Issues in Controlled Substance Use: Papers and Commentary, Conference on Issues in Controlled Substance Use (1980)

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.

TOWARD THE ACQUISITION OF DATA ON CONTROLLED SUBSTANCE USE Peter K. Levison and Albert J. Stunkard Introduction You will not have difficulty convincing most people that compul- sive users of opiates or alcohol are "out of control." It is implicit in the concept of "out of control" in this context that the substance exercises such compelling effects that there are insufficient forces "inside" or "outside" the victim to restrain excessive use. In the same terms, the moderate user is regarded as able to limit consumption to "safe" amounts without great difficulty and to adjust patterns of use to situational requirements. This common-sense formulation assumes that there are factors in the moderate user's individual development, current experience, or knowledge about drugs and environment that ensure the rational self-management of substance use so that it does not get "out of control." This common-sense view is not quite ade- quate; the out-of-control addict has periods—sometimes long ones—of refraining from use. And many "moderate" users have occasions when the amount taken and its effects under the existing conditions produce NOTE: We are indebted to the following people for their help with this paper: to the members of the work group of the Committee on Substance Abuse and Habitual Behavior—who had a primary role in the original shaping of the ideas presented here—Peter Dews, Jerome Jaffe, Stanley ^ Schachter, Thomas Schelling, and Frank Stanton; to the group of ad- visers who generously gave us their time and thoughtful suggestions at a meeting in Boston—Jack Elinson, Roger Meyer, Kenneth Rothman, and George Vaillant, and especially to Elinson and Rothman for their addi- tional help after the meeting; to Abraham Lilienfeld for his willing and valuable advice; to Donald Treiman for his helpful suggestions on secondary analysis; and especially to committee staff members Deborah Maloff, Dean Gerstein, and Arlene Fonaroff for their hard intellectual and editorial work on drafts of the paper. The authors, however, take full responsibility for the final form and deserve any blame that is due for errors of fact or judgment. The paper in this volume has been modified considerably from the original presented at the conference. 47

48 unwanted results or the risk of them. In other words, control factors operate in all substance use. Scientific analyses of control factors will proceed best if we look for common controls among all patterns of substance use behaviors and try to understand why they are effective in some contexts and have weakened or failed in others. What are control factors? The psychotherapist and the biologist have traditionally looked inside the individual: the former at atti- tudes, coping skills, personality traits, motivational states, and the like; the latter at physical dependence, metabolism, side effects, genetic factors, and more recently, neurochemical receptors and trans- mitter systems. Social scientists have focused on variables outside the individual such as cultural practices, group norms, family and peer pressures, stratification, and the social function of substance use. Some behavioral scientists have attempted to integrate internal factors such as past learning, conditioning, and reinforcement with current environmental signals. Policy studies have emphasized availability, legality, enforcement, price, etc. in the context of governmental options for controlling use. The categories in the foregoing lists are not necessarily con- trols in themselves. However, the categories specify variables that are supposed to increase or decrease the probabilities of substance use, and in this sense contribute to the control of habitual behaviors. This paper will deal with the acquisition of knowledge about control factors and the ways in which policy changes can encourage development of a fuller knowledge base in this important area. Specifi- cally, we will discuss (l) the need for far more data on control factors in substance use in the general population as compared with that in special user populations from which most of our information has been derived; (2) the present possibilities and limitations for acquiring such data; and (3) the ways in which relevant policies might be changed to encourage data acquisition. The substances that are of the immedi- ate concern include tobacco, alcohol, narcotic drugs, and food, but the discussion framework does not exlude others. This paper does not intend to exclude any category of potential controls from its argument. However, it emphasizes the need for more information on situational variables, learned patterns of substance use, and individual strategies for controlling use. Thus, it emphasizes both self-control and collective control. Still, the individual user in each case is the starting point for the inquiry. Background The Committee on Substance Abuse and Habitual Behavior has a strong interest in encouraging the investigation of common properties in the use of substances and in associated patterns of habitual be- havior. In this context, the study of control factors is important because of its centrality in problems of substance abuse. The idea for this paper originated with a work group of the committee interested in control factors outside formal treatment settings. For complex histor- ical reasons, treatment for narcotic drugs and, more recently, for

49 alcohol problems has been a focus for government intervention and support. As a consequence, treatment programs tend to be patterned after a relatively limited set of models that are publicly endorsed and eligible for support with public funds. In order to escape these constraints and to get an overview of the whole problem, the work group decided to emphasize control factors outside "official" treatment settings. In particular, it focused on individuals who enter organized self-help programs such as Weight Watchers, SmokEnders, and Alcoholics Anonymous. It soon became apparent that there are huge gaps in our information on the substance use histories of individuals joining such groups, the circumstances that led to completion of or attrition from the programs, the post-treatment levels of use, and the long-term maintenance of the effects. However, even the large number of persons in self-help groups represents only a small fraction of individual efforts in the general population to change substance use behaviors. How representative are the self-help group members of users in the general population? For example, what proportion of the millions of Americans who have stopped smoking cigarettes in the past 20 years did so on their own, relative to those who receive organized help? In pursuing these questions, the work group immediately encoun- tered problems in data acquisition. Organized self-help groups are all too often unwilling to share data with outsiders. Part of their success in recuiting members depends on a public image of success in dealing with a problem with which most prospective members have suffered repeated failures. But objective studies of treatment outcome by disinterested researchers have found that self-help programs are considerably less effective than is generally believed. Moreover, those profit-making programs that are enjoying monetary success have little to gain from any evaluation and perhaps much to lose. In any case, it became clear to the work group that, irrespective of the data acquisition problems particular to self-help groups, knowledge of substance use behaviors in the general population is limited. The Needs for Data on Control Factors A major problem in assessing research on substances and substance users is the fact that the research has focused on extremes. Opiate addiction and alcoholism have been focal points for studies that des- cribe histories of use, lifestyles, and attempts to control use through treatment (e.g., Hughes, l977; Preble and Casey, l972; Wiseman, l970). Studies that include more moderate users, mainly national surveys aimed at prevalence estimates, still tend to reflect primary interest in the excessive user (for example, Abelson et al., l977). The extensive literatures on treatment evaluation have been primarily concerned with how well overuse can be changed to non-use, abuse to abstinence. Largely overlooked have been those factors that are important for understanding controlled substance use and sources of control. These may be studied in a variety of contexts: biological and psychological susceptibilities and immunities; individual decision making about use; social settings and pressures; access to substances; and formal or

50 informal sanctions on use. In whatever domain we chose to examine, we found insufficent information about what promotes and what retards substance use, why individuals differ in their degress of control over excessive use, and what strategies are suggested by this information toward prevention of or successful intervention in excessive use. There are several arguments for the acquisition of additional data on control factors in substance use. The state of any field is in part dependent on its points of departure. Current sociobehavioral research about substance users appears to be in an empiricist phase, little influenced by major theories, such as the psychoanalytic model of the addictive personality provided in earlier days. In this theoretical vacuum, hypotheses about habitual use tend to be derived entirely from work on the most thor- oughly researched population, the very special and biased sample of excessive users participating in organized treatment programs. How- ever, those designated as patients in treatment contexts may offer particularly unpromising sources for theories about how control factors operate in the lives of different kinds of people with different kinds of problems. In the attempt to learn about general principles governing con- trols in the population, more valuable information may be obtained from relatively moderate users, or even largely abstaining populations, than from treatment populations. One strategy could be to compare the same control factors among abstainers, light, moderate, and heavy users, and addicts. As we know, classification according to amount and frequency of use is a difficult problem per se. Studying levels and patterns of use in the context of control factors may be a productive way to classify risks. Studies of the general population designed to examine factors in the control of substance use might be strengthened by incorporating a cross-substance perspective. Support of research by categorical agencies (each concerned with only one class of substances) has pro- duced literatures with little interest in cross-substance problems, and valuable information has been lost in the process. Studies in multiple substance use have largely involved polydrug users and the association of alcohol with heroin use patterns, but there are many other interesting possibilities: the heavy use of tobacco and coffee in Alcoholics Anonymous meetings has often been noted; cessation of cigarette smoking predictably leads to increases in body weight. Multiple substance use should be fertile ground for discovering common control factors. Adequate information on control factors is important for public policy. Current debate over marijuana laws provides an example. Some data suggest that marijuana intoxication may be increasingly associated with driving accidents (Sterling-Smith, l976). From these data alone, a policy of tightening controls over marijuana supply might be indi- cated. But suppose, for example, that studies on control factors show that thrill seeking or risk taking is characteristic of the people principally responsible for the increase in accidents, and that these individuals also use alcohol and are willing, even likely, to sub- stitute it if marijuana is difficult to obtain. Then policy would more

5l effectively be focused on driving practices rather than on the availa- bility of the substance; one might choose to increase deployment of traffic police rather than drug enforcement officers. Data on control factors would help to make appropriate distinc- tions among users in the general population with respect to the hazards of use and, conversely, the effectiveness of controls. Studies of many kinds are needed to discover how much of what substance is enough or too much for an individual in a situation. Longitudinal studies could be important, for an inquiry confined to a single period in a user's life may be misleading. A moderate amount of a substance is too much for some users if it is a point on an increasing dosage schedule that is leading to severe dependence problems. Societal and individual values are central in substance use issues. Although there is societal consensus that addicted states are bad, more intermediate levels of use are the source of considerable disagreement. The President of the United States has banned hard liquor from White House social events, but his brother has become a national symbol of the virtues of the male beer-drinking friendship group. Moderate use for recreational purposes is generally condoned and even encouraged in user populations of, for example, alcohol and marijuana. However, regulatory policies and intervention programs aimed at curtailing excessive use necessarily have some restraining effects upon recre- ational users as well. Value positions held by groups and individuals are important control factors in substance use, particularly for defining temperance in use. Value issues come to the forefront when policies or programs are being formulated to control substance use. More research alone will not provide complete resolution of conflicting value positions. But better knowledge of user behavior, including circumstances affecting use in the great majority who are not addicts, can help frame more rational policies and more effective interventions. Knowledge of the incidence in the general population of spontane- ous changes in patterns of use, such as stops and starts, periods of prolonged abstinence, self-administration of special techniques for controlling use, or controlled use as a replacement for heavy use, are necessary to understand and evaluate the effects of treatment. One might assert that almost all "true addicts" are netted by treatment programs and that the remainder of the user population is different in kind (i.e., have less severe problems). Without evidence from well- designed studies using appropriate no-treatment control groups, the argument is circular: The worst problems are in programs; others are not so badly off or they would be in treatment too. The required control groups may be almost impossible to construct under present policies, for reasons of financial constraints, time limitations, and ethical issues in the assignment of troubled persons to no-treatment groups. Therefore, good estimates of spontaneous changes in the general population are necessary to evaluate the size of the apparent treatment effects of normal programs. If a smoking cessation program, for example, is no more successful than spontaneous cessation in the client's population subgroup, then the presence of a treatment effect

52 is questionable. (The basis on which clients will be assigned to such subgroups is a very important problem itself, however.) The work group's original interest was in methods for determining the effectiveness of individual attempts to terminate or reduce sub- stance use in the general population. This line of inquiry has an advantage in that the individual provides the definition of whether the amount and/or the frequency of use is excessive. We are not arguing that self-judgments are always preferred to those made by others or those based on objective indices, but a user has unique access to some of the effects of substances as well as their use in private settings. Self-definitions of excessive use and self-descriptions of controls and how and when they fail are therefore important sources of data on control factors. Studies to learn how individuals or groups are self- regulating their use of substances are important for understanding more about how to proceed with policies and programs to alert individuals to inherent risks and to control excessive use. Studies are also essential to better define limits of use that, if exceeded, greatly increase risks, such as health problems (including accidents), increasing physical dependence on a substance, decrements in job performance, or neglect of social responsibilities. Because most existing information about patterns of substance use comes from addicted individuals, who may be atypical of the general user popula- tion along some psychological and demographic dimensions, studies of control factors and their effects in the general population are needed to estimate risks—especially the less catastrophic ones—in the general population. For years heroin addiction was considered virtually irreversible because Lexington patients were believed to represent the universal heroin user in the final phase of the addiction process. We know now that there is a spectrum of intensity of involvement with heroin, and even the heavy user population includes subgroups that have prospects for stopping drug use spontaneously. There appear to be latent control factors in the life circum- stances of some addicts that must be discovered or further analyzed. Demographic variables appear to be better predictors of recovery and long-term abstinence than are particular treatment program variables. Those who start using heroin relatively late in their lives and/or have stable family and employment backgrounds are more likely to remain abstinent after treatment than those who started early and are in unstable environments. In Lee Robins's (l974a) studies of returning Vietnam veterans who became addicted in the service, the history of early experimentation with drugs, including alcohol, was the best predictor of continued use upon returning home. It would be valuable to know whether in general early exposure to drug use is a major factor in predicting later light vs. heavy use patterns. Also, more detailed knowledge about the lives of those individuals who have been heavy users but have managed to avoid chronicity can contribute substantially to understanding the role of limits preventing loss of control in different individuals (Zinberg et al., l977). Studies of addicts show that social and physical settings have strong effects on patterns of use and even on physiological responses

53 that accompany the experience of craving. The ability of heroin users to remain abstinent is strengthened when they are out of the environ- ment in which heroin is obtained. Studies of family and ethnic group drinking practices are important for understanding the effects of settings and informal learning experiences on moderate use. Signifi- cant as they are, these studies emphasize comparisons with substance abusers or intergroup comparisons of consumption per se. Studies that provide more detail and deeper analyses of the operation of controls in these settings are needed. For example, we know little about how individuals control their own tendencies toward excessive use by avoiding places conducive to substance use. Under what circumstances will heavy social drinkers actively avoid cocktail parties and bars as a self-control procedure or will obese individuals avoid smorgasbord restaurants? Another consequence of the paucity of accurate normative data on control factors, especially on the self-regulation of use, is that policies regarding excessive use of substances have been dominated by abstentionist thinking. Excessive use is emphasized as an issue for a problem group of people who can avoid being out of control only by complete abstinence from the substance (or categories of it in the case of food). Treatment episodes are regarded as failures unless the subject abstains for an extended period, usually more than a year after discharge from the treatment program. High probabilities of relapse by people in formal treatment programs are emphasized. The political storm following the publication of the Rand report (Armor et al., l976), which suggested that some former alcohol abusers could success- fully practice controlled drinking, demonstrates the strength and depth of these ideas. However, better understanding of general principles involved in controlling use could provide treatment alternatives to lifelong abstinence that might be beneficial to some excessive users. The demoralization accompanying repeated failures when perpetual ab- stinence is held out as the only acceptable alternative probably contributes to the maintenance of compulsive patterns in some users. The prevalence of the belief that the first misstep for the addict leads directly to a new cycle of addiction may be a self-fulfilling prophesy. Studies on control factors might reveal self-control princi- ples that could be taught to compulsive users so that an occasional misstep would not upset a general balance in which abstinence is largely maintained. The abstentionist, all-or-nothing view of success overlooks the real benefits that may be obtained from even temporary periods of abstinence or reduced use. Substances, such as tobacco and alcohol, that have known damaging effects when used excessively may, beyond some threshold point, produce cumulative damage with each episode of use. Any period of abstinence of diminished use probably flattens this cumulative effect.

54 Issues in Data Acquisition We have listed a general set of issues in the control of sub- stance use that deserve more attention. It is now our responsibility to look ahead to the requirements for serious work—that is, conceptu- alizing problems, specifying variables, and selecting research strategies, taking into account such factors as the availability of resources, existing information, time frames, requirements for pre- cision, and the like. We need to be especially clear about populations and settings selected for study, the methodologies to be used, and how well our proposals mesh with ongoing programs of sponsored research. Populations and Settings: In populations in which the use of a substance is negligible, control factors may, like the air we breathe, be so ubiquitous that analysis of specific control effects is very difficult. For example, among middle class white American adults over 30, legal controls keep the heroin marketplace at a distance; sellers will not take risks in so unfavorable a market. Even among those who habitually seek thrills and risks, anticipated disapproval of one's peers, exchanges of beliefs about heroin's dangers, and risks of detec- tion and punishment abort any tendencies to start experimenting with heroin. On the other hand, in populations where heavy use is normative and controls are minimal, for example, cigarette smoking in the United States in the post-World War II period, studies of controls are unlikely to provide clear results. A promising focus for studies of control factors is those popula- tions in which patterns of substance use are quite varied, such as occasional use in recreational settings, moderate use associated with specific rituals, periodic occasions when the usual controls are de- clared null and void, violations of norms in situations in which use is disapproved. There is inadequate information on situations in which the effect of use may markedly alter ongoing social processes or the performance of important tasks and conditions that alert individuals to exercise self-control. Methodologies: Because of the paucity of systematic data on control factors, surveys and field studies may play a prominent part in developing a knowledge base. Moreover, there appears to be a large number and variety of important controls that affect individual sub- stance use behavior. Surveys and field studies are well suited for the initial sorting out of this complexity; many different situational variables, individual characteristics, and consequences of use can be analyzed in the framework of a single, properly designed study. This emphasis does not rule out parallel programmatic support for other methodologies such as laboratory experiments on controls, studies of individual differences in self-control, response to imposed external controls, or clinical studies of learning new controls. However, large sample surveys and field studies will be likely to generate new hypotheses to be tested in experimental or clinical paradigms. The analysis of clinical histories that include rich descriptions of pathways to excessive use are also valuable, but addicts' perspec- tives on their past are strongly influenced by the overwhelming fact of

55 the current addiction. The importance of having a personally satisfy- ing explanation for one's present condition often colors the past too strongly. A generation of scholars at Lexington may have been misled about the nature of heroin addiction on this basis. Emphasis on surveys and field studies as a starting point for programmatic support of research on control factors is also based on policy considerations. Agencies responsible for research on substance use patterns and/or health effects have invested heavily in large surveys and epidemiological studies, so this approach is well estab- lished. Also, political pressures to catalog the national experience in substance use ensure the continuation of large-sample studies. Strategy: We hope to persuade potential sponsoring agencies and investigators in substance abuse of the importance of more work on control factors. But we cannot be so optimistic as to believe that large new sums will be appropriated for such studies. Nor does it appear likely—even if desirable—that many existing resources for studying substance abuse will be redeployed to follow these recommen- dations. Therefore, it is important to review the options that do exist, to suggest how studies of control factors can take advantage of existing data and ongoing studies. The sections that follow provide a brief review of study alternatives and discuss their strengths and weaknesses. Design Alternatives New Studies: The most desirable alternative, time and cost considerations aside, is to design new studies specifically aimed at important questions on control factors. Good studies can reveal those control factors that, when present, are powerful in preventing ex- cessive substance use. It is necessary to examine a variety of populations and situations to separate robust controls from those that are strong only within the fabric of very special controlling environ- ments. Speculatively, the former might include sound health training combined with salient warning labels or other reminders; the latter might include anti-substance sentiments among members of a highly structured, isolated religious community. As national policy, it seems best to emphasize controls that are effective in the widest variety of situations and that follow individuals from one environment to another. Reanalysis of existing data: Although new studies are the preferred strategy, they are also usually the most expensive and time-consuming—and we may need answers relatively quickly for policy reasons. Under these circumstances we can reanalyze existing data sets that promise to yield information on control factors. These data sets may generate useful hypotheses about habitual behavior patterns and sources of control and permit the testing of analytic models. There are two limitations to exploiting existing data sets, however. One is that they are fixed; completed studies cannot be modified to build on their own results. Second, questions on substance use have usually been constructed from different perspectives, for other objectives than those of the current investigators.

56 Utilization of existing data has a practical drawback that is well known to practitioners of secondary analysis: researchers who design, obtain sponsorship for, and carry out studies have an under- standably proprietary interest in the data. "Mining" a large data set may provide substantial professional rewards for the principal investi- gators, their colleagues, and/or their students. "Ownership" of a data set may be a sensitive subject, particuarly when the study has been sponsored by a government agency and may include agency employees as collaborators. Access to data for secondary analysis may depend on such factors as pre-existing collegial relationships, mutual respect among investigators (no "owner" wants to see his or her data sullied by incompetent work), or a contractual agreement in which the original sponsor may require data sharing. On the positive side, the stature of an investigator may be enhanced by the frequent literature citations to the original study that are a by-product of secondary analyses. Re-use of the data also testifies to the confidence of other scientists in the soundness of the primary study design and execution. Piggybacking: The combined advantages of the relatively low cost of using existing studies and of the flexibility of primary data .acquisition can be obtained by inserting questions on control factors in substance use in ongoing, repeated surveys. The best-known example is the decennial national census. However, principal investigators for surveys may put up barriers to access by any new inerests. An example of productive coupling of different research interests is a study by Parry and associates (l974), in which questions on alcohol use as a coping mechanism were inserted in a national survey concerned primarily with psychotherapeutic drugs sponsored by the National Institute of Mental Health (NIMH). Retesting: If an existing set includes adequate information on substance use to be the basis of a study on control factors, and if the data is dated enough to make another effort worthwhile, re-examination of the sample population may be an efficient way to assess longitudinal trends. Also, new items on substance use could be added, including retrospective questions, to supplement the information obtained in the original interviews. In the latter case, both the original and the retrospective data might be combined to increase the information available on the first point(s) in the time series. A major risk with retesting an old sample is loss of sample size. Experience with relatively small samples such as O'Donnell's (l969) shows that almost the entire sample can be recaptured if the investigators are persistent. The practicality of this option declines with increasing sample size and/or geographic dispersion. Migration is a complicating factor. Some populations are very mobile and the longer the interval between interviews, the greater the likelihood of address changes. A decision on whether to follow up an established sample to obtain longitudinal information on substance use will depend in part on the costs and the difficulties involved. However, if it appears important to recover most of the sample in order to control bias due to subject loss, one can estimate these effects at relatively little ex- pense before proceeding. For example, if an orginal sample size is

57 l0,000, and if 8,000 of these are relatively easy to recover but 2,000 are very difficult to recover (for example, no known current address), one can draw a sample of 200 from the 2,000, go to the trouble of finding them, and study them thoroughly to determine if they differ significantly from the other 8,000. Examples of Potential Data Sources There are many data sets and continuing studies with potential for the investigation of issues in substance use. These include open- ended, ongoing longitudinal studies and ongoing cross-sectional studies. Examples of such studies follow. The Health and Nutrition Examination Survey (HANES) and the Health Interview Survey (HIS) are sponsored by the National Center for Health Statistics and conducted by the Census Bureau (USPHS, l976). Other health surveys based on interests in cancer or cardiovascular diseases may also be valuable for substance use research since causal factors in these diseases may include excessive use of such substances as tobacco, alcohol, and food. Therefore, some questions on related habits are included. Also, the data sets frequently represent large national samples and may include the results of medical examinations. Examples of such data are the American Cancer Society panel; the Rosewell Park/University of Buffalo admissions data for all suspected cancers between l956 and l965; and the cardiovascular epidemiology field studies sponsored by the National Institutes of Health (NIH) over the past three decades (for example, Dawber et al., l95l). The NIH field studies include populations in Framingham, Massachusetts (for which there are some data tapes available); Tecumsah, Michigan; Los Angeles, California; Albany, New York; Evans County, Georgia; Puerto Rico; Hawaii; Israel; and Yugoslavia. Caroline Thomas's longitudinal studies of Johns Hopkins University medical students is a very complete set of data on a highly selected population. The NIMH national survey on the acquisition and use of psychotherapeutic drugs and psychotropic substances is a cross- sectional study (l970-l97l) with importance for substance use research. Surveys on smoking behavior are sponsored by the National Clearinghouse for Smoking and Health. Data collection systems that emphasize substance abuse are spon- sored by a number of government agencies. An example is the Drug Abuse Epidemiology Data Center (DAEDAC) (l976) at Texas Christian University, which is supported by the National Institute on Drug Abuse. DAEDAC maintains the Drug Abuse Reporting Program, which receives treatment- related data from many programs in the field. Issues in Selecting Design In principle the basic issues in the study design are simple to sort out. Questions that bear on development, maintenance, termina- tion, and relapse in habitual patterns of use, particularly multiple

58 substance use and substitution, require longitudinal data. Questions related to demographic and regional differences, changing substance use practices (including shifts, for example, from one recreational drug to another), the effectiveness of prevention programs, and the frequency of attempts to reduce use can be addressed most efficiently in cross- sectional studies. Retrospective designs, on the other hand, can encompass the broadest range of issues with the least expense, includ- ing both current practices and longitudinal perspectives, but one risks a loss in the accuracy of reports for many substance use issues. In practice, however, the selection of design is complicated by cost, time, and other practical considerations. Accurate description and classification of substance use patterns and control factors over different phases in the life cycle of a user are particularly important for prevention, regulation, and treatment policies. Therefore, longitudinal studies assume great importance for research on control factors in substance use. Vaillant's (l973) and Robins's (l974a; l974b; Robins and Murphy, l967) studies on heroin users are examples of how prospective methods helped to destroy the myth that narcotic addicts never recover and helped to pave the way for more permissive treatment strategies. However, the pace at which a pattern emerges in a prospective study may not be suited to the needs and demands of rapidly changing policy arenas. An illustration of cost, time, and purpose issues in selecting longitudinal designs is provided in the following discussion of three basic epidemiological designs. Prospective panel study. This is the most expensive, time- consuming method and should only be considered if the other designs prove infeasible. Unlike the other two, its major problem is the delay in waiting to measure whether events of interest occur. Retrospective panel follow-up. In this method, a sample is selected from past records and a follow-up survey employs retrospective recall to establish biographical events. In this case the events of interest have already occurred, and no lengthy follow-up period is required. The major expense is in the detective work necessary to locate the subject. Cross-sectional retrospective study. This design involves inter- viewing a group of people and learning about the natural history from those among them who were once users based on aided recall of events. A major problem with this method is the loss of individuals by death, migration, etc. An interesting possibility would be to combine the second and third approaches. For example, a series of former users could be identified, perhaps thorugh data from the National Center for Health Statistics. These individuals could be followed up and, at specified intervals, interviewed for a history of their substance use behavior. Comparison of the current interview data with the original records would serve as a built-in check on the accuracy of recall. Those who died before the follow-up could be included and information about them obtained from friends or relatives. Despite drawbacks to prospective studies described above, there are distinct advantages for life cycle studies of substance use.

59 Periodic data collection has a positive effect on maintaining the stability of the sample and the process becomes institutionalized, making the subject part of a reliable reporting system. Except for deaths, it is not unusual for almost entire samples to remain intact over many years. Advantages include gathering of periodic health and other data that relate to substance use. Demographic changes and critical life events, particularly major decisions to alter levels of use, can be assessed in relation to recorded changes in substance use patterns with an accuracy and completeness not likely to be obtained from data that depends on subject recall of remote events. For some purposes, composite life cycle profiles assembled from different age cohorts can be used to simulate true longitudinal data, for example, in estimating the extent of the "maturing out" of heroin use. However, such simulations are risky. For example, Jellinek's (l952) influential description of phases in the development of alcohol- ism appears to have been based on composite, cross-sectional clinical data. Subsequent factor analytic studies, however, suggest that moderate, heavy, and alcoholic drinkers are likely to be members of different populations whose patterns of use are relatively stable, not members of the same population observed at different points in time (for example, Norm and Wanberg, l969). Distortion of responses by the motivation to appear socially acceptable is often a methodological problem in social science. The problem may be particularly severe when the research topic involves illicit or normatively disapproved substances. Substance use investi- gators should be alert to shifts in social approval or disapproval of specific substances over time. Cigarette smoking is a good example of an activity that was once positively or neutrally valued but now is increasingly being judged as negative. The existence of current data on attitudes toward substances is the best defense against undetected changes in response bias due to social acceptability factors. There are a variety of interesting ways in which survey informa- tion can be coupled with other kinds of investigation. Identification of hypothetical control factors can be the basis for organized inter- vention programs using behavioral training principles; these methods lend themselves well to field studies in which control factor hypotheses are tested and the effectiveness of the interventions are assessed simultaneously. The current movements for behavioral medicine and prevention provide a favorable ideological climate. Participant observer studies in tandem with a survey (on the same population) could considerably amplify information on the operation of controls. Conclusions We need more studies on control factors in substance use in populations not in formal treatment programs. We particularly need studies that emphasize cross-substance comparisons and relationships. We urge that attempts to reduce substance use and their successes and failures be studied extensively in the general population rather than

60 concentrating solely on the narrow domain of treatment studies with excessive users. The steps that need to be taken to develop and carry out specific policies responding to the needs we have identified in large part involve technical matters for policy scholars and government special- ists. Therefore, our recommendations for action are in the most general form, with the understanding that we are offering compass readings, not road maps. l. In funding agencies concerned with substance abuse, there should be more support for studies of control factors in substance use, emphasizing populations not in formal treatment programs. 2. Studies that include use patterns in individuals across many substances are both important and scarce, since most funding for surveys and examinations comes from categorical substance abuse agencies. A mechanism is needed to coordinate some of these funding programs so that cooperative, cross-substance studies will receive the needed level of support. 3. It is both feasible and cost-effective to use existing data sets containing substance use information in secondary analyses. Moreover, relevant existing samples ought to be studied again to supplement the original data and/or to collect prospective data. 4. Cross-agency cooperation is needed for data exchange, so that profiles of control factors in substance use can be constructed across substances or so that data sets collected under the sponsorship of one agency may be available to the grantees or contractors of all agencies supporting work in substance use. 5. Precedents and agreements are needed to make the products of very expensive government-supported studies available to a wide group of researchers. Such agreements should protect the legitimate professional interests of those who have carried the burden of study design and data collection, while changing the prevalent pattern of perpetual ownership. BIBLIOGRAPHY Abelson, H. I., P. M. Fishburne, and I. H. Cisin (l977) National Survey on Drug Abuse: l977. A Nationwide Study - Youth, Young Adults, and Older People: Vol. l, Main Findings. Report prepared for the National Institute on Drug Abuse, DHEW Publication No. (ADM) 78-6l8. Washington, DC: U.S. Government Printing Office. Armor, D. J., P. Johnson, S. Polich, and H. Stambul (l977) Trends in U.S. Adult Drinking Practices: Summary Report. "Working Note" prepared for the National Institute on Alcohol Abuse and Alcoholism. Santa Monica, CA: Rand Corporation. Armor, D., M. Polich, and H. Stambul (l976) Alcoholism and Treatment. R-l739-NIAAA. Santa Monica, CA: Rand Corporation.

6l Cahalan, D., I. H. Cisin, and H. M. Crossley (l969) American Drinking Practices: A National Study of Drinking Behavior and Attitudes. New Brunswick, NJ: Rutgers Center of Alcohol Studies: Dawber, T. R. et al. (l95l) Epidemiological approaches to heart disease; the Framingham study. American Journal of Public Health 4l:279-286. Drug Abuse Epidemiology Data Center (l976) Information Package. Fort Worth, TX: Institute of Behavioral Research, Texas Christian University. Frederickson, D. T., A. McAlister, and B. G. Danaher (l976) Giving up smoking: how the various programs work. Medical World News l7(24):52-57. Greene, M. H. and R. L. Dupont (l974) Heroin addiction trends. American Journal of Psychiatry l3l:545-550. Hammond, E. C. and D. Horn (l958) Smoking and death rates - report on forty-four months of follow-up of l87,783 men. Journal of the American Medical Association l66(ll):l294-l308. Horn, D. (l97l) Epidemiology and psychology of cigarette smoking. Chest 59(5, supplement):225-245. Horn, D. (l968) Some factors in smoking and its cessation. Pp. l2-2l In E. F. Borgatta and R. R. Evans, (eds.) Smoking, Health, and Behavior. Chicago: Aldine. Horn, J. L. and K. W. Wanberg (l969) Symptom patterns related to excessive use of alcohol. Quarterly Journal of Studies on Alcohol 30:35-58. Hughes, P. H. (l977) Behind the Wall of Respect. Chicago: University of Chicago Press. Jaffe, J. H. (l975) Drug addiction and drug abuse. Pp. 284-324 In L. S. Goodman and G. Gilman, (eds.) The Pharmacological Basis of Therapeutics. New York: Macmillan. Jarvik, M. E. et al., eds. (l977) Research on Smoking Behavior. National Institute on Drug Abuse Research Monograph No. l7. U.S. Government Printing Office. Jellinek, E. M. (l952) Phases of alcohol addiction. Quarterly Journal of Studies on Alcohol l3:673-684. Jellinek, E. M. (l960) The Disease Concept of Alcoholism. New Haven, CT: College and University Press.

62 Johnston, L. (l973) Drugs and American Youth. Report to the U.S. Office of Education. Ann Arbor, MI: Institute for Social Research. Josephson, E., and E. Carroll, (eds.) (l974) Drug Use: Epidemiological and Sociological Approaches. Washington, DC: Hemisphere - Halsted - J. Wiley & Sons. Kandel, D. (l973) Adolescent marijuana use; role of parents and peers. Science l8l:l067-l070. Kandel, D., (ed.) (l978) Longitudinal Research on Drug Use: Empirical Findings and Methodological Issues. Washington, DC: Hemisphere - Halsted - J. Wiley & Sons. Knupfer, G. and R. Room (l967) Drinking patterns and attitudes of Irish, Jewish, and White Protestant American men. Quarterly Journal of Studies on Alcohol 28(4):676-699. Meyer, R. E. and S. M. Mirin (l979) The Heroin Stimulus: Implications for a Theory of Addiction. New York: Plenum. National Cancer Institute (l977) The Smoking Digest: progress report on a nation kicking the habit. Washington, DC: U.S. Department of Health, Education, and Welfare. O'Donnell, J. A. (l969) Narcotic Addiction in Kentucky. Public Health Services Publication No. l88l. Washington, DC: U.S. Government Printing Office. Parry, H. J. et al. (l974) Increasing alcohol intake as a coping mechanism for psychic distress. Pp. ll9-l44 In R. Cooperstock, (ed.) Social Aspects of the Medical Use of Psychotropic Drugs. Toronto, Ont.: Addiction Research Foundation. Preble, E. and J. J. Casey (l972) Taking care of business: the heroin user's life on the street. In D. E. Smith and G. R. Gay, (eds.) It's So Good, Don't Even Try It Once: Heroin in Perspective. Englewood Cliffs, NJ: Prentice-Hall. " Rittenhouse, J., ed. (l977) The Epidemiology of Heroin and Other Narcotics. National Institute on Drug Abuse Monograph No. l6. U.S. Public Health Service. Washington, DC: U.S. Government Printing Office. Robins, L. N. (l976) Estimating addiction rates and locating target populations: how decomposition into stages helps. A Report of the Task Force on the Epidemiology of Heroin and Other Narcotics. Rockville, MD: National Institute on Drug Abuse.

63 Robins, L. N. (l974a) The interaction of setting and predisposition in explaining novel behavior: drug initiation before, in, during, and after Vietnam. Pp. l79-l96 in D. Kandel, ed., Longitudinal Research on Drug Use: Empirical Findings and Methodological Issues. Washington, DC: Hemisphere - Halsted - J. Wiley & Sons. Robins, L. N. (l974b) The Vietnam drug user returns: Final Report, Special Action Office Monograph, Series A, No. 2. Washington, DC: U.S. Government Printing Office. Robins. L. N. and Murphy, G. E. (l967) Drug use in a normal population of young negro men. American Journal of Public Health 57(9): l580-l596. Roizen, R., Cahalan, D., and Shanks, P. (l978) "Spontaneous remission" among untreated problem drinkers. Pp. l97-22l in D. Kandel, ed., Longitudinal Research in Drug Use: Empirical Findings and Methodological Issues. Washington, DC: Hemisphere - Halsted - J. Wiley & Sons. Rothman, K. J. (l976) Causes. American Journal of Epidemiology l04(6):587-592. Schachter, S. (l978) Pharmacological and psychological determinants of cigarette smoking. Annals of Internal Medicine 88(l):l04-ll4. Sells, S. B. et al. (l976) A national follow-up study to evaluate the effectiveness of drug abuse treatment: a report on Cohort l of the Drug Abuse Reporting Program five years later. American Journal of Drug and Alcohol Abuse 3(4):545-556. Smart, R. (l974) Addiction, dependency, abuse, or use: which are we studying with epidemiology? Pp. 23-4l In E. Josephson and E. Carroll, (eds.) Drug Use: Epidemiological and Sociological Approaches. Washington, DC: Hemisphere - Halsted - J. Wiley & Sons. Smith, G. and C. P. Fogg (l978) Longitudinal study of teenage drug use. In D. Kandel (ed.) Longitudinal Research on Drug Use: Empirical Findings and Methodological Issues. Washington, DC: Hemisphere - Halsted - J. Wiley & Sons. Sterling-Smith, R. S. (l976) A special study of drivers most responsi- ble in fatal accidents. Summary for Management Report, Contract No. DOT HS 3l0-3-595, April, l976. In Seventh Annual Report on Marijuana and Health, l979. National Institute on Drug Abuse (l979). Washington, DC: U.S. Government Printing Office.

64 Stunkard, A. J. (l975) Obesity and the social environment. Pp. l78-l9l In A. Howard, (ed.) Recent Advances in Obesity Research: l. Proceedings of the First International Conference on Obesity. London: Newman. U.S. Public Health Service (l976) The Health and Nutrition Examination Survey. DHEW Publication No. (HRA) 76-l235. Washington, DC: U.S. Department of Health, Education, and Welfare. Vaillant, G. (l973) A 20-year follow-up of New York City heroin addicts. Archives of General Psychiatry 29:237-24l. Wiseman, J. P. (l970) Stations of the Lost. Englewood Cliffs, NJ: Prentice Hall. Zinberg, N., W. M. Harding, and M. Winkeller (l977) A study of social regulatory mechanisms in controlled ilicit drug users. Journal of Drug Issues 7:ll7-l33.

COMMENTS ON "TOWARD THE ACQUISITION OF DATA ON CONTROLLED SUBSTANCE USE" Byron W. Brown, Jr. I would like to compliment Drs. Stunkard and Levison on a very stimulating paper; I appreciate the chance to comment on it. In the past, biostatisticians have been concerned for the most part with the design and analysis of animal experiments, clinical trials, and classical epidemiological studies of a retrospective or prospective nature. In recent years, however, there has been a new movement in medical research toward population research in preventive medicine and health care, with an emphasis on the study of health care systems and the development and the scientific evaluation of public health programs. These interests are readily apparent in the paper by Stunkard and Levison, who are concerned with the broad study of sub- stance use and abuse, the measurement of the extent of the problem in the population, and the natural history of the problem in individuals in the general population. Such studies would provide the basis for the three steps toward preventive medicine and protection of the public health in this area—development of preventive measures for the in- dividual, development of public health programs that incorporate these measures, and evaluation of the costs and benefits of such programs. Throughout the Stunkard-Levison paper there is a recurrent theme, the implication or assumption that there is a broad population of substance users, who are potential abusers. Stunkard and Levison propose to gauge the size of this population and to sort these users into those who are and will remain healthy and those who do or will need help. There is a heavy emphasis on the idea that multiple substance use may be a key to this classification. As a member of the Department of Family, Community and Preventive Medicine at Stanford, I am attracted to the ideas of the paper as a basis for preventive medicine. As a biostatistician, I am concerned NOTE: The comments in this discussion in part respond to the original version of the paper, "Toward the Acquisition of Data on Controlled Substance Use," which was presented at the conference. The paper in this volume has been modified considerably from the original, to some extent on the basis of comments in this discussion. 65

66 with the technological problems involved. Stunkard and Levison give us some very sound guidance on these difficulties. I am much interested in their ideas since I have gained some related experience of my own in a mind-boggling experiment in preventive medicine that we are carrying out at Stanford. A very large, one-campus team of investigators— psychiatrists, epidemiologists, psychologists, communications experts, cardiologists, biostatisticians, biochemists, and others—are putting together all that is known about the causes of heart disease and methods of behavior modification in an effort to persuade and enable whole communities to modify their living habits, to reduce their risks of heart disease. Our experimental units are cities; our treatments involve mass media and community organization; our end points are community changes in habits, reduction in risk factors, and decreases in morbidity and mortality rates. Our last study involved two treated communities and one control community. Our current study involves five cities, two treated and three control cities, containing a total of nearly .5 million people. Our goal in heart disease research is going to be difficult to achieve, but I think that the research problems of substance use are more difficult. Let me review some of the diffi- culties mentioned by Stunkard and Levision. l. We know far more about the causes of heart disease than is known about the risk factors for substance use and abuse in general, though, of course, the two areas overlap. In heart disease we have an extensive background of retrospective studies and a number of very good cross-sectional and longitudinal studies, Framingham being the best known of the latter. Similar studies are lacking in substance use and abuse and only now are being called for by Stunkard and Levison. I can assure you that heart disease studies such as the Framingham and Tecumseh studies, in which whole populations of people have been followed carefully over many years to discover who ultimately develops disease, are invaluable for further research in two respects. First, they furnish the basis for deciding which are the important risk factors that might be amenable to change through health education and other preventive medicine programs. Second, they furnish the data on population variation in risk factors and on incidence rates of desease that allow for sound estimates of sample size requirements for further research—in terms of the numbers of cities required for study, sizes of cities, sizes of surveys within the cities, and so forth. Obviously, this same kind of information on risk factors for the general population, with incidence rates, must be available for substance abuse if any research in preventive medicine is to be done. 2. Another point that makes substance abuse itself a more difficult area for research than, for example, heart disease is the sensi- tivity of the subject and the involvement with issues of morality and legality. We know, in dealing with smokers and with the

67 overweight, that guilt and sensitivity cause the unsuccessfully treated to lie about their activities and to drop out of studies, making analysis and interpretation of research results difficult. The problems of studying behavior that is regarded by the community as immoral or illegal is many times more difficult. There are tricks for dealing with these problems, but they are expensive and do not provide a complete answer to the questions. For example, blood levels and/or urine levels can be monitored for evidence of smoking or heroin use. In administering ques- tionnaires, randomized response methods can be used so that the answers to the questions are given but not divulged even to the interviewer and are available only for statistical tabulation by solving a set of equations involving the final tabulations of answers to a set of questions. And, of course, there are the usual statistical techniques for testing for dropout biases and for adjusting for such biases, techniques that become critical in studies in which the dropout rates, from recruitment to termina- tion of the study, are substantial. 3. A third difficulty in carrying out research in the area of sub- stance use and abuse stems from the low incidence rates. A large proportion of the United States population is at astonishingly high risk of developing clinically manifest heart disease. It isn't difficult to specify fairly large subgroups in the popula- tion who have an incidence rate that is as much as 5 to l0 percent per year. However, although substance use and abuse is widespread in this country, particularly if one includes the use of cigarettes, alcohol, and food, the incidence of actual measur- able trouble stemming from use may be another matter. Though the impact of use on individuals may be huge and such problems for the community may be of public health significance, if the incidence rates for these cases are low, the sample sizes required for reliable preventive medicine research may be prohibitive. Because of the difficulties mentioned above, I would suggest that Stunkard and Levison are right on the mark when they propose a no-holds- barred approach to the study of the natural history of substance use and abuse in the United States population. They suggest that existing data banks be mined, that population samples already selected and studied by others be followed up, and that studies be piggybacked by inserting items in questionnaires of surveys being planned by govern- ment agencies and other organizations for other purposes. I would favor exploration of all three of these avenues, though I am not sanguine about results. Existing data banks seem to be very useful to economists, but in health statistics I have been disappointed to find that data in an existing data bank usually fall too far short in quality or detail to answer any but the main objective precipitating the study. As to following up on a sample of persons selected by some other investigator for another purpose, I feel that this will become

68 more and more difficult as we learn to respect the privacy of our sample subjects by making identification and trace of the subjects impossible as soon as a study is finished. As to piggybacking by inserting items in another survey, I think this has some potential, but only if some serious preliminary work is done. I would suggest that a committee of investigators interested in substance abuse set about developing questions or hypotheses. Such items could then prepare and pretest items relevant to these hypotheses. Such items could then be proposed, adopted, and readily incorporated into the questionnaire of a cooperating agency or investigator. If the investigator has to develop and pretest the item himself, he is not likely to bother. As to the choice of strategy for a primary epidemiological study of the causes of substance abuse, I disagree with Stunkard and Levison when they say that retrospective study is the method of choice. The retrospective approach will usually be the least expensive, but problems in the choice of control groups and questions of response and investigator bias are always much more acute in retrospective studies, where the end point is the start of the study. The prospective study, in which subject are selected and then followed to the occurrence of end point, has the advantage in both respects; in addition, the pro- spective study has the upper hand in offering the chance to discover end points unsuspected at the start of the study. An example of the latter is the first prospective study of smoking as a cause of lung disease; a public health finding of much greater significance than the lung cancer association was the discovery that death due to heart disease was more prevalent among smokers than among nonsmokers. I believe that substance use research in the general population will be so expensive and so difficult that all strategies should be carefully weighed, even piloted, before deciding on an approach in any given instance. Stunkard and Levison have given us the beginnings of a taxonomy for study designs in use research in populations. I believe this to be well worth additional effort. A fairly detailed catalog of approaches, with examples and costs specific to substance abuse re- search, could be invaluable to both investigators and those who are asked to support studies in this area. Now suppose that the Stunkard-Levison plea for broad scientific study of substance use in the population is met. Suppose risk factors for serious abuse are uncovered and fairly well understood, risk factors that might be altered through health education or other control efforts at feasible cost to the public. Is this the point at which scientific inquiry stops and public policy makers take over? I hope not. I feel that this has been the case in other areas too often in the past. It worries me that Stunkard and Levison seem to weaken at several points in their paper when they speak of the time pressures exerted when the issues involve public policy. I believe it is just at this point of pressure that scientific inquiry becomes most important, for here the losses through wrong decision can be greatest. There is a great leap from a conviction that a factor causes abuse to the conclusion that a particular effort at control, through education, legislation, or some other means will have a substantial and cost- effective impact on the problem. Public money invested in verifying

69 the validity of this leap, through sound scientific study may be appropriate if the stakes are high. But valid study of this kind of question can be extremely expensive. This is where the whole new field of program evaluation, methods of quasi-experimentation, and so on enter—a topic for another time. Let me conclude by saying that I enjoyed the Stunkard and Levison paper very much. I found the recommendations sensible and I do support them all.

BETTER THEORIES AND METHODOLOGIES—THEN MORE DATA Bruce D. Johnson The Stunkard and Levison paper, "Toward the Acquisition of Data on Substance Abuse," reviews many aspects of controlled or excessive use of substances or habitual behaviors. They correctly note that much of the current literature is substance-specific, or deals with general drug classes (i.e., nonmedical use of legal substances and illegal drug use versus medical drug use, alcohol, cigarettes, beverages such as coffee, tea, cola, and over-the-counter drugs). Their observation that current research is in an empirical phase and not heavily influenced by major theories accords with my obserations. Moreover, the committee's insistence on studying controlled substance use as well as excessive use is needed. They raise many tantalizing questions for further re- search. Their identification of longitudinal data sets, such as HANES and HIS, collected by the National Center for Health Statistics was previously unknown to me. Nevertheless, I have an uncomfortable feeling that the paper raises many intriguing questions that need research, but the authors fail to develop or suggest a framework for analyzing theories and methodologies for studying both controlled and excessive substance use, however defined. My unease with the paper lies less with what they present than with what they neglect. Although they touch on three of the areas I will mention, a fuller development of major ideas with recommendations that flow from such an analysis is needed. My field of expertise lies mainly in illegal drug-using behavior, particularly studies of drug use in the general population plus the behavior of opiate addicts. After about a decade of experience and relatively wide reading, I am aware of four major problems that appear to currently inhibit major advances in substance use research. NOTE: The comments in this discussion in part respond to the original version of the paper, "Toward the Acquisition of Data on Controlled Substance Use," which was presented at the conference. The paper in this volume has been modified considerably from the original, to some extent on the basis of comments in this discussion. 70

7l l. To the extent that effective theory exists in this field, it is mainly post hoc. Far too little research is conducted to assess major theoretical model(s); even less theoretical work is eventually published. 2. Government-sponsored and funded research, mainly through academic institutions and/or government agencies, appears unaware of a major source: many experts are almost entirely unaware of private market research on substance use, which, I suspect, may have preliminary answers to some questions that Stunkard and Levison raise. 3. Surveys relying on respondent self-reported behavior must over- come major methodological obstacles that need to be addressed by a constructive program of intensive research in the near future. 4. The results of major studies seldom have any apparent impact upon the American mass media and populace, although the impact on policy makers may be more substantial at times. Despite the best research to the contrary, popular folk beliefs persist unchanged. Stunkard and Levison state "so little is well known about sub- stance use patterns in the general population that important informa- tion could be obtained from ... fairly obvious questions." They then list l0 questions. This assertion strikes me as lacking a factual basis. Given three months of full-time effort, a scholar who knows the relevant literature could provide a mass of data from the published literature relevant to these questions. The answers might not be clear because much of the data might contain contradictory findings. Further, even better answers could be obtained from relevant data sources that are currently available and accessible for secondary analysis. My office recently completed data collection on a survey of drug use representative of public and Catholic junior and senior high school students in New York state. To aid in comparison with other studies, the Drug Abuse Epidemiology Data Center (DAEDEC) was asked to provide us with a list of those studies they have on file. We received an index more than l0 pages long and a computer output of about 2,l00 pages listing the data elements in those studies of high school popula- tion available at DAEDEC. Mountains of data, good data, have barely been presented, much less analyzed for theoretical implications. For example, the Vietnam veterans data, while masterfully analyzed by Lee Robins (l973, l974, l975, l977), could be mined for many more important papers and even books. This data set is available at DAEDEC. Response Analysis (l973, l975, l976, l977) has conducted biannual surveys of the United States population and Lloyd Johnston (l977) conducts annual surveys of high school seniors that have only been described but not carefully analyzed. Another example, John O'Donnell and associates' (l976) study of young men contains, to the best of my knowledge, the only data on self- reported criminal behaviors among a relatively normal population of adults. These fine studies beg for more intensive analysis. Thus

72 Stunkard and Levison's recommendation to limit investigator "ownership," while reasonable, avoid the major problem—few persons conduct second- ary or even intensive analyses of available data. More Theorizing Needed To the extent that any theory is employed, research on drugs and habitual behaviors uses what Robert Merton (l957) calls theories of the middle range. What passes for theory consists of "general orientations toward data, suggesting types of variables which need somehow to be taken into account, rather than clear, verifiable statements of the relationships between specific variables." Research in illicit drug use has generated several relatively good general orientations toward types of variables that have been compiled into a relatively integrated theory; the theoretical and empirical work of the Jessors (l977), Kandel and associates (l975, l976), Ginsberg and Greenley (l978), and Johnson (l973) have been particularly fruitful. A quick appraisal of the published literature, however, should quickly convince almost any scientist that most substance abuse re- search is relatively atheoretical (see Johnson and Nishi, l976, for more documentation). Most of the published work is clearly descriptive; data relevant to a series of concrete questions are presented. No at- tempt, possibly excepting a few clearly post hoc statements, will be made to show that the data offer support or refute some theoretical idea. My rough estimate is that for every article or monograph that attempts to assess one or more theoretical propositions, there are approximately l0 atheoretical, descriptive articles or monographs. I suspect your ratios may differ from mine, but only in magnitude and not direction. Why is there so little middle-range theorizing in research on substance use and, I suggest, in other fields of habitual behavior? I suggest three major factors: the sociology of grantsmanship; the lack of extended collegia! relationships; and the difficulties of bringing adequate theories to one or more data sets. The Sociology of Grantsmanship Although a variety of parties may be at fault (i.e., investigators overpromise and procrastinate, research organizations are underfunded, etc.), the federal government—which funds much research on substance abuse—must bear the major responsibility for the slow advances in theory development. Why? Because most of the funds for research are almost always committed, via grants and contracts, to instrument devel- opment, data collection, and extremely condensed report writing. So little time and resources are committed to report writing that only a bare description of the massive data can be generated. Then the grant funds run out. At this point, three major routes of activity are open to the investigator(s) with mountains of important data conceived to measure some relatively clear middle-range theoretical variables: l)

73 apply for a grant extension to conduct continued analysis of the data; 2) apply for a new research grant that will initiate a new round of instrument development, data collection, and rapid report writing; or 3) return to other channels of support (teaching, administration, new job, etc.) and attempt to mine the data by writing reports on theoret- ical topics as time permits—which it seldom does. For reasons that I do not fully understand, the third option seems to be the most common, followed by the second. The first option, of obtaining continued support for further theoretical analyses, is seldom awarded. Such continued funding is essential if major theoret- ical advances are to be made. I suggest that the physical sciences do better at ensuring the theoretical and empirical advances by patterns of continuous funding support to major laboratories. Lack of Extended Collegial Relationships Most of the researchers who will make major theoretical and methodological advances are based in universities and occasionally in government. Although drug researchers may have departmental and pro- fessional colleagues who are sociologists, psychologists, etc., they seldom have contact at their university with other drug experts. They know of others' work mainly through published articles. Occasionally they meet and discuss current work at professional conferences. A few are invited to serve on federal research review committees or as gov- ernment consultants. But it is a rare occasion when colleagues spend even one week working together on mutually useful exchanges of ideas, critiques of working papers, clarifying theories and comparing research findings. Yet I would suggest that routine collegial exchanges could greatly enhance the advancement of theory and research in drug re- search, a field with much disciplinary fragmentation. Again, I would single out the federalgovernment for failure to promote serious working collegial exchanges; current efforts via professional conferences, while promoting this effort, could be more effectively focused to generate results. Difficulties in Theorizing The major problem, however, in future theoretical advances is conceptual and analytical. Everyone who has tried to develop a theoretical structure and then prove or disprove it with data knows that this is an extremely diffiuclt task. Major rewrites of the theory are frequently in order, which may change the entire approach to the data. Further, the data do not always conform to the initial theory so revisions in the theory need to be made. In short, a great deal more time must be poured into writing a theory that fits the data fairly well. And once it is worked out, reactions from colleagues may bring a whole new wealth of ideas to the theory and data. If the theory and/or data are not subject to neglect, reactions from colleagues, however, may open up new challenges. Every colleague has his own theory to

74 propound or to ask you to test. But it is precisely at this juncture between colleagues that major advances are apt to be made. Why? Because the principal author must sharpen his definitions, develop new distinctions, or revamp them entirely. New interpretations may lead to a new theory. Empirical evidence may suggest that theories for ex- plaining one phenomenon (say marijuana use) may be less important in explaining a similar phenomenon (say barbiturate use). The Examination of Market Research Data Stunkard and Levison rightly emphasize the need for more second- ary analyses of existing data and locate several important data sets for such efforts. They seem to badly neglect a major data source, the private market research industry. Since almost all market research is proprietary and owned by private industry, published reports on current data seldom, if ever, appear in major academic journals. Academic and government officials know very little about the scope and activity of market research on alcohol, tobacco, over-the-counter drugs, beverages, and food. I, for one, admit to being quite ignorant. I have, however, interviewed at several market research companies, listened to a market advertising radio program in New York, and have talked with an aca- demic, Charles Winick, who has experience in and contacts with the market research industry. My guess is that the market research industry must have prelim- inary answers to several of the questions that Stunkard and Levison raise, although there is probably even more substance specialization than in government surveys. I particularly suspect that liquor and tobacco market researchers know more about shifts in consumer behavior and how this is influenced by supply and demand or advertising and gov- ernment policy than does the federal government. Some of the larger market research firms or industries conduct continuing surveys of consumer behavior on a monthly or annual basis; they have panels of respondents, sometimes extremely large (20,000) and carefully selected. Many respondents may exhibit the patterns of con- trolled use or excessive use in which this committee is interested. In market research, any study that is more than one year old is dated and almost useless; most companies, I suspect, would be willing to release such "old" data for secondary analysis. Further, for a few thousand dollars, the government could easily buy into prospective longitudinal consumer surveys that may be extremely economical on a per interview basis when compared with the current grant process. I would also suspect that market research has also dealt with and developed preliminary methodologies for overcoming some of the major shortcomings of respondent self-reported behaviors and sample selection problems to which I will now turn.

75 Toward Better Methodologies for Studying Substance Consuming Behavior The suggestions and recommendations made by Stunkard and Levison regarding methodological issues are generally quite practical and po- tentially useful. Even here, however, their suggestions seem to deal with minor issues (an examination of the trees), rather than outlining the major issues (analyzing the forest) that need research support in the immediate future. I will attempt to suggest what I see as the major issues that need to be addressed. Quantitative survey research techniques and qualitative studies (variously referred to as ethnography, field studies, observations, or anthropology) have provided important theoretical and methodological insights about substance use. Unfortunately, the two methodologies are seldom combined and relatively separate literates exist in both areas. Large samples and statistical analyses (Kandel, l975, l976; Johnston, l977; Jessor, l977, Response Analysis, l977) can provide generally more convincing proof that a relationship exists between one or more vari- ables but field studies (see Preble and Casey, l969; Waldorf, l977; Weppner, l977; du Toit, l977) provide fuller description of the pro- cesses, motivations, and patterns of behavior that emerge during the course of any study. Future research needs to employ statistical measures of major factors, but also needs to obtain observational and motivational insights to provide a fuller understanding of substance abuse and habitual behavior. At the current time, national and other surveys can provide us with highly accurate cross-sectional and trend data about the preva- lence and/or incidence of drug use patterns (Response Analysis, l972, l974, l976, l978; Cahalan et al., l974; Johnston, l977). In addition to improved theory development, I predict that the next major advances in substance use research—which have already begun (McGlothlin, l977; Nurco, l976)—will occur when the phenomenon of interest becomes the unit of analysis; that is, when we can accurately count the number of marijuana joints, cups of coffee, "bags" or amount of pure heroin, criminal acts committed, criminal victimizations experienced, etc. Most of the experimental research on animals and humans has shown strong does response curves. The Maloff et al. paper on informal processes (in this volume) also indicates that sociocultural patterns may decisively influence the does response patterns demonstrated in laboratories. Major theoretical and empirical advances will emerge when social scientists using survey-like techniques can measure with a precision roughly similar to the laboratory how much of a substance a person or class of persons consumes over the short and long run. The central finding of research on illicit drug use (Kandel, l975, l976; Johnson, l973; Jessor, l977) is that a person's immediate clique of friends (also called peer group, reference group, best friends, etc.) has the greatest impact on the use and regular use of a substance by the respondent. Reference groups and friendship cliques probably change over time, but our ability to measure and characterize these cliques is woefully underdeveloped, especially on a longitudinal basis. Only careful thinking, clear theorizing, and careful testing of

76 alternative metholologies may provide methods to measure such changes in clique behavior. Survey research, while relatively accurate in measuring any use versus non-use, is much less accurate in assessing the quantities con- sumed during different time periods and in different settings. The problems of measuring or counting the phenomenon instead of the person is fundamental, of course, to distinguishing between controlled sub- stance use and excessive use; where and how much substance can be con- sumed before it is controlled versus execessive is a central question raised by the papers in this volume by Stunkard and Levison, Maloff et al., and Fonaroff et al. Stunkard and Levison note, and much evidence suggests, that cur- rent survey research techniques cannot accurately measure quantities because of strong social norms toward moderation. That is, respondents both forget (memory fade) what they did and report what they think they should have used or done, rather than report on their actual behavior. Thus, respondent reports may frequently error on the side of under- reporting (Perry, l97l; National Research Council, l976). Such under- reporting cannot be precisely measured and undoubtedly varies from person to person. Thus, a central goal for the immediate future should be development of a methodology to measure the objective reality of what a person consumes independently of how much that same person reports himself as consuming. National surveys of alcohol use find, when projected to the national population, that total alcohol consumption is 40-50 percent lower when estimated from survey data than from the beverage industry sales of alcohol (Room, l975); similar comparisons of survey-reported consumption and industry sales could be made for cigarettes, coffee, tea, over-the-counter drugs, and prescription drugs. Of course, not all substances manufactured and sold by industry are actually consumed; nevertheless, a variety of checks on respondent behavior is needed. Low-income respondents are also quite mistrustful of interviewers, especially persons exhibiting patterns of what society may consider excessive consumption. Edward Preble has studied opiate addicts in the ghetto of East Harlem. His experience is that the first interview should be mainly devoted to establishing rapport and trust with such respondents. While not denying heroin use, respondents tend to severely minimize their consumption in the first interview. For example, I.F. is currently enrolled in a methadone program, although he has been off and on several times. When asked about his use of heroin during the first interview, he indicated that he only "dipped and dabbed" in heroin. By the third session, when he felt more confortable with the interviewer, the following conversation occurred: Q. But you just can't make it on the methadone alone? A. You get a craving. I always said that the majority of people that still go for heroin got like a desire. They like the spike. They like to see that blood come up or whatever it is. They got a needle craze. They like to stick that needle in their arm.

77 Q. Were you taking dope when you were working and on a methadone program? A. I was always messing around. The majority of the times guys do it out of boredom. ... You're laying around all day and you have to get into something and start, you know, running for that bag. Probably the most extensive efforts to measure the phenomenon of substance consumption has been done for liquor, cigarettes, and other commercial products. Here I suspect that the market researchers have methodologies (Winick, l977) that may be more advanced than academic or government research. Efforts to improve criminological research are under way; the National Research Council (l976) has recently analyzed and recommended major methodological research in the National Crime Panel Surveys to measure more accurately criminal victimizations in the United States. The paper by Maloff et al. on informal social controls raises yet another methodological problem to be investigated: How can social scientists measure and quantify "informal social controls" or the various norms controlling substance abuse behavior? That is, what are the cultural recipes? Given that these can be described statistically with a known range of variation (i.e., average age at initiation of substance use, average dose per occasion), perhaps those who initiate use at an earlier age or who use twice as much as average may be des- cribed as having tendencies toward heavy or excessive use. The main difficulty is how to describe ethnographically and statistically what these norms are for the general population, for those who ever use a given substance, and for "consumers" (those who use on a weekly or daily basis or some other criteria of regular use). Thus, a methodology for measuring the conduct or operational norms (i.e., norms that ef- fectively govern group behavior) as opposed to moral norms (i.e., norms expressing what the group wishes would or would not occur) is badly needed and would probably provide major advances in both theory and methodology. The suggestions that Stunkard and Levison make about research designs for future longitudinal studies is well informed and accurate. Despite all the problems they identify with prospective longitudinal designs, however, I suspect that future longitudinal research will almost have to adopt such designs—even with treatment or identified excessive user populations. Federal legislation protecting respondent privacy and requiring informed consent now makes it very difficult, if not impossible, to gain access to standard agency files necessary for tracking potential respondents and validating respondent self-reported behavior. Without written permission from respondents, access to such data is almost impossible and very difficult with such permission. Thus, research designs in which respondents given written informed con- sent at the first interview for future interviews will almost necessi- tate prospective longitudinal designs. A related design issue they neglect is how to sample enough con- trolled and/or excessive users of substances that are used by small

78 proportions of the population. Random samples of the general popula- tion, regardless of how large, will generally not produce enough cases for extensive cross-sectional analysis, and even fewer cases for longi- tudinal analysis (assuming considerable shifts in respondent substance use). Current national studies collect too much data from nondrug users or experimental-light users (possibly excepting alcohol and tobacco). Future national studies need to be carefully stratified to obtain data from respondents in groups identified from previous surveys as being at highest risk for controlled or excessive substance use. Many other methodological improvements may be needed before initiating another round of major nationwide surveys, but the above discussion emphasizes some of the more important advancements that need to be made. Reaching the American Populace America is a democracy in which current institutional arrange- ments both legislate morality and protect bad habits. One result of such democratic action is that widespread legal access to substances that negatively effect health and the nation's economy (e.g., alcohol and tobacco) coexist with severe laws against substances whose harmful effects to health and economy are hard to document (e.g., LSD). During the past decade, major research efforts have informed academic social scientists and some policy making elites about patterns of substance use and abuse. The Robins (l973, l974, l975, l977) studies of Vietnam veterans, among others, have been particularly effective in discredit- ing the "once an addict, always an addict" belief. But this news sel- dom reaches mass media attention; the public seems entirely unaware of these findings. The important point is that findings from the research effort envisioned in the Stunkard and Levison paper will, in all probability, be ignored by the media and public as widely as the Robins or O'Donnell study. The current gap between researched reality and misinformed be- liefs will remain. Perhaps bridging this gap and attempting to educate the public is as important as more research. A major hindrance to educating the public, I am convinced, is an excessive reliance on statistical data—that the average layman does not understand or relate to. In addition, statistical data remove the phenomenon (such as alcohol, heroin, or marijuana use) to the intellec- tual level, whereas most citizens relate to feelings, strong emotion, and support for their own morality. Journalists usually emphasize strong feelings and identification, both positive and negative, with case histories to make a story. Perhaps future research efforts should include among their sta- tistics some exemplary case histories of subjects (real identities con- cealed of course) who seem representative of the main findings reported in the study. With such cases, statistical and academic findings can be exemplified by snapshots of real people with whom the layman can relate. More feelings and human warmth written into major reports may

79 make major studies easier for the mass media to report and for the public to maintain interest. Suggestions for the Future Like Stunkard and Levison, I am not sure what steps should be taken to make major advances in substance use research. The following suggestions are only my opinion. Moreover, they are very general sug- gestions on which I could elaborate if asked. l. Funding agencies concerned with substance use should provide more support to study target high-risk groups in the general popula- tion, not only those identified by official action as "abusers." 2. Federal funding agencies should place more emphasis on funding completed reports that prove or disprove theories of substance use rather than provide only descriptions of data. This will generally mean more money allocated to support researchers to write books and theoretical papers after the original grant ends. Perhaps a contract with a very clearly specified end product and planned deadlines would speed up current research productivity and theory advances. 3. Federal funding agencies need to plan more interinstitution and interdisciplinary working conferences between major researchers and ensure that the results of such efforts get published and distributed rapidly. 4. Major emphases in the near future should be directed at develop- ing reliable and valid measures and methodologies to measure the amount of substances consumed by respondents, describing and sta- tistically analyzing the stability and changes in social cliques and major reference groups and statistically measuring group norms. 5. A pressing issue is to identify the 5 to l0 major theoretical models or major sets of concepts that may explain much of the variation in substance use, controlled use, and excessive use for a variety of substances or habitual behaviors. 6. Using these theoretical models, various data sets currently existing should be reanalyzed to show whether support for each model (or set of variables) appears to be statistically related to use patterns. 7. Current ongoing efforts, especially prospective longitudinal sur- veys by government agencies and market research firms, should be used for preliminary tests of the methodological and theoretical issues outlined above while also providing a baseline of sub- stance use obtained at the same time.

80 The analyses I present reflect my focal interest in illicit drug use. I know much too little about related fields of habitual behavior such as alcohol, tobacco, over-the-counter drugs, coffee, tea, gambling, medical drug use, and others. Nevertheless, I suspect that the problems I identify above are similar in these fields as well. If substance use and habitual behavior research is to advance as markedly in the l980s as it has in the l970s, the recommendations advanced by the committee may play a major role. REFERENCES Cahalan, Donald, et al. (l974) American Drinking Practices. New Brunswick, N.J.: Rutgers Center for Alcohol Studies. du Toit, Brian (ed.) (l977) Drug, Rituals, and Altered States of Consciousness. Rotterdam: A. A. Balkena. Ginsberg, Irving J. and James Greenley (l978) Competing theories of marijuana use: a longitudinal study. Journal of Health & Social Behavior l9 (March): 22-34. Jeasor, Richard and Shirley Jessor (l977) Problem Behavior and Psychosocial Development: A Longitudinal Study of Youth. New York: Academic Press. Johnson, Bruce D. (l973) Marijuana Users and Drug Subcultures. New York: Wiley Interscience. Johnson, Bruce D. and Sutsuko Nishi (l976) Myths and realities of drug use by minorities. In Iliyama, Nishi, and Johnson (eds.) Drug Use and Abuse Among U.S. Minorities. New York: Praeger. Johnston, Lloyd, et al. (l977) Drug Use Among American High School Students l975-77. Rockville, MD: National Institute on Drug Abuse. Kandel, Denise and Richard Faust (l976) Sequences and stages in patterns of adolescent drug use. Archives of General Psychiatry 32 (July): 923-932. " Kandel, Denise, et al. (l976) The epidemiology of drug use among New York state high school students: distribution, trends, and changes in rates of use. American Journal of Public Health 66 (l): 43-53. McGlothlin, William, et al. (l977) An Evaluation of the California Civil Addict Program. Services Research Monograph Series. Rockville, MD: National Institute on Drug Abuse.

8l Merton, Robert (l957) Social Theory and Social Structure. New York: Free Press. National Research Council (l976) Surveying Crime, Report of the Panel for the Evaluation of Crime Surveys. Washington, DC: National Academy of Sciences. Nurco, David (l976) Crime and addiction: methodological approaches taken to correct for opportunity to commit crime. In Report of the National Panel on Drug Use and Criminal Behavior, Drugs and Crime, Washington, DC: National Technical Information Service. O'Donnell, John, et al. (l976) Young Men and Drugs. NIDA Monograph (Series No. 5). Washington, DC: U.S. Government Printing Office. Perry, Hugh, et al. (l977) Primary levels of underreporting psycho- tropic drug use. Public Opinion Quarterly 34 (Winter)i 589-592. Preble, Edward and John Casey (l969) Taking care of business. International Journal of Addictions 4 (l): l-24. Response Analysis (l973) Drug experience, attitudes and related behaviors among adolescents. In the National Commission on Marijuana and Drug Abuse, Drug Use in America, Appendix I. Washington, DC: U.S. Government Printing Office. Response Analysis (l975) Public Experience with Psychoactive Sub- stances. Princeton, NJ. Response Analysis (l976) Nonmedical Use of Psychoactive Substances. Princeton, NJ. Response Analysis (l977) National Survey on Drug Abuse: l977. Washington, DC: U.S. Government Printing Office. Robins, Lee (l973) A Follow-up of Vietnam Drug Users. Washington, DC: Special Action Office for Drug Abuse Prevention. Robins, Lee (l974) The Vietnam Drug User Returns. Washington, DC: Special Action Office for Drug Abuse Prevention. Robins, Lee (l975) Veteran's Drug Use Three Years After Vietnam. Washington, DClNational Institute on Drug Abuse (unpublished). Robins, Lee (l977) Estimating addiction rates and locating target population. In Rittenhouse, Joan (ed.). The Epidemiology of Heroin and Other Narcotics. Washington, DC, National Institute on Drug Abuse. Research Monograph l6. Room, Robin (l975) Personal communication.

82 Waldorf, Dan, et al. (l977) An Ethnography of Cocaine Users and Sellers. Washington, DC: Drug Abuse Council. Weppner, Robert (l977) The Ethnography of Drugs and Crime. Beverly Hills, CA: Sage Publications. Winick, Charles (l978) Personal communication.

ACQUISITION OF DATA ON CONTROLLED SUBSTANCE USE: GENERAL CONFERENCE DISCUSSION Milton Terris In his comments on the Stunkard and Levison paper, Johnson dis- agreed with the authors on the importance of acquiring more survey data on substance use in the general population. Johnson asserted that rich stores of survey data were already in hand; what is needed is more analysis, more theory, and complementary observational studies. Much of the general discussion included rather sharply diverging statements about whether data on substance use in the general popula- tion are very inadequate or in relatively good supply. Johnson emphasized the NIDA-sponsored national surveys covering a wide range of substances (e.g., Abelson, Cisin, and Fishburne, l977), the alcohol surveys (e.g., Calahan et al., l969), and the Vietnam veterans data. Krasnegor cited Lloyd Johnston's (l973, l975), and Lillian Backford's (l975) studies and the National Clearinghouse for Smoking and Health data on smoking (e.g., l973). Much of the disagreement appeared to be focused on differing interpretations of what Stunkard and Levison meant by patterns of use in the general population. Levison stated that the intended thrust in their paper was toward persistence or changes in use over time, in- cluding attempts by individuals to curtail amount of use and the success or failure of such attempts. He asserted that existing cross- sectional surveys have limited utility for these purposes. Dews emphasized that it was the original purpose of the working group to find better ways of measuring use in the general population than those that currently exist. Schachter introduced a parallel issue that also points toward the need for data on use patterns in the general population. Most existing data on patterns of substance use are based on self-selected popula- tions, i.e., specific subgroups such as addicts, those in treatment, and others who come to the attention of formal agencies and treatment programs. Lee Robins's studies were attempts to obtain natural histories in a population that was not self-selected. However, as Kaplan noted, even this group is not an unbiased sample since the soldiers in Vietnam were aware that their urine would be tested for the presence of heroin, and those individuals who constituted the sample chose to continue use. Schachter emphasized that because the data we have are based on highly selected groups of excessive users, we tend to 83

84 regard addiction as incurable. We need data to tell us what happens to the whole spectrum of individuals who use various substances. The need for longitudinal data obtained through prospective studies was raised by a number of speakers. Cross-sectional data provides little help in the attempt to determine the patterns over time of users who start or stop using substances or change their mode of use. O'Brien commented that he would like to learn about the natural history of addiction: the incidence of spontaneous remissions and controlled use in addicts and nonaddicts, and about those users who don't come in for treatment. How many treatment successes would have occurred anyway, without intervention? The answers to such questions are not provided by the multitude of existing cross-sectional surveys, he said. Zinberg commended the paper for its attention to the wide range of users who fall between abstinence and compulsive use, and for its description of substance use as a socially evolving process. However, he stated that a reliance on surveys and quantitative data to meet our information needs is inadequate. In order to learn about the quality of use, rather than the quantity, the investigator must ask when, where, with whom, and under what circumstances people use substances. Experiential studies are required to get this kind of information. De Rios reinforced this suggestion, stating that there was a need to wed survey and ethnographic techniques in finding out about substance use in the normative population. Perhaps the most significant argument for prospective studies of substance use in the general population was presented by Solomon: billions of dollars are being spent for treatment, but the value of treatments cannot be measured without knowledge of how many persons stop using substances without treatment. Data are urgently needed on these base rates if treatment services are to be effectively evaluated. In view of the massive funds being spent on treatment, the issue is not just academic but has major practical significance for public policy. The issue of data use was also discussed briefly. Some conferees argued that existing data are underused and that secondary analysis of these data sets are rare. Maccoby stated that people do make use of existing data; for example, the data collected at the University of Michigan's Institute for Social Research are widely available and used. Krasnegor commented that NIDA is also sensitive to the issue of data use and sees its research monograph series as a successful attempt to collect and circulate research findings. REFERENCES Abelson, H., P. Fishburne, and I. Cisin (l977) National Survey on Drug Abuse: l977. National Institute on Drug Abuse, DHEW Publication No. (ADM) 78-6l8. Washington, DC: U.S. Government Printing Office.

85 Backford, L. (l975) Student drug use surveys, San Mateo County, California, l968-75. San Mateo: Department of Public Health and Welfare. Cahalan, D., I. Cisi.ii, and H. Crossley (l969) American Drinking Practices. New Brunswick, NJ: Rutgers Center for Alcohol Studies. Johnston, L. (l973) Drugs and American Youth. Ann Arbor, MI: Institute for Social Research. Johnston, L. (l975) Some Preliminary Results from Drugs and American Youth II: A Longitudinal Resurvey. Ann Arbor, MI: Institute for Social Research. National Clearinghouse for Smoking and Health (l973) Adult Use of Tobacco: l970. Washington, DC: U.S. Department of Health, Education, and Welfare.

SMOKING AS INDUCED BEHAVIOR Thomas C. Schelling It is often observed that smoking is a somewhat contagious activ- ity. The likelihood that a person smokes may depend on the number or the proportion of people who smoke in some socially relevant environ- ment; the amount one smokes may depend on how many surrounding people smoke or how much they smoke. The assertion of this principle is most emphatic with respect to ~young people. Smoking does not appear to be something that one takes up and becomes habituated to merely by learning that there are ciga- rettes and that they can be smoked. Cigarette companies are said to have promoted smoking among women by inducing film companies to have women smoke in movies and to have hired women to smoke on the streets 50 years ago. Recently the observation that certainly culturally homogeneous populations, like white male professionals, are smoking noticeably less than a decade ago suggests that we are affected by the habits of the people we associate with; and if some of us stop or re- duce smoking some more of us may do likewise, and if many of us quit we are less likely to go back to smoking than if only a few of us quit. This would not be peculiar to cigarettes. It is probably true of pizza, Scrabble, or yoga. Discovering whether this is so, and the speed and potency of the social influence; identifying the relevant influential populations for different kinds of people; and finding out if it affects only whether one smokes or affects as well what one smokes and how much and when and where, could be important. It might clear up some mysteries about what is happening and improve our understanding of current trends. It might point to strategies of social control—how or where to concentrate ef- forts, and techniques to influence smoking that would not otherwise occur to us. These are tantalizingly difficult things to discover through observation. It is unlikely that people can tell us, or we can tell each other, how one's own smoking is influenced by others' smoking. There is anecdotal evidence that, when a person gives up smoking, a husband or wife sometimes does, too. In schools the phenomenon may be recognizable but the parameters too obscure for the concept to be of much help. I once studied a behavior that has some of the same character- istics, neighborhood "tipping." The idea was that, if a white 86

87 neighborhood began to be penetrated by blacks, a few of the least tolerant of the old residents would move away, and their departure would induce a few more to move, and then still more would move, until the whole neighborhood had "tipped" from white to black. The idea was also applied to school systems in which the entry of blacks, beyond some number, would induce some whites to withdraw their children and, as those withdrawals shifted the ratio, others would withdraw their children, further shifting the ratio, in a cumulative process that might eventually cause the departure of most of the white children. I was impressed with how difficult it is to know what to look for to tell whether tipping was taking place. Was it evidenced by the accelerated departure of whites, or by their failure to be replaced by other whites in the neighborhood or the school? What territory con- stituted a neighborhood? Would the phenomenon look different according to whether people owned their homes or lived in rented homes? And in areas in which people live in apartments, would it occur apartment by apartment or neighborhood by neighborhood? How would you distinguish tipping from some discontinuity in the rate of normal turnover? I became convinced that familiarity with underlying models was a prerequisite to knowing what to look for, knowing how to formulate the questions to which investigation should be addressed. I also dis- covered that when one did formulate a model and worked out its implica- tions, there were some surprises. So let's see what the model would look like, what alternative forms the model might take, what the models imply we ought to observe, whether we might identify the phenomenon by observation or experiment, and what the implications are for policy. For the basic model we divide the relevant population into three parts: those who would smoke whether or not anybody else did, those who would not smoke whether or not anybody else did, those who would not smoke no matter how many did, and those who smoke if enough others do around them or whose probability of smoking goes up with the number of smokers in their environment. This model contains only the binary variable, smoking or not smoking, and not the amount, which we can allow for later. It assumes that smoking is positively related to the smoking of others, nobody quitting if the smoke gets too dense. It assumes that people differ in their susceptibility to the smoking of others—some will smoke if a few others do, some only if nearly every- body does. And we assume that only numbers matter, not the identities; the population is homogeneous with respect to smoking influence, every- body is equally "close to" everybody else for purposes of smoking. For everybody who would ever smoke, either independently or if enough others did, we now have a number—the number of others whose smoking would induce this individual to smoke. When it comes to chang- ing population sizes or sequestering parts of the population, it will matter whether absolute numbers or proportions are the influential variables. Let's talk in percentages. Everybody has his or her per- centage. Those who would smoke even if nobody else did have a critical percentage of zero. Somebody who would smoke if 20 percent of the pop- ulation did has the number 20, and so forth. For those whose smoking

88 depends on others we can plot a frequency distribution of their cri- tical numbers. It could be bell-shaped, U-shaped, J-shaped, hori- zontal, or triangular, skewed on either side of its modal value, which may be anywhere from 0 to l00 percent, and bimodal if the population is hetereogeneous with respect to susceptibility even though homogeneous with respect to influence. We diagram the model by converting that frequency distribution into cumulative form. This is a curve on the same horizontal axis that shows, for any percent smoking, the percent who would smoke if at least that many did. It starts at zero on the horizontal axis and rises steadily to the right (or at least does not decline); and on the right l00-percent side of the diagram its height is equal to the percentage of the population that smokes if and only if enough others smoke. Notice that it merely cumulates individual data. It does not tell us how many people will smoke. It does not contradict itself by saying that if 40 people smoke 30 people will smoke, but only that, for example, there are 30 individuals, each of whom would smoke if and only if at least 40 did. (The model is timeless; we are not concerned yet with how quickly somebody takes up smoking if his critical percentage is met by the population around him, or whether, if his critical number is 30 percent, he will more quickly take up smoking at that number if he has been surrounded for a long time by 25 percent smokers rather than 5 percent.) To finish the diagram, we shift this cumulative curve upward so that it intersects the vertical axis at the percentage who smoke inde- pendently of the rest. It then traverses, upward to the right, the horizontal scale from 0 to l00, and at the right side is the whole population less the number who never smoke at all. This curve can be a straight line, S-shaped, concave from above or from below, etc. For reference we draw the 45-degree line denoting equal vertical and hori- zontal distances from the origin. Wherever our curve is about that 45 degree line it indicates that the percentage of people who would smoke if jc did is greater than jc. Where the curve is below the 45-degree line, the percentage who would smoke if x did is less than x. Some shapes for that curve are labelled A, B, C, and D in the adjoining diagram. Curve A corresponds to a horizontal frequency dis- tribution for susceptible smokers amounting to half the population: there are as many for whom l0 is critical as for whom 20 or 30 or 60 is. A quarter smoke unconditionally and a quarter never smoke. Curve B corresponds to a bell-shaped distribution covering 75 percent of the population, with a modal value at about a third smoking. Curve C is also bell-shaped, but there is nobody who would smoke unless some others did, while a third of the population never smokes. And D has a shape similar to that of C. Let's articulate the model with Curve A, the straight line that rises from 25 on the left to 75 on the right. Imagine that 35 percent of the population were smoking—the 25 who smoke no matter what, and l0 among the more susceptible potential smokers. Among the potential smokers there are l7.5 percent who would be induced to smoke by the number currently smoking. Among them only l0 are smoking, so we expect another 7.5 percent to take up smoking. As they do they raise the

100 75 100 Horizontal axis: Vertical axis: Y 100 75 50 » number smoking in the social environment (independent var. number who would smoke if X were smoking (dependent variable 100 25 89

90 number, inducing more to take it up. For any number short of 50 there are some potential smokers whose critical number has been achieved and are not yet smoking, who can be expected to take up smoking, inducing others to do so until the curve crosses that 45-degree line at 50. If more than 50 were actually smoking there would be some smokers whose environment does not contain enough smokers to induce them to continue, and they will give it up, reducing the number and inducing others to give it up, until the number converges on 50. Generally, the dynamic interpretation of the model is that where the curve is above the 45- degree line, for some number smoking, the number will increase; and where the curve is below the number will decrease. There is a single equilibrium. We can ask what happens to the equilibrium number if, say, 5 per- cent of the confirmed nonsmokers are replaced by unconditional smok- ers. The curve shifts up 5 points, parallel to where it was. The new intersection is at 60 percent, l0 more than before. The 5 percent whose critical susceptibilities are between 50 and 60 have been induced to smoke, in addition to the 5 percent replacements. In the same way, if unconditional smokers numbering 5 percent are induced to quit abso- lutely, the curve drops 5 points parallel to itself and an additional 5 percent are induced to quit. We have a "multiplier effect." If Curve A ran from l0 on the left to 90 on the right, with a slope of 0.8, it would still intersect at 50, but now if we shift 5 percent of the population from uncondi- tional smoking to nonsmoking and the curve moves down parallel by 5 points, the intersection shifts to where only 25 are smoking. The multiplier is 5.0. (If the slope of the curve is S, the multiplier is equal to l/(l-S).) Curve B offers two stable equilibria, at the intersections where it crosses the 45-degree line from above. (The intersection in between is an equilibrium, but any divergence from that number would lead cumu- latively right or left to one of the stable equilibria.) We now have a population in which either a large percentage will engage in collec- tively self-induced smoking, or a large percentage will engage in self- induced nonsmoking. If the percentage is anywhere below 35 or 40, the process will converge on 20, while if the percentage is above 40 it will converge on 85. If the 85 were smoking and with inducements we could get more than half of them to quit and to stay quit long enough for the influence to have its effect, the number wanting to smoke would converge on that lower equilibrium and we could remove the inducements. Conversely, if the number were at that lower equilibrium but half of the population were temporarily moved into an environment in which three-quarters smoked, four-fifths of them would end up smoking, and on returning to their original environment would raise the proportion to over 50, at which point a few of the new smokers would quit but more of those who had stayed behind would be induced to take it up, and the percentage would shift to the higher equilibrium. Curve C is smiliar to B, the difference being that the lower equilibrium is at 0. There are no unconditional smokers. Any number below that first intersection is not self-sustaining; any number in excess of l5 will converge on the higher equilibrium at 65.

9l Curve D reflects an activity, of which smoking does not seem to be an illustration, that is not a viable custom. There are many who would smoke if enough others did, but no number is "enough" to be self- sustaining. A quarter of the population would smoke if half did, half if about two-thirds did, and two-thirds if everybody did, but not everybody will. It is worth noticing that if all the susceptibles represented in B, together with the confirmed nonsmokers, could be removed from the influence of those unconditional smokers, the result could be the pop- ulation of Curve D, and smoking would become an extinct activity. Those are some of the phenomena that the model can illuminate. Some of the things left out are: l. Is the minimum number that induces a person to smoke the same as the one below which the person will give up smoking; or may a person take up smoking only if half do but, having acquired the habit, give it up only when the proportion falls to a much lower figure? 2. Does the process—in either direction, taking up smoking or giving it up under the influence of other smokers—take months or years or decades; and does the time it takes to become a smoker, or to give up smoking, depend on how much above or below one's critical percentage the actual number is? 3. Is there really a "relevant population," or is the influence of smokers on a potential smoker related to social or geo- graphical distance or frequency of contact, so that a "domino model" is needed instead of critical mass? 4. Is the influential variable the percentage who smoke or the amount of smoking; and is an individual's own smoking an on-off variable or quantitatively influenced by the smoking going on. (People's smoking or not smoking could be in- fluenced by the quantity, even though each individual, once a smoker, smoked his own fixed amount; smokers in the pop- ulation would then be weighted, in our model, according to how much they smoked.) 5. Is the influence that smokers have on others affected by where or when or how or how conspicuously they smoke? 6. What constitutes the relevant "population" or "social environment" for an individual? How many social environ- ments does one participate in, and how do they get averaged together—e.g., peers where one works, others where one works, family, friends, neighbors, customers, business contacts, the doctor, taxi driver, or fellow airline passenger, or characters on television?

92 7. Even if the functional relation has the form of Curve A or B, starting at some positive value and ending short of l00 percent on the right, is the population really divided into unconditional smokers, susceptibles, and nonsmokers? Or is everybody's smoking determined by a number of probabilistic variables—job, income, health, climate, recreation, and the company one keeps—"environmental" variables that are sub- ject to change. (The exposition is neater if everybody has a critical number and for some the number is zero; but the description validity may be greater with a probabilistic interpretation for a population in which everybody's life and environment can undergo continual change.) The hardest thing is to recognize the critical-mass or multiplier phenomenon by looking at the data, even knowing what to look for. Knowing how many smoke doesn't tell us anything. The dynamics of change in the number smoking in some population may reveal something about the underlying mechanism, and comparisons of apparently similar populations in which the incidence of smoking is very different may tell us something. Some special-purpose information might help: know- ing who, in a given population, is aware of the frequency of smoking, or aware of recent changes in the frequency of smoking, might allow some inferences, but we cannot even be sure in advance that the people most susceptible to the smoking of others are aware of their own sus- ceptibility, or even aware in a conscious way of how much smoking goes on around them and whether it has been changing lately. Experiments might produce less ambiguous results, possibly deci- sive results. Experiments are hard to carry out but easy to imagine. Thinking up experiements is one way of trying to discover what we should be looking for when all we have is observation. And there are occasionally "natural experiments," when a discontinuous change occurs exogenously and we can study before and after. Furthermore, some anti- smoking strategies suggested by these models would be like experiments except in their motivation, and testing a suggested strategy might take the form of introducing it on an experimental scale. So we can begin by thinking about experiments and strategies. One experiment is to pick a random sample from a definable popu- lation, smokers and nonsmokers, and redistribute them among other cul- tures or population groups, noting the frequency of smoking in these destination cultures. We watch what happens to smokers and nonsmokers in the subsamples that go to these destinations. A "natural experiment" occurs when, say, a graduating class disperses into a variety of occu- pations and cultures and localities, or soldiers are discharged from service, or a firm goes out of business, as long as we think smoking behavior is not a determinant of the ensuing choice of occupation or location. (One might even follow law graduates into law firms if they vary in the incidence of smoking.) Another experiment would change the smoking-nonsmoking ratios in various subpopulations, preferably at random but if necessary through some process thought not to be correlated with the propensity to smoke, and to observe the effects on smokers and nonsmokers. Find two similar

93 nursing homes in which the proportions of smoking are about the same and exchange some smokers from one for nonsmokers from the other, or allocate the smokers and the nonsmokers among new residents to the one and the other, raising the percentage of smokers in one and lowering it in the other. If originally one-third of the population smoked, redis- tribute them so that in the one place half do and in the other one- sixth do. The contagion or critical-mass hypothesis suggests that pro- portions will diverge further, going below one-sixth in the one place and above one-half in the other. If we thought the model worked for less "total" environments, we could try the experiment in dormitories or places of work; military units, prisons, veterans hospitals, and possibly groups that are physically or socially isolated, like people who work night shifts, and people in activities that demand intensive teamwork, might be potential subjects for experiment. An interesting question is whether such discrimination would be in violation of civil rights, fair employment, etc. A third experiment would use some program known to be effective either in getting more people to smoke or in getting more people to quit. In one set of subpopulations the program would be conducted at a uniform level of intensity. In the other set of subpopulations the effort would be more concentrated, perhaps at quadrupled intensity, where "intensity" measures the investment and resources. If the inten- sified program is disproportionately successful, four times the re- sources producing more than four times the reduction in smoking—or four times the increase, if the program is promotional—one interpreta- tion could be the multiplier effect suggested by our model. That in- terpretation would be the more justified if "intensity" could measure the percentage of the population targeted rather than the concentration on each individual. (Whatever the interpretation, the experiment could generate criteria for the concentration or dispersion of program ef- forts. ) A variant of that last experiment, if the intensified program made a difference even though not a disproportionate one, would be to see whether the intensified program would have more permanent results than the less concentrated program. If the influence is of the kind depicted in Curve B, and intense program that shifts the number of smokers into the region of a new equilibrium can leave a permanently changed percentage of smokers even after the resources have moved on to another target population, while if the curve has the shape of A, there may be a multiplier effect that is reversed once the program is discon- tinued. The model proposes that some people may smoke independently of others' smoking, some may smoke if a few others do, some may smoke only if many others do. Suppose we not only knew this were true but knew how to recognize the "marginal" smokers who would quit if the propor- tion were moderately reduced, the ones who would quit if enough others did, and the ones who would go on smoking even if everybody else quit. And suppose we have an opportunity either to remove some people from a subculture, or to direct an effective antismoking program at some. If we remove the marginal smokers, or induce them to quit, we have little effect on anybody else; if we remove some who are not susceptible to

94 the smoking of others, or can induce them to quit, we'll get an extra reduction because of their influence on the marginal smokers. There's a possible dilemma here. It may be that the smokers most susceptible to moderate changes in the number smoking are also most susceptible to efforts to promote or to discourage smoking; if they are, the best strategy may be to concentrate on the marginal smokers and foregoing the multiplier effect, but if they are not, or not enough more vulner- able to outweigh the multiplier effect, our campaign should be directed at the others. It seems plausible that the people most amenable to antismoking campaigns, or the people who could most easily give up smoking, are the ones most susceptible to the smoking of others, but, though plausible, it could be altogether wrong. Except for people in the cigarette business, most people inter- ested in the susceptibility of smokers to the smoking of others are motivated against smoking. Experiments in how to reduce smoking, though subject to the usual legal and moral constraints on human exper- imentation, are at least believed not to conflict with the health and welfare of the subjects, smoking being dangerous to their health. But there may be two reasons for experimenting with ways to induce people to smoke. One is that prevention may be as good as cure, or better, and knowing how to induce smoking may help to know how to guard against induced smoking. The other is that if the inducement process is at all reversible, experiments that increase smoking can test hypotheses and suggest strategies as well as, or at least in addition to, experiments that reduce smoking, and it's even possible that experiments to in- crease smoking are easier or cheaper. But they may be objectionable. As presented above, the contagion model suggests that it is other people's smoking that influences one's own tendency to smoke, but cer- tainly some kind of awareness of that smoking must mediate the in- fluence. If everybody smokes and I don't know it, and I never see evi- dence of it, I'm not likely to be much influenced by it. There are probably many bad habits, or habits subject to social disapproval, that are somewhat contagious either through suggestion and imitation or through the reassurance to people that they are not alone in a practice considered depraved (or that people pretend to consider depraved), that are substantially quarantined against contagion because the shame or disapproval causes them to be kept out of sight. Perhaps among "smokers anonymous," people who backslide smoke so surreptiously that they do not endanger each other by smoking contagiously, but generally people smoke quite openly, except perhaps for children. This suggests an experiment. If we could induce a large fraction of the smokers to smoke only very privately, offering no inducements or encouragement to each other, their smoking habits might be reduced. If we could selec- tively forbid smoking at times and places where smoking had the great- est contagious influence, we might do little directly to reduce smoking but quite a bit indirectly; if everybody repairs to his room to smoke an after-dinner cigarette in private, they may collectively reduce the social inducement to smoke. A campaign to reduce smoking that cannot monitor against cheating may nevertheless benefit from sending the smoking underground; if people can be paid not to smoke at times and

95 places where the contagion is greatest they may contribute to the anti- smoking campaign even though they compensate by smoking at other times. It may be hard to enroll people in a reduced smoking campaign by paying them rewards if they have to be monitored around the clock to make sure they really do not smoke, much easier if they are merely en- joined not to smoke where it can be noticed, because we can notice it if they do. If it should turn out that some changes in smoking habits that do not reduce the total amount smoked can be more easily induced through various blandishments than changes that require actual reduc- tion, it may be possible to test the contagion model and also possible to base strategies on the quarantine principle.