the National Cnme Survey. Unpublished manuscript. Bureau of Social Science Research, Washington, D.C. Bihari, B., and Ottomanelli, G. (1989) Defense mechanisms and HIV risk related behaviors in substance abusers. Presented at the Fifth International Conference on AIDS, Montreal, June =9. Billy, J. O. G., and Udry, J. R. (1985) Patterns of adolescent friendship and effects on sexual behavior. Social Psychology Quarterly 48:27~1. Bishop, Y. M. M., Fienberg, S., and Holland, P. W. (1975) Discrete Multivariate Analysis. Cambndge, Mass.: Massachusetts Institute of Technology Press. Bledsoe, C. (1990) The politics of AIDS and condoms for stable heterosexual relations in Africa: Recent evidence from the local print media. In W. P. Handwerker, ea., Births and Power: The Politics of Reproduction. Boulder, Colo.: Westview Press. Blumstein, P., and Schwartz, P. (1977) Bisexuality: Some social psychological issues. Journal of Social Issues 33:30~5. Blumstein, P. W., and Schwartz, P. (1983) American Couples; Money, Work, and Sex. New York: Morrow.

METHODOLOGICAL ISSUES ~ 299 Bonito, A. J., Nurco, D. N., and Shaffer, J. W. (1976) The veridicality of addicts' self-reports in social research. International Journal of the Addictions 11:719-724. Borgatta, E. F., Blumstein, P., and Schwartz, P. (1987) Research Methodologies for Sensitive Data Collection: Final Report submitted to the Centers for Disease Control. Department of Sociology, University of Washington, Seattle. Boruch, R. F. (1989) Resolving privacy problems in AIDS research: A primer. In L. Sechrest, H. Freeman, and A. Mulley, eds., Health Services Research Methodology: A Focus on AIDS. Conference Proceedings. Rockville, Md.: National Center for Health Services Research and Health Care Technology Assessment. Bradburn, N. W., Rips, L. J., and Shevell, S. K. (1987) Answenug autobiographical questions: The impact of memory and inference on surveys. Science 236:157-161. Bradburn, N. M., Sudman, S., and Associates (1979) Improving interview Method and Questionnaire Design. San Francisco: Jossey Bass. Brady, J. P., and Levitt, E. E. (1965~1~e scalability of sexual experiences. Psychological Record 15:275-279. Brewer, M. B., Dull, V. T., and Jobe, J. B. (1989) Social cognition approach to reporting chronic health conditions. Vital awl Health Statistics, Series 6, Whole No. 3. Brooks, C. A., and Bailar, B. (1978) An Error Profile: Employment as Measured by the Current Population Survey. Statistical Policy Working Paper 3. Washington, D.C.: U.S. Department of Commerce. Brown, L. S., Phillips, R., Ajulchukwu, D., Battjes, R., Prinun, B. J., and Nemoto, T. (1989) Demographic and behavioral features of HIV infection in intravenous drug users in New York City drug treatment programs: 1985-1988. Presented at the Fifth International Conference on AIDS, Montreal, June =9. Cain, V. S., and Baldwin, W. (1989) The national study of health and sexual behavior. Presented at the Fifth International Conference on AIDS, Montreal, June =9. Campbell, A. A. (1987) Randomized response technique (letter). Science 236:1049. Cannell, C. F., Fisher, G., and Bakker, T. (1965) Reporting of hospitalization in the Health Interview Survey. Vital and Health Statistics, Series 2, Whole No. 6. Cannell, C. F., Marquis, K. H., and Laurent, A. (1977) A summary of studies of interviewing methodology. Vital and Health Statistics, Senes 2, Whole No. 69. Cannell, C. F., Miller, P. V., and Oksenberg, L. (1981) Research on interviewing techniques. In S. Leinhardt, ea., Sociological Methodology 1981. San Francisco: Jossey-Bass. Cannell, C., Oksenberg, L., Kalton, G., Bischoping, K., and Fowler, F. J. (1989) New techniques for pretesting survey questions. Final Report to the National Center for Health Services Research and Health Care Technology Assessment. Survey Research Center, University of Michigan, Ann Arbor, Mich. Camer, J. M. (1989) Sexual behavior and the spread of AIDS in Mexico. Medical Anthropology 10:129~142. CaITier, J. M., and Magana, J. R. (n.d.) Applied anthropology and AIDS in a heals care agency. Unpublished manuscnpt. Orange County Health Care Agency, Orange County, Calif. Cassidy, C. M., and Porter, R. W. (1989) Ethnographic Perspectives on Nonpenetrative Sexual Behavior. Research Tnangle Park, N.C.: AIOSCOM, Family Health International. Castro, K. G., Lieb, S., Jaffe, H. W., Narkonas, J. P., Calisher, C. H., et al. (1988) Transmission of HIV in Belle Glade, Flonda: Lessons for other communities in the United States. Science 239:19~197.

300 ~ APPENDIX C Catania, J. A., McDermott, L. J., and Pollack, L. M. (1986) Questionnaire response bias and face-to-face interview sample bias in sexuality research. Journal of Sex Research 22:52-72. Catania, J. A., Gibson, D. R., Chitwood, D. D., and Coates, T. J. (199Oa) Methodological problems in AIDS behavioral research: Influences on measurement error and participation bias in studies of sexual behavior. Psychological Bulletin 108:339- 362. Catania, J. A., Gibson, D. R., Marin, B., Coates, T. J., and Greenblatt, R. M. (199Ob) Response bias in assessing sexual behaviors relevant to HIV transmission. Evaluation and Program Planning 13:19-29. Centers for Disease Control (CDC). (1988) V-related beliefs, knowledge, and be- haviors among high school students. Morbidity and Mortality Weekly Report 37:717-721. Chaisson, R. E., Osmond, D., Moss, A. R., Feldman, H. W., and Bernacki, P. (1987) HIV, bleach, and needle sharing. Lancet 20 June:1430. Chaisson, R. E., Osmond, D., Bacchetti, P., Brodie, B., Sande, M. A., and Moss, A. R. (1988) Cocaine, race and HIV infection in IV drug users. Presented at the Fourth International Conference on AIDS, Stockholm, June 12-16. Chaisson, R. E., Bacchetti, P., Osmond, D., Brodie, B., Sande, M. A., and Moss, A. R. (1989) Cocaine use and HIV infection in intravenous drug users. Journal of the American Medical Association 261:561~09. Cisin, I. H., and Park, H. L. (1979) Sensitivity of survey techniques in measuring illicit drug use. In J. D. Rittenhouse, ea., Developmental Papers: Attempts to Improve the Measurement of Heroin Use in the National Survey. Washington, D.C.: National Institute on Drug Abuse. Citro, C. F., and Cohen, M. L. (1985) The Bicentennial Census: New Directions for Methodology in 1990. Report of the National Research Council Panel on Decennial Census Methodology. Washington, D.C.: National Academy Press. Clark, A. L. and Wallin, P. (1964) Lee accuracy of husbands' and wives' reports of the frequency of marital coitus. Population Studies 18:165-173. Clark, J. P., and Tifft, L. L. (1966) Polygraph and interview validation of self-reported deviant behavior. American Sociological Review 31:51~523. Coates, R. A., Soskoine, C. L., Calzavara, L., Read, S. E., Fanning, M. M., et al. (1986) The reliability of sexual histories in AIDS-related research: Evaluation of an interview-administered questionnaire. Canadian Journal of Public Health 77:343-348. Coates, R. A., Calzavara, L. M., Soskolne, C. L., Read, S. E., Fanning, M. M., et al. (1988) Validity of sexual histories in a prospective study of male sexual contacts of men with AIDS or an AIDS-related condition. American Journal of Epidemiology 128:719-728. Cochran, W. G., Mosteller, F., and Tukey, J. W. (1953) Statistical problems of the Kinsey report. Journal of the American Statistical Association 48:67~716. Communication Technologies, Inc. (1987) A Report on Designing an Effective AIDS Prevention Campaign Strategy for San Frar~cisco: Results from the Fourth Proba- bility Sample of an Urban Gay Male Community. Son Francisco: Communication Technologies, Inc., July 31, 1987. Communication Technologies, Inc. (1988) A Report on Planning for the AIDS Epidemic in California: A Population-Based Assessment of Knowledge, Attitudes, and Behavior. San Francisco: Communication Technologies, Inc., June 30, 1988.

METHODOLOGICAL ISSUES ~ 301 Conant, F. P. (1989) AIDS and beyond: The need, and a design, for a targeted cultural information management system. Unpublished manuscript. Department of Anthropology, Hunter College, City University of New York. Connors, M. (1989) Perception of risk and HIV infection among intravenous drug users (IVDUs) in Worcester, Mass. Unpublished manuscript. Department of Anthropology, University of Massachusetts, Amherst. Conte, H. R. (1983) Development and use of self-report techniques for assessing sexual functioning: A review and antique. Archives of Sexual Behavior 12:555-576. Converse, J. M., and Schuman, H. (1974) Conversations at Random: Survey Research as Interviewers See It. New York: Wiley. Coombs, L. C. (1977) Levels of reliability in fertility survey data. Studies irz Family Planning 8:218-232. Council of American Survey Research Organizations (CASRO) (1982) On the Definition of Response Rates. A Special Report of the CASRO Task Force on Completion Rates. Port Jefferson, N.Y. Cox, T. J., and Longwell, B. (1974) Reliability of interview data concerning current heroin use from heroin addicts on methadone. International Journal of the Addictions 9: 161-1 65. Coxon, A. P. M. (1986) Report of a Pilot Study: Project on Sexual Lifestyles of Non- Heterosexual Males. Unpublished manuscript. Social Research Unit, University College, Cardiff, Wales. Coxon, A. P. M. (1988) Something sensational: The sexual diary as a tool for mapping detailed sexual behavior. Sociological Review 35:353-367. Coxon, A. P. M., and Carballo M. (1989) Editorial Review: Research on AIDS: Behavioral perspectives. AIDS 3:191-197. Coxon, A. P. M., and Davies, P. M. (1989) Using structured sexual diaries data to estimate rates, predictors and context of high-risk sexual behaviour. Presented at the Fit International Conference on AIDS, Montreal, June =9. Crider, R. A. (1985) Heroin incidence: A trend comparison between National Household Survey data and indicator data. In B. A. Rouse, N. J. Kozel, and L. G. Richards, eds., Self-Report Methods of Estimating Drug Use: Meeting Current Challenges to Validity. DELIS Publication No. (ADM) 85-1402. National Institute on Drug Abuse Research Monograph No. 57. Washington, D.C.: U.S. Government Printing Office. Criqui, M. H. (1979) Response bias and risk ratios in epidemiologic studies. American Journal of Epidemiology 109:39~399. Cronbach, L. J., and Gleser, G. C. (1965) Psychological Tests and Personnel Decisions. 2nd ed. Urbana: University of Illinois Press. Cynamon, M. L. (1989) The national survey of health and sexual behavior: The pretest experience. Presented at the Fifth International Conference on AIDS, Montreal, June =9. Czaja, R. (1987-88) Asking sensitive behavioral questions in telephone surveys. Applied Research and Evaluation 8:23-32. Daniel, W. W. (1975) Nonresponse in sociological surveys. Sociological Methods and Research 3:291-307. Darrow, W. W., Jaffe, H. W., Thomas, P. A., Haverkos, H. W., Rogers, M. F., et al. (1986) Sex of interviewer, place of interview, and responses of homosexual men to sensitive questions. Archives of Sexual Behavior 15:79~8.

302 ~ APPENDIX C Davies, P. M. (1986) Some problems in defining and sampling non-heterosexual males. Working Paper No. 21. Social Research Unit, University College, Cardiff, Wales. Davis, K. H., Hawks, R. L., and Blanke, R. V. (1988) Assessment of laboratory quality in urine drug testing. Journal of the American Medical Association 260:1749-1754. Day, N., Houston-Hamilton, A., Taylor, D., Lemp, G., and Rutherford, G. (1989) Tracking survey of AIDS knowledge, attitudes and behaviors in San Francisco's Black communities. Presented at the Fifth International Conference on AIDS, Montreal, June ~9. ~ De Gruttola, V., and Fineberg, H. V. (1989) Estimating prevalence of HIV infection. Considerations in the design and analysis of a national seroprevalence survey. Journal of Acquired Immune Deficiency Syndromes 2:472~80. DeLamater, J. D. (1974) Methodological issues in the study of premarital sexuality. Sociological Methods and Research 3:3~61. DeLamater, J., and MacCorquodale, P. (1975) The effects of interview schedule variations on reported sexual behavior. Sociological Methods and Research 4:215-236. DeLamater, J., and MacCorquodale, P. (1979) Premarital Sexuality: Attitudes, Rela- tionships, Behavior. Madison, Wisc.: University of Wisconsin Press. DeMaio, T. J., ed. (1984) Approaches to Developing Questionnaires. Statistical Policy Working Paper No. 10. Washington, D.C.: Office of Management and Budget. DeMaio, T. J. (1984) Social desirability and survey measurement: A review. In C. F. Turner and E. Martin, eds., Surveying Subjective Phenomena. Vol. 2. New York: Russell Sage. Derogatis, L. R. (1980) Psychological assessment of psychosexual functioning. Psv~hi- atric Clinics of North America 3:11~131. O , . Derogatis, L. R., and Melisaratos, N. (1979) The DSFI: A multidimensional measure of sexual functioning. Journal of Sex and Marital Therapy 5:244 281. Des Jarlais, D. C., and Friedman, S. R. (1988a) Intravenous cocaine, crack, and HIV infection. Journal of the American MedicalAssociation 259:1945-1946. Des Jarlais, D. C., and Friedman, S. R. (1988b) Needle sharing among IVDUs at risk for AIDS. American Journal of Public Health 78:1498. lees Jarlais, D. C., and Friedman, S. R. (1988c) The psychology of preventing AIDS among intravenous drug users: A social learning conceptualization. American Psychologist 43:865 - 870. Des Jarlais, D. C., Friedman, S. R., and Stoneburner, R. L. (1988) HIV infection and intravenous drug use: Critical issues in transmission dynamics, infection outcomes, and prevention. Reviews of Infectious Diseases 10:151-158. Des Jarlms, D. C., Friedman, S. R., Sotheran, J. L., and Stoneburner, R. (1988) The sharing of drug injection equipment and the AIDS epidemic in New York City: The first decade. In R. J. Battjes and R. W. Pickens, eds., Needle Sharing Among Intravenous Drug Abusers: National anal International Perspectives. National Institute on Drug Abuse Research Monograph 80. Washington, D.C.: U.S. Government Printing Office. Des Jarlais, D. C., Tross, S., AWul-Quader, A., Kouzi, A., and Friedman, S. R. (1989a) Intravenous drug users and maintenance of behavior change. Presented at the Fifth International Conference on AIDS, Montreal, June =9.

METHODOLOGICAL ISSUES ~ 303 Des Jarlais, D. C., Hagen, H., Purchase, D., Reid, T., and Friedman, S. R. (1989b) Safer injection among participants in the first North American syringe exchange program. Presented at the Fifth International Conference on AIDS, Montreal, June =9. Dougherty, J. A. (1988) Prevention of HIV transmission in IV drug users. Presented at the IV International Conference on AIDS, Stockholm, June 12-16. Drug Enforcement Administration (DEA). (1988) Crack Cocaine Availability and Traf- ficking in the United States. Washington, D.C.: ITS. Department of Justice. January. Eckerman, W. C., Bates, J. D., Rachal, J. V., and Poole, W. K. (1971) Drug Usage and Arrest Charges. Washington, D.C.: Bureau of Narcotics and Dangerous Drugs. Eisenberg, D. (1981) A scientific gold rush. Science 213:1104 1105. Ericsson, K. A., and Simon, H. A. (1980) Verbal reports as data. Psychological Review 87:215-251. Exner, T. M., Meyer-Bahlburg, H. F. L., Gwen, R. S., Ehrhardt, A. A., and Gorman, J. M. (1989) Peer norms for safer sex as a predictor of sexual risk behaviors in a cohort of gay men. Presented at the Fifth International Conference on AIDS, Montreal, June =9. Eyre, S. L. (1989) Metaphors of AIDS: Social and existential function. Presented at the Annual Meeting of the American Anthropological Association, Washington, D.C., November 1~19. Farmer, P. (1990) Sending sickness: Sorcery, politics, and changing concepts of AIDS in rural Haiti. Medical Anthropology Quarterly 4:6-27. Fanner, P., and Kleinman, A. (1989) AIDS as human suffenng. Daedalus 118:135-160. Pay, R. E., Turner, C. F., Klassen, A. D., and Gagnon, J. H. (1989) Prevalence and patterns of same-gender sexual contact among men. Science 243:338-348. Feldman, H. F., and Biernacki, P. (1988) The ethnography of needle shanug among intravenous drug users and implications for public policies and intervention strategies. In R. J. Battjes and R. W. Pickens, eds., Needle Sharing Among Intravenous Drug Abusers: National and International Perspectives. DHHS Publication No. (ADM) 8~1567. National Institute on Drug Abuse Research Monograph 80. Washington, D.C.: U.S. Government Printing Office. Ferber, R. (1966) The Reliability of Consumer Reports of Financial Assets and Debts. Bureau of Economic and Business Research, University of Illinois, Urbana. Fineberg, H. (1988) Education to prevent AIDS: Prospects and obstacles. Science 239:592-596. Flaskerud, J. H., and Nyamathi, A. M. (1989) An AIDS education program for black and Latina women. Presented at the Fifth International AIDS Conference, Montreal, June =9. Fowler, F. J., and Mangione, T. W. (1990) Standardized Survey Interviewing. Newbury Pam, Calif: Sage. Fox, J. A., and Tracy, P. E. (1986) Randomized Response: A Method for Sensitive Surveys. Beverly Hills, Calf: Sage Publications. Franks, F. (1981) Polywater. Cambridge, Mass.: Massachusetts Institute of Technology Press. Friedman, S. R., and Des Jarlais, D. C. (1989) Measurement of Intravenous Drug Use Behaviors that Risk HIV Transmission. Unpublished manuscript. Narcotic and Drug Research, Inc., New York.

304 ~ APPENDIX C Friedman, S. R., Sotheran, J. L., Abdul-Quader, A., Primm, B. J., Des Jarlais, D. C., et al. (1987a) The AIDS epidemic among blacks and Hispanics. The Milbank Quarterly 65:455~99. Friedman, S. R., Des Jarlais, D. C., Sotheran, J. L., Garber, J., Cohen, H., and Smith, D. (1987b) AIDS and self-organizai~on among intravenous drug users. International Journal of the Addictions 22:201-219. Friedman, S. R., Dozier, C., Sterk, C., Williams, T., Sotheran, J. L., et al. (1988) Crack use puts women at risk for heterosexual transmission of HIV from intravenous drug users. Presented at the Fourth International Conference on AIDS, Stockholm, June 12-16. Friedman, S. R., Rosenblum, A., Goldsmith, D., Des Jarlais, D. C., Sufian, M., et al. (1989) Risk factors for HIV-1 infection among street-recruited intravenous drug users in New York City. Presented at the Fifth International Conference on AIDS, Montreal, June =9. Fullilove, R. E. m, Fullilove, M. T., Bowser, B., and Gross, S. A. (1989) Crack use and risk for AIDS among black adolescents. Presented at the Fifth International Conference on AIDS, Montreal, June =9. Fullilove, R. E., Fullilove, M. T., Bowser, B., and Gross, S. (1990) Crack users: The new AIDS risk group? Journal of Cancer Prevention and Detection 14:363-368. Gagnon, J. H. (1988) Sex research and sexual conduct in the era of AIDS. Journal on AIDS 1:593~01. Gawin, F. H., and Ellinwood, E. H., Jr. (1988) Cocaine and other stimulants: Actions, abuse, and treatment. New England Journal of Medicine 318:117~1182. Gebhard, P. H. (1972) Incidence of overt homosexuality in the United States and Western Europe. ~ National Institute of Mental Health Task Force on Homosexuality, Final Report and Background Papers. Washington, D.C.: U.S. Government Printing Office. Gebhard, P. H., and Johnson, A. B. (1979) The Kinsey Data: Marginal Tabulations of 1938-1963 Interviews Conducted by the institute for Sex Research. Philadelphia: W. B. Saunders. Gelmon, K., Schechter, M. T., Sheps, S. B., Hershler, R., and Craib, K. J. P. (1989) Recall bias and memory failure: An empiric demonstration in persons with acquired immune deficiency syndrome. Presented at the Fifth International Conference on AIDS, Montreal, June =9. Gibson, D., Wermuth, L., Sorensen, J. L., Menicucci, L., and Bernal, G. (1987) Approval need in self-reports of addicts and family members. International Journal of the Addictions 22:895-903. Gibson, D. R., Sorensen, J. L., Lovelle-Drache, J., Catania, J., Kegeles, S., and Young, M. (1988) Psychosocial predictors of AIDS high risk behavior among intra- venous drug users and their sexual pamers. Presented at the For International Conference on AIDS, Stockholm, June 12-16. Gibson, D. R., Lovelle-Drache, J., Derby, S., Garcia-Soto, M., Sorensen, J. L., and Melese-d'Hospital, I. (1989) Brief counseling to reduce AIDS risk in IV dmg users: Update. Presented at the Fifth International Conference on AIDS, Montreal, June =9. Gonzalez, M., Ogus, J., Shapiro, G., and Tepping, B. (1975) Standards for discussion and presentation of errors in survey and census data. Journal of the American Statistical Association 70~351, whole pt. 2~.

METHODOLOGICAL ISSUES ~ 305 Goodstadt, M. S., and Gruson, V. (1975) The randomized response technique: A test on drug use. Journal of the American Statistical Association 70:814 818. Goodstadt, M. S., Cook, G., and Gruson, V. (1978) The validity of reported drug use: The randomized response technique. International Journal of the Addictions 13:359-367. Goyder, J. (1987) The Silent Minority: Nonrespondents on Sample Surveys. Cambridge, U.K.: Polity Press. Groves, R. M., and Kahn, R. L. (1979) Surveys by Telephone: A National Comparison with Personal Interviews. New York: Academic Press. Gulliksen, H. (1950) Theory of Mental Tests. New York: Wiley. Haberman, P. W., Josephson, E., Zanes, A., and Ellinson, J. (1972) High school drug behavior: A methodological report on pilot studies. In S. Einstein and S. Allen, eds., Proceedings of First International Conference on Student Drug Surveys. New York: Baywood Publishing. Hamid, A. (1990) The political economy of crack-related violence. Contemporary Drug Problems Spring 1990. Hamilton, M. (1960) A rating scale for depression. Journal of Neurological and Neurosurgical Psychiatry 23:56 61. Harbison, J. J. M., Graham, P. J., Quinn, J. T., McAllister, H., and Woodward, R. (1974) A questionnaire measure of sexual interest. Archives of Sexual Behavior 3:357-366. Harrell, A. V. (1985) Validation of self-report: The research record. In B. A. Rouse, N. J. Kozel, and L. G. Richards, eds., Self-Report Methods of Estimating Drug Use: Meeting Current Challenges to Validity. National Institute on Drug Abuse Research Monograph No. 57. Washington, D.C.: U.S. Government Printing Office. Heckert, K., Shultz, J., and Salem, N. (1989) The Minnesota general public AIDS survey: A behavioral risk factor survey supplement. Presented at the Fifth International Conference on AIDS, Montreal, June =9. Heitzman, C. A., Sorensen, J. L., Gibson, D. R., Morales, E. R., Costantin, M., et al. (1989) AIDS prevention among IV drug users: Behaviors changes. Presented at the Annual Meeting of the Society of Behavioral Medicine, San Francisco, Calif. Henson, R., Cannell, C. F., and Lawson, S. (1973) Effects of Interviewer Style and Question Form on Reporting of Automobile Accidents. Ann Arbor: Survey Research Center, University of Michigan. Herb, F., Watters, J. K., Case, P., and Petitti, D. (1989) Endocarditis, subcutaneous abscesses, and other bacterial infections in intravenous drug users and their association with skin-cleaning at drug injection sites. Presented at the Fifth Intemational Conference on AIDS, Montreal, June =9. Herold, E. S., and Way, L. (1988) Sexual self-disclosure among university women. Journal of Sex Research 24:1-14. Hesselbrock, M., Babor, T. F., Hesselbrock, V., Meyer, R. E., and Workman, K. (1983) Never believe an alcoholic? On the validity of self-report measures of alcohol dependence and related constructs. International Journal of the Addictions l8:S93~09. Himmelweit, H. T., and Turner, C. F. (1982) Social and psychological antecedents of depression. In P. Baltes and O. Brim, eds. Life-span Development arid Behavior. New York: Academic Press.

306 APPENDIX C Ho, C. Y., Powell, R. W., and Liley, P. E. (1974) Thermal conductivity of the elements: A comprehensive review. Journal of Physical and Chemical Reference Data 3: 1-244. Hochhauser, M. (1987) Readability of AIDS educational materials. Presented at the Annual Meeting of the American Psychological Association, New York, August. Hofferth, S. L., Kahn, J. R., and Baldwin, W. (1987) Premarital sexual activity among U.S. teenage women over the past three decades. Family Planning Perspectives 19:46-53. Hoon, E. F., Hoon P. W., and Wincze, J. P. (1976) An inventory for the measurement of female sexual arousability: The SAI. Archives of Sexual Behavior 5:291-300. Huang, K. H. C., Watters, J. K., and Case, P. (1988) Psychological assessment and AIDS research with intravenous drug users: Challenges in measurement. Journal of Psychoactive Drugs 20:191-195. Huang, K. H. C., Watters, J., and Case, P. (1989) Compliance with AIDS prevention measures among intravenous drug users: Health beliefs or sociaVenvironmental factors? Presented at the Fifth International Conference on AIDS, Montreal, June =9. Hubbard, R. L, Marsden, M. E., and Allison, M. (1984) Reliability and Validity of TOPS Data. Research Triangle Park, N.C.: Research Tnangle Institute. Hubbard, R. L., Eckennan, W. C., Rachal, J. V. and Williams, J. R. (1977) Factors affecting the validity of self-reports of drug use: An overview. Proceedings of the American Statistical Association (Social Statistics Section) 1977:36~365. Chicago, Ill., August 15-18. Hunter, J. S. (1977) Quality assessment of measurement methods. In National Research Council, Environmental Monitoring, Vol. 4a. Washington, D.C.: National Academy of Sciences. Hunter, J. S. (1980) The national system of scientific measurement. Science 210:869- 874. Hyman, H. H., Cobb, W. J., Feldman, J. J., Hart, C. W., and Stember, C. H. (1954) Interviewing in Social Research. Chicago: University of Chicago Press. Jabine, T. B., Straf, M. L., Tanur, J. M., Tourangeau, R., eds. (1985) Cognitive Aspects of Survey Methodology. Washington, D.C.: National Academy Press. Jackson, D. D., Lee, W. B., and Liu, C. (1980) Aseismic uplift in southern California: An alternative interpretation. Science 210:53~536. Jacobson, N. S., and Moore, D. (1981) Spouses as observers of the events in their relationship. Consulting and Clinical Psychology 49:269-277. Johnson, A., Wadsworth, J., Elliot, P., PIior, L., Wallace, P., et al. (1989) A pilot study of sexual lifestyle in a random sample of the population of Great Britain. AIDS 3:135-141. Johnston, L. D., and O'Malley, P. M. (1985) Issues of validity and population coverage in student surveys of drug use. In B. A. Rouse, N. J. Kozel, and L. G. Richards, eds., Self-report methods of estimating drug use: Meeting current challenges to validity. DHHS Publication No. (ADM) 85-1402. National Institute on Drug Abuse Research Monograph No. 57. Washington, D.C.: U.S. GoveInment Printing Office. Josephson, E. (1970) Resistance to community surveys. Social Problems 18:117-129. Kahn, J. R., Kalsbeek, W. D., and Hofferth, S. L. (1988) National estimates of teenage sexual activity: Evaluating the comparability of three national surveys. Demography 25:189-204.

METHODOLOGICAL ISSUES ~ 307 Kahn, R. L., and Cannell, C. F. (1957) The Dynamics of Interviewing. New York: Wiley. Kane, S. (1989a) Heterosexuals, AIDS and the heroin subculture. Unpublished manu- script. Kane Agency, New York, N.Y. Kane, S. (1989b) Interviews with two women: AIDS, addiction and hetero-sex on Chicago's south side. Unpublished manuscnpt. Kane Agency, New York, N.Y. Kann, L., Nelson, G. D., Jones, J. T., and Kolbe, L. J. (1989) Establishing a system of complementary school-based surveys to annually assess HIV-related knowledge, beliefs, and behaviors among adolescents. Journal of School Health 59:55-58. Kann, L., Nelson, G., Jones, J., and Kolbe, L. (1989) HIV-related knowledge, beliefs and behaviors among high school students in the United States. Presented at the Fifth International Conference on AIDS, Montreal, June 09. Kaplan, H. B. (1989) Methodological problems in the study of psychosocial influences in the AIDS process. Social Science and Medicine 29:277-292. Kaplan, H. B., Johnson, R. J., Bailey, C. A., and Simon, W. (1987) The sociological study of AIDS: A critical review of the literature and suggested research agenda. Journal of Health and Social Behavior 28:14~157. Keeter, S., and Bradford, J. B. (1988) Knowledge of AIDS and related behavior change among unmarried adults in a low-prevalence city. American Journal of Preventive Medicine 4:14~152. Kelly, J. A., St. Lawrence, J. S., Hood, H. V., and Brasfield, T. L. (1989) Behavioral intervention to reduce AIDS risk activities. Journal of Consulting and Clinical Psychology 57:6~67. King, F. W. (1970) Psychology in action: Anonymous versus identifiable questionnaires in drug usage surveys. American Psychologist 25:982-985. Kinsey, A. C., Pomeroy, W. B., and Martin, C. E. (1948) Sexual Behavior in the Human Male. Philadelphia: W. B. Saunders. Kinsey, A. C., Pomeroy, W. B., Martin, C. E., and Gebhard, P. H. (1953) Sexual Behavior in the Human Female. Philadelphia: W. B. Saunders. Kipke, M. D., and Drucker, E. (1988) A method for assessing needle-shanng behavior in intravenous drug users. Presented at the Fourth International Conference on AIDS, Stockholm, June 12-16. Klassen, A. D., Williams, C. J., and Levitt, E. E. (1989) Sex and Morality in the U.S., edited by H. J. O'Gorman. Middletown, Conn.: Wesleyan University Press. Klassen, A. D., Williams, C. J., Levitt, E. E., Rudkin-Miniot L., Miller H., and Gunjal, S. (1989) Trends in premarital sexual behavior. In C. F. Turner, H. G. Miller, and L. E. Moses, eds., AIDS, Sexual Behavior, and Intravenous Drug Use. Washington, D.C.: National Academy Press. Koblin, B., McCusker, J., Lewis, B., Sullivan, J., Birch, F., and Hagen, H. (1988) Racial differences in HIV infection in IVDUs. Presented at the Fourth International Conference on AIDS, Stockholm, June 12-16. Koester, S. (1989a) The risk of HIV transmission from sharing water, drug-mixing con- tainers and cotton filters among intravenous drug users. Unpublished manuscript. University of Colorado School of Medicine, Boulder, Colo. Koester, S. (1989b) When push comes to shove: Poverty, law enforcement and high risk behavior. Unpublished manuscript. University of Colorado School of Medicine, Boulder, Colo. Koss, M. P., and Gidycz, C. A. (1985) Sexual experiences survey: Reliability and validity. Journal of Consulting and Clinical Psychology 53:422~23.

308 ~ APPENDIX C Lange, W. R., Snyder, F. R., Lozovsky, E., Kaistha, V., Jaffe' J. H., et al. (1988) Geographic distribution of human immunodeficiency virus markers in parenteral drug abusers. American Journal of Public Health 78:443- 446. Leach, C., Milker, S., Kuhls, T., Pams, N., Cherry, J., and Christenson, P. (1989) Changes in sexual behavior of a cohort of female health care workers during the AIDS era. Presented at the Fifth International Conference on AIDS, Montreal, June =9. Leonard, T. L. (1990) Male clients of female street prostitutes: Unseen partners in sexual disease transmission. Medical Anthropology Quarterly 4:41-55. Lessler, J., Tourangeau, R., and Salter, W. (1989) Questionnaire design in the cognitive research laboratory. Vital and Health Statistics Series 6, Whole No. 1. Lever, J., Rogers, W. H., Carson, S., Hertz, R., and Kanouse, D. E. (1989) Behavioral patterns of bisexual males in the U.S., 1982. Presented at the Fifth International Conference on AIDS, Montreal, June =9. Levinger, G. (1966) Systematic distortion in spouses' reports of preferred and actual sexual behavior. Sociometry 29:291-299. Lide, D. R., Jr. (1981) Critical data for critical needs. Science 212:1343-1349. Lofts E. F. (1975) Leading nIle.ctinn~ ~nr1 the Inch -~_ I::._ D_.~L_~___ 7:56~572. ——C ~ a—~O lo. ~U8I~~~V~ <~yC~10~ Loftus, E. F., and Marburger, W. (1983) Since the eruption of Mt. St. Helens, has anyone beaten you up? Improving the accuracy of retrospective reports win landmark events. Memory and Cognition 11:11~120. Loftus, E. F., and Palmer, J. C. (1974) Reconstruction of automobile destruction: An example of the interaction between language and memory. Journal of Verbal Learning and VerbalBehavior 13:585-589. LoPiccolo. J. and ~~;~es~rr ~ ~ (] C)7~N Tl~^ ¢~.,^~ ;~ :_ : ~ =,_., _. _. ~^,,~, 4~- o-~ ult~la~ul1 ulv~n[~; A new Instrument for assessment of sexual dysfunction. Archives of Sexual Behavior 3:585-595. Luetgert, M. J., and Armstrong, A. H. (1973) Methodological issues in drug usage sur- veys: Anonymity, recency, and frequency. International Journal of the Addictions 8:683~89. MacKuen, M. B. (1981) Social communication and He mass policy agenda. ~ M. B. MacKuen and S. L. Coombs, eds., More than News: Media Power in Public Affairs. Beverly Hills, Calif.: Sage. MacKuen, M. B. (1984) Reality, the press, and citizens' political agendas. En C. F. Turner and E. Martin, eds., Surveying Subjective Phenomena. VoL 2. New York: Russell Sage. MacKuen, M. B. and Turner, C. F. (1984) The popularity of presidents: 1963 80. C. F. Turner and E. Martin, eds., Surveying Subjective Phenomena. Vol. 2. New York: Russell Sage.. Maddox, J., and Desmond, D. (1975) Reliability and validity of information from chronic heroin users. Journal of Psychiatric Research 12:87-95. Madow, W. G., Nisselson, H., and Ollcin, I., eds. (1983) Incomplete Data in Sample Surveys. Vol. 1, Report and Case Studies; Vol. 2, Theory and Bibliographies; Vol. 3, Proceedings of the Symposium. New York: Academic Press. Magura, S., Goldsmith, D., Casriel, C., Goldstein, P. J., and Lipton, D. S. (1987) The validity of methadone clients' self-reported drug use. International Journal of the Addictions 22:727-749.

METHODOLOGICAL ISSUES ~ 309 Maisto, S. A., and O'Farrell, T. J. (1985) Comment on the validity of Watson et al.'s "Do alcoholics give valid self-reports?" Journal of Studies on Alcohol 46:447-453. Maisto, S. A., Sobell, L. C., and Sobell, M. B. (1982-83) Corroboration of drug abusers' self-reports through the use of multiple data sources. American Journal of Drug and Alcohol Abuse 9:301-308. Marasca, G., D'Arcangelo, E., De Candido, D., Della Giusta, G., Liseo, B., et al. (1989) Sexual behaviour and HIV related knowledge among a random sample of young population of Italy. Presented at the Fifth International Conference on AIDS, Montreal, June 4-9. Marks, I. M., and Sartorius, N. H. (1968) A contribution to the measurement of sexual attitude. Journal of Nervous and Mental Disease 145:441-451. Marshall, P., O'Keefe, J. P., Fisher, S. G., Caruso, A. J., and Surdukowski, J. (1990) Patients' fear of contracting the acquired immune deficiency syndrome from physicians. Archives of Internal Medicine 150:1501-1506. Martin, J. L., and Vance, C. S. (1984) Behavioral and psychosocial factors in AIDS: Methodological and substantive issues. American Psychologist 39:1303-1308. Mason, T. (1988) AIDS prevention among black IV drug users and their sexual partners in a Baltimore public housing project. Presented at the First International Symposium on Infonnation and Education on AIDS, Ixtapa, Mexico, October 18. Mason, T. (1989) The politics of culture: Drug users, professionals, and the meaning of needle sharing. Presented at the Annual Meeting of the Society for Applied Anthropology, Santa Fe, April. Mason, T. (1989) A preliminary look at social and economic dynamics influencing drug markets, drug use patterns, and HIV risk behaviors among injecting drug users in two Baltimore networks. In Proceedings of the Community Epidemiology Work Group: Chicago, Illinois, June 1989. National Institute on Mug Abuse, Division of Epidemiology and Statistical Analysis, Rockville, Md. May, R. M., and Anderson, R. M. (1987) Transmission dynamics of HIV infection. Nature 326:137-142. May, R. M., Anderson, R. M., and Blower, S. M. (1989) The epidemiology and transmission dynamics of HlV-AlOS. Daedalus 118:163-201. McCombie, S. C. (1986) The cultural impact of the AIDS test: The American experience. Social Science and Medicine 23:455~59. McCombie, S. C. (1989) Rituals of infection control among health care workers. Unpublished manuscnpt. Annenberg School of Communications, University of Pennsylvania. McCombie, S. C. (1990) Patterns of condom use in Trinidad and Tobago. Unpublished manuscript. Annenberg School of Communications, University of Pennsylvania. McGlothlin, W. H., Anglin, M. D., and Wilson, B. D. (1977) An Evaluation of the California Civil Addict Program. Rockville, Md.: National Institute on Drug Abuse. McLaws, M. L., McGirr, J., Croker, W., and Cooper, D. A. (1988) Risk factors for HIV and HBV infections in intravenous drug users. Presented at the Fourth International Conference on AIDS, Stockholm, June 12-16. McNemar, Q. (1946) Opinion-athtude methodology. Psychological Bulletin 43:289-374. Means, B., Nigam, A. Zarrow, M., Loftus, E. F., and Donaldson, M. S. (1989) Autobiographical memory for heals related events. Vital and Health Statistics Series 6, Whole No. 1.

310 ~ APPENDIX C Michael, R. T., Laumann, E. O., Gagnon, J. H., and Smith, T. W. (1988) Number of sex partners and potential risk of sexual exposure to human immunodeficiency virus. Mortality and Morbidity Weekly Report 37:565-568. Miller, H. G., Turner, C. F., and Moses, L. E. (1990) AIDS: The Second Decade. Washington, D.C.: National Academy Press. Millstein, S., and Irwin, C. (1983) Acceptability of computer-acquired sexual histories in adolescent girls. Journal of Pediatrics 103:815~19. Minkoff, H., McCalla, S., Delke, I., Feldman, J., Stevens, R., and Salwen, M. (1989) Cocaine use and sexually transmitted diseases including HIV. Presented at the Fifth International Conference on AIDS, Montreal, June =9. Moatti, J. P., Tavares, J., Durbec, J. P., Bajos, N., Menard, C., and Seward, C. (1989) Modifications of sexual behavior due to AIDS in French heterosexual "at risk" population. Presented at the Fifth International Conference on AIDS, Montreal June =9. Mooney, H. W., Pollack, B. R., and Corsa, L. (1968) Use of telephone interviewing to study human reproduction. Public Health Reports 83:1049-1060. Moser, C. A., and Kalton, G. (1972) Survey Methods in Social Investigation. 2d ed. New York: Basic. Moss, A. R., Bacchetti, P., Osmond, D., Meakin, R., Keffelew, A., and Gorter, R. (1989) Seroconversion for HIV in intravenous drug users in San Francisco. Presented at the Fifth International Conference on AIDS, Montreal, June =9. Mukherjee, B. N. (1975) Reliability estimates of some survey data on family planning. Population Studies 29:127-142. Myers, V. (1977a) Survey methods for minority populations. The Journal of Social Issues 33: 11-19. Myers, V. (1977b) Toward a synthesis of ethnographic and survey methods. Human Organization 36:244 251. National Research Council (NRC) (1979) Privacy and Confidentiality as Factors in Survey Response. Washington, D.C.: National Academy of Sciences. Needle, R. H., Jou, S., and Su, S. S. (1989) The impact of changing methods of data collection on the reliability of self-reported drug use of adolescents. American Journal of Drug and Alcohol Abuse 15:275-289. Needle, R., McCubbin, H., Lorence, J., and Hochhauser, M. (1983) Reliability and valid- ity of adolescent self-reported drug use in a family-based study: A methodological report. International Journal of the Addictions 18:901-912. Neisser, U. (1981) John Dean's memory: A case study. Cognition 9:1-22. Newcomer, S., and Udder, J. R. (1988) Adolescents' honesty in a survey of sexual behavior. Journal of Adolescent Research 3:419~23. Newman, R. G., Cates, M., Noun A., and Werbell, B. (1976) Reliability of self- reported age of first drug use: Analysis of New York City narcotics register data. International Journal of the Addictions 11:611-618. Newmeyer, J. A. (1988) Why bleach? Development of a strategy to combat HIV contagion among San Francisco intravenous drug users. In R. J. Battjes, and R. W. Pickens, eds., Needle Sharing Among Intravenous Drug Abusers: National and International Perspectives. DHHS Publication No. (ADM) 88-1567. National Institute on Drug Abuse Research Monograph 80. Washington, D.C.: U.S. Government Printing Office. Nisbett, R. E., and Wilson, T. D. (1977) Telling more than we can know: Verbal reports on mental processes. Psychological Review 84:231-259.

METHODOLOGICAL ISSUES | 311 Novick, D. M., Tngg, H. L., Des Jarlais, D. C., Fnedman, S. R., Vlahov, D., et al. (1989) Drug abuse patterns and ethnicity in IVDA during the early years of the REV epidemic. Presented at the Fifth International Conference on AIDS, Montreal, June =9. Nurco, D. N. (1985) A discussion of validity. In B. A. Rouse, N. J. Kozel, and L. G. Richards, eds., Self-report Methods of Estimating Drug Use: Meeting Current Challenges to Validity. DHHS Publication No. (ADM) 8~1402. National Lnstitute on Drug Abuse Research Monograph No. 57. Washington, D.C.: U.S. Government Printing Office. Getting, E. R., Edwards, R., and Beauvais, F. (1985) Reliability and discriminant validity of the children's drug-use survey. Psychological Reports 56:751-756. O'Reilly, K. R. (1989) Risk behaviors and their determinants. In R. Kulstad, ea., AIDS 1988: American Association for the Advancement of Science Symposia Papers. Washington, D. C.: American Association for the Advancement of Science. Page, B. P., Chitwood, D. D., Smith, P. C., Kane, N., and McBride, D. C. (1990) Intravenous drug use and HIV infection in Miami. Medical Anthropology Quarterly 4:5~71. Page, W. F., Davies, J. E., Ladner, R. A., Alfassa, J., and Tennis, H. (1977) Urinalysis screened versus verbally reported drug use: The identification of discrepant groups. International Journal of the Addictions 12:439~50. Parker, R. (1987) Acquired immunodeficiency syndrome in urban Brazil. Medical Anthropology Quarterly 1:15~175. Parker, R. (1989) Bodies and pleasures in the construction of erotic meaning in contemporary Brazil. Anthropology and Humanism Quarterly 14:504. Pearson, R. W., Ross, M., and Dawes, R. (1989) A theory of personal recall and the limits of retrospection in surveys. Unpublished manuscript. Social Science Research Council, August 1, 1989. Pelto, P., and Pelto, G. (1978) Anthropological Research: The Structure of Enquiry. London: Cambridge University Press. Peterson, J. L., and Bakeman, R. (19~9) AIDS and IV drug use among ethnic minorities. Journal of Drug Issues 19:27-37. Petzel, T. P., Johnson, J. E., and McKillip, J. (1973) Response bias in drug surveys. Journal of Consulting and Clinical Psychology 40:437~39. Podell, L., and Perkins, J. C. (1957) A Guttman scale for sexual experience A methodological note. Journal of Abnormal and Social Psychology 54:420~22. Poti, S. J., Chakraborti, B., and Malaker, C. R. (1960) Reliability of data relating to contraceptive practices. In C. V. Kiser, ea., Research In Family Planning. Princeton: Princeton University Press. Public Heals Service (PHS). (1988) Report of the Second Public Heals Service AIDS Prevention and Control Conference. Public Health Reports 103, Supplement No. Quart, A. M., Small, C. B., and Klein, R. S. (1989) Local destruction of labial surface of mandibular teeth by direct application of cocaine in drug users with AIDS. Presented at the Fifth International Conference on AIDS, Montreal, June ~9. Reining, P. (1989) Male circumcision status in relationship to seroprevalence data in Africa: A review of method. Unpublished manuscript. Department of Anthropol- ogy Catholic University. Research Tnangle Institute (RTI). (1989) National Household Seroprevalence Survey: Pilot Study Report. Research Triangle Park, N.C.: Research Triangle Institute.

312 ~ APPENDIX C Robertson, 3. R., Skidmore, C. A., and Roberts, J. J. K. (1988) HIV infection in intravenous drug users: A follow-up study indicating changes in risk-taking behaviour. British Journal of Addiction 83:387-391. Robinson, T. W., Davies, P., and Beveridge, S. (1989) Sexual practices and condom use amongst male prostitutes in London: Differences between streetworking and non-streetworking prostitutes. Presented at the Fifth International Conference on AII)S, Montreal, June =9. Rodgers, J. L., Billy, J. O. G., and Udry, J. R. (1982) The rescission of behaviors: Inconsistent responses in adolescent sexuality data. Social Science Research 11:28~296 Rolnick, S. J., Gross, C. R., Garrard, J., and Gibson, R. W. (1989) A comparison of response rate, data quality, and cost in the collection of data on sexual history and personal behaviors. American Journal of Epidemiology 129:1052-1061. Rossi, P. H., Wright, J. D., and Anderson, A. B., eds. (1983) Handbook of Survey Research. New York: Academic Press. Rounsaville, B., Kleber, H. D., Wilber, C., Rosenberger, D., and Rosenberger, P. (1981) Comparisons of opiate addicts' reports of psychiatric history with reports of significant-other informants'. American Journal of Drug and Alcohol Abuse 8:51~9. Rouse, B. A., Kozel, N. J., and Richards, L. G., eds. (1985) Self-report Methods of Estimating Drug Use: Meeting Current Challenges to Validity. National Institute on Drug Abuse Research Monograph No. 57. Washington, D.C.: U.S. Government Printing Office. Saltzman, S. P., Stoddard, A. M., McCusker, J., Moon, M. W., and Mayer, K. H. (1987) Reliability of self-reported sexual behavior risk factors for HIV infection in homosexual men. Public Health Reports 102:692-697. Schaeffer, N. C. and Thomson, E. (1989) The discovery of grounded uncertainty: Developing standardized questions about strength of fertility motivation. Center for Demography and Ecology Working paper No. 8~19. Madison, Wisc.: Center for Demography and Ecology, University of Wisconsin. Schiavi, R. C., Derogatis, L. R., Kuriansky, J., O'Connor, D., and Sharpe, L. (1979) The assessment of sexual function and marital interaction. Journal of Sex and Marital Therapy 5:169-224. Schilling, R. F., Schinke, S. P., Nichols, S. E., Zayas, L. lI., Miller, S. O., et al. (1989) Developing strategies for AIDS prevention research with black and Hispanic drug users. Public Health Reports 104:2-11. Schmidt, K. W., Krasnik, A., Brends~up, E., Zof~nan, H., and Larsen, S. O. (1988) Occurrence of sexual behaviour related to the risk of HIV-infection. Danish Medical Bulletin 36:8W38. Schoepf, B. G., Walu, E., Rukarangira, Wn., Payanzo, N.' and Schoepf, C. (In press) Action research on AIDS with women in Central Africa. Social Science and Medicine. Schofield, M. (1965) The Sexual Behavior of Young People. Boston: Little, Brown, and Co. Schurnan, H. and Presser, S. (1981) Questions and Answers in Attitude Surveys: Experiments in Question Form, Wording, and Context. New York: Academic Press.

METHODOLOGICAL ISSUES ~ 313 Seage, G. R., m, Mayer, K. H., Horsburgh, C. R., Cal, B., and Lamb, G. A. (1989) Validation of sexual histories of homosexual male couples. Presented at the Fifth International Conference on AIDS, Montreal, June =9. Siegel, K., and Bauman, L. J. (1986) Methodological issues in AIDS-related research. In D. A. Feldman and T. M. Johnson, eds., The Social Dimensions of AIDS: Method and Theory. New York: Praeger. Single, E., Kandel, D. B., and Johnson, B. D. (1975) The reliability and validity of drug use responses in a large scale longitudinal survey. Journal of Drug Issues 5:426~43. Skinner, H. A., and Allen, B. A. (1983) Does the computer make a difference? Computerized versus face-to-face self-report assessment of alcohol, drug, and tobacco use. Journal of Consulting and Clinical Psychology 51:267-275. Smith, A. F., Jobe, J. B., and Mingay, D. J. (In press) Retrieval from memory of dietary information. Applied Cognitive Psychology. Smith, T. W. (1988) A Methodological Review of the Sexual Behavior Questions on the 1988 General Social Survey (GSS). GSS Methodological Report No. 58, National Opinion Research Center: University of Chicago. Sorensen, J. L., and Leder, D. (1978) Measuring the readability of written information for clients. In G. Landsberg, W. D. Neigher, R. J. Hammer, C. Kindle, and J. R. Way, eds., Evaluation in Practice: A Sourcebook of Program Evaluation Studies from Mental Health Care Systems in the United States. DHEW Publication No. (ADM) 7~763. Washington, D.C.: U.S. Government Printing Office. Sorensen, J. L., Gibson, D., Heitzmann, C., Calvillo, A., Dumontet, R., et al. (1988) Pilot trial of small group AIDS education with IV drug abusers (abstract). In L. S. Harris, ea., Problems of Drug Dependence 1988: Proceedings of the 50th Annual Scientific Meeting, Committee on the Problems of Drug Dependence. National Institute on Drug Abuse, Research Monograph 90. Washington, D.C.: U.S. Government Printing Office. Sorensen, J. L., Gibson, D. R., Heitzmann, C., Dumontet, R., London, J., et al. (1989a) AIDS prevention: Behavioral outcomes with outpatient drug abusers. Presented at the Annual Meeting of the American Psychological Association, New Orleans, La. Sorensen, J. L., Guydish, J., Costantini, M., and Batki, S. L. (1989b) Changes in needle sharing and syringe cleaning among San Francisco Drug Abusers. New England Journal of Medicine 320:807. Sorensen, J. L., Batki, S. L., Gibson, D. R., Dumontet, R., and Purnell, S. (1989c) Methadone maintenance and behavior change in seropositive drug abusers: The San Francisco General Hospital Program for AIDS Counseling and Education (PACE). Presented at the Fifth International Conference on AIDS, Montreal, June =9. Spencer, B. D. (1989) On the accuracy of current estimates of the numbers of intravenous drug users. In C. F. Turner, H. G. Miller, and L. E. Moses, eds., AIDS, Sexual Behavior,ar~Intravenous Drug Use. Washington,D.C.: NationalAcademy Press. Spencer, L., Faulkner, A., and Keegan, J. (1988) Talking About Sex. (Publication P. 5997) London: Social and Community Planning Research. Steger, K., Comella, B., Forbes, J., McLaughlin, R., Hoff, R. A., and Craven, D. E. (1989) Use of a fingerstick paper-absorbed blood sample for HIV serosurveys in intravenous drug users. Presented at the Fifth International Conference on AIDS, Montreal, June =9.

314 ~ APPENDIX C Stephens, R. (1972) The truthfulness of addict respondents in research projects. Inter- national Journal of the Addictions 7:549-558. Sterk, C. (1989) Fieldwork among prostitutes in the AIDS era. In C. Smith and W. Kornblum, eds., In the Field: Readings on the Field Research Experience. New York: Praeger. Stimson, G. V., Donoghoe, M., Alldritt, L., and Dolan, K. (1988a) HIV transmission risk behaviours of clients attending synnge-exchange schemes in England and Scotland. British Journal of Addiction 83:1449-1455. Stimson, G. V., Alldritt, L. J., Dolan, K. A., Donoghoe, M. C., and Lmt, R. A. (1988b) Injecting Equipment Exchange Schemes: Final Report. London: Monitoring Research Group, Goldsmiths' College. Strunin, L., and Hingson, R. (1987) Acquired immunodeficiency syndrome and adoles- cents: Knowledge, beliefs, attitudes, and behaviors. Pediatrics 79:825-828. Sudman, S., and Bradburn, N. M. (1974) Response Effects in Surveys. Chicago: Aldine. Sundet, J. M., Kvalem, I. L., Magnus, P., Grommesby, J.K., Stigum, H., and Bakketeig, L. S. (1989) The relationship between condom use and sexual behavior. Presented at the Fifth International Conference on AIDS, Montreal, June =9. Tracy, P. E., and Fox, J. A. (1981) The validity of randomized response for sensitive measurements. American Sociological Review 46:187-200. Traeen, B., Rise, J., and Kraft, P. (1989) Condom behavior in 17, 18 and 19 year-old Norwegians. Presented at the Fifth Intemat~onal Conference on AIDS, Montreal, June =9. Tross, S., Abdul-Quader, A., Des Jarlais, D. C., Kouzi, A., and Friedman, S. R. (1989) Dete~ll~inants of sexual risk reduction in female IV drug users recruited from the street. Presented at the Fifth International Conference on AIDS, Montreal, June =9. Tucker, C., ~trano, F., Miller, L., and Doddy, I. (1989) Cognitive issues and research on the consumer expenditure diary survey. Presented at the 1989 Annual Meetings of the American Association for Public Opinion Research, St. Petersbllrg, Fla., May. Turner, C. F. (1978) Fallible indicators of the subjective state of the nation. American Psychologist 33:456~70. Turner, C. F. (1984) Why do surveys disagree? Some preliminary hypotheses and some disagreeable examples. In C. F. Turner and E. Mariin, eds., Surveying Subjective Phenomena. Vol. 2. New York: Russell Sage. Turner, C. F. (1989) Research on sexual behaviors that transmit HIV: Progress and problems. AIDS 3:563-569. Tu~ner, C. F., and Fay, R. E. (1987/1989) Monitor~ng the spread of HIV infection. Background paper for ad hoc advisory group, Centers for Disease Control. Atlanta, Gal, July 7, 1987. Reprinted in C. F. Tllrner, H. G. Miller, and L. E. Moses, eds., (1989) AIDS, Sexual Behavior, and Intravenous Drug Use. Washington, D.C.: National Academy Press. Tu~ner, C. F., and Martin, E., eds. (1984) Surveying Subjective Phenomena. Two volumes. New York: Russell Sage. Turner, C. F., Miller, H. G., and Barker, L. F. (1989) AIDS research and the behavioral and social sciences. In R. Kulsad, ea., AIDS, 1988: AAAS Symposium Papers. Washington, D.C.: American Association for the Advancement of Science. Turner, C. F., Miller, H. G., and Moses, L. E., eds. (1989) AIDS, Sexual Behavior, and Intravenous Drug Use. Washington, D.C.: National Academy Press.

METHODOLOGICAL ISSUES 315 Udry, J. R., and Morris, N. M. (1967) A method for validation of reported sexual data. Journal of Marriage and the Family 29:442~46. Vessey, M. P., Johnson, B., and Donnelly, J. (1974) Reliability of reporting by women . . taking part In a prospective contraceptive study. British Journal of Preventive and Social Medicine 28:104 107. V~nokur, A., Oksenberg, L., and Cannell, C. F. (1977) Effects of feedback and reinforcement on the report of health information. In C. F. Cannell, L. Oksenberg, and J. M. Converse, eds., Experiments in Interviewing Techniques. Washington, D.C.: National Center for Health Surveys Research. Wagennar, W. A. (1986) My memory: A study of autobiographical memory over six years. Cognitive Psychology 18:225-252. Waterton, J. J., and Ouffy, J. C. (1984) A comparison of computer interviewing tech- niques and traditional methods in the collection of self-report alcohol consumption data in a field survey. international Statistical Review 52:17~182. Watson, C. G. (1985) More reasons for a moratorium: A reply to Maisto and O'Farrell. Journal of Studies on Alcohol 46:450~53. Watters, J. K. (1989) Observations on the importance of social context in HIV transmission among intravenous drug users. Journal of Drug Issues 19:9-26. Webb, E. J., Campbell, D. T., Schwartz, R. D., and Sechrest, L. (1966) Unobtrusive Measures: Nonreactive Research in the Social Sciences. Chicago: Rand McNally. Weibel, W. W. (1988) Combining ethnographic and epidemiologic methods in targeted AIDS interventions: The Chicago model. In Needle Sharing Among intravenous Drug Abusers: National and international Perspectives. National Institute on Drug Abuse Monograph 80. Washington, D.C.: U.S. Government Printing Office. Wells, J. A., Wilensky, G. R., Valleron, A. J., Bond, G., Sell, R. L., and DeFilippes, P. (1989a) Population prevalence of AIDS high risk behaviors in France, the United Kingdom and the United States. Presented at the Fifth International Conference on AIDS, Montreal, June =9. Wells, J. A., Sell, R. L., Will, A., and DeFilippes, P. (1989b) Readability analysis of AIDS brochures and pamphlets from the United States. Presented at the Fit Intemational Conference on AIDS, Montreal, June =9. Wermuth, L., and Ham, J. (1989) Women don't wear condoms: Coping with AIDS risk among women partners of intravenous drug users. Presented at He Annual Meeting of the American Sociological Association, San Francisco, Calif. Whitehead, P. C., and Smart, R. G. (1972) Validity and reliability of self-reported drug use. Canadian Journal of Criminology and Corrections 14:83~9. Wigdor, A. K., and Garner, W. R., eds. (1982) Ability Testing. Two vole. Washington, D.C.: National Academy Press. Wiley, J. (1989) Studying Nonresponse in the AMEN Cohort Survey (unpublished notes). Survey Research Center, University of California at Berkeley, July. ~nkelstein, W., Jr., Samuel, M., Padian, N. S., Wiley, J. A., Lang, W., et al. (1987a) The San Francisco Men's Health Study. m. Reduction in human im- munodeficiency virus transmission among homosexuallbisexual men, 1982-1986. American Journal of Public Health 77:685~89. W~nkelstein, W., Jr., Samuel, M., Padian, N. S., and Wiley, J. A. (1987b) Selected sexual practices of San Francisco heterosexual men and risk of infection by the human immunodeficiency virus. Journal of the American Medical Association 257:147~1471.

316 ~ APPENDIX C Winkelstein, W., Jr., Lyman, D. M., and Padian, N. S. (1987c) Sexual practices and risk of infection by the AIDS-associated retrovirus: lithe San Francisco Men's Health Study. Journal of the American Medical Association 257:321-325. Wish, E., Johnson, B., Strug, D., Chedekel, M., and Lipton, D. (1983) Are Urine Tests Good indicators of the Validity of Self-Reports of Drug Use? It Depends on the Test. New York: Narcotic and Drug Research, Inc. Wolf, J. S., Wodak, A., and Guinan, J. (1989) The effect of a needle and syringe exchange on a methadone maintenance unit. Presented at the Fifth International Conference on AIDS, Montreal, June =9. Walk, J. S., Wodak, A., Guinan, J. J., Morlet, A., Gold, J., et al. (1988) HIV seroprevalence in syringes of intravenous drug users using syringe exchanges in Sydney, Australia, 1987. Presented at the Fourth International Conference on AIDS, Stockholm, June 12-16. Woodward, J. A., Retka, R. L., and Nig, L. (1984) Construct validity of heroin abuse estimators. international Journal of Addictions 19:9~117. World Health Organization: Global Programme on AIDS (1988) Progress Report No. 4. Geneva: World Health Organization. Worth, D. (1989) Sexual decisionmaking and AIDS: Why condom promotion among vulnerable women is likely to fail. Presented at the Population Council, New York. March 7. Wyatt, G. E., and Peters, S. D. (1986) Methodological considerations in research on the prevalence of child sexual abuse. Child Abuse and Neglect 10:241-2S1. Youden, W. (1961) How to evaluate accuracy. Materials Research and Standards I:26~271. Youden, W., and Steiner, E., eds. (1975) Statistical Manual of the Association of Official Analytical Chemists. Washington, D.C.: Association of Official Analytical Chemists. Zelnik, M., and Kantner, J. F. (1980) Sexual activity, contraceptive use and pregnancy among metropolitan-area teenagers: 1971-1979. Family Planning Perspectives 12:23~237. Zeugin, P., Dubois-Arber, F., Hausser, D., and Lehmann, P. H. (1989) Sexual behaviour of young adults and the effects of AIDS-prevention campaigns in Switzerland. Presented at the Fifth International Conference on AIDS, Montreal, June 4-9 . Zich, J., and Temoshok, L. (1986) Applied methodology: A pruner of pitfalls and opportunities in AIDS research. In D. A. Feldman and T. M. Johnson, eds., The Social Dimensions of AIDS: Method and Theory. New York: Praeger. Zuckerman, M. (1973) Scales for sex experience for males and females. Journal of Consulting and Clinical Psychology 41:27-29. Zuckennan, M., Tushup, R., and Finner, S. (1976) Sexual attitudes and expenence: Attitude and personality correlates and changes produced by a course in sexuality. Journal of Consulting and Clinical Psychology 44:7-19. Zuckerman, B. S., Hingson, R. W., Morelock, S., Amaro, H., Frank, D., et al. (1985) A pilot study assessing maternal marijuana use by urine assay during pregnancy. In B. A. Rouse, N. J. Kozel, and L. G. Richards, eds., Self-report Methods of Estimating Drug Use: Meeting Current Challenges to Validity. DHHS Publication No. (ADM) 85-1402. National Institute on Drug Abuse Research Monograph No. 57. Washington, D.C.: U.S. Government Printing Office.