5
Survey Design Issues

The collection of time-use data presents many interesting survey design issues. Time-use studies may have the goal of not only sampling across the population, but also across hours of the day, days of the week, and seasons of the year. Surveys often have a goal of completely accounting for time use in a specified period, usually a day. For some studies, however, a sample of a day’s activities may be sufficient to achieve the goals of the study. Often time-use studies need to collect information on where the respondent was during the activity, who the respondent was with, and whether the respondent was doing anything else in addition to the primary activity being reported. Some time-use studies also need to collect information on respondents’ characteristics, how respondents felt during the activity, and other behaviors of the respondent.

One session of the workshop was devoted to discussing the various methods used to collect data on time use and survey design issues surrounding these methods. This session focused on two methods of collecting time-use data, the time diary method and the experiential sampling method. Participants also discussed how the quality of data and the feasibility of these methods compare with other methods of collecting time-use data, such as stylized questions on surveys (questions that ask respondents to estimate how much time they spend in certain activities) and observational approaches to measuring time use. This section first briefly describes each of these methods and then discusses some sampling and questionnaire content issues that relate to the methods.



The National Academies | 500 Fifth St. N.W. | Washington, D.C. 20001
Copyright © National Academy of Sciences. All rights reserved.
Terms of Use and Privacy Statement



Below are the first 10 and last 10 pages of uncorrected machine-read text (when available) of this chapter, followed by the top 30 algorithmically extracted key phrases from the chapter as a whole.
Intended to provide our own search engines and external engines with highly rich, chapter-representative searchable text on the opening pages of each chapter. Because it is UNCORRECTED material, please consider the following text as a useful but insufficient proxy for the authoritative book pages.

Do not use for reproduction, copying, pasting, or reading; exclusively for search engines.

OCR for page 36
Time-Use Measurement and Research: Report of a Workshop 5 Survey Design Issues The collection of time-use data presents many interesting survey design issues. Time-use studies may have the goal of not only sampling across the population, but also across hours of the day, days of the week, and seasons of the year. Surveys often have a goal of completely accounting for time use in a specified period, usually a day. For some studies, however, a sample of a day’s activities may be sufficient to achieve the goals of the study. Often time-use studies need to collect information on where the respondent was during the activity, who the respondent was with, and whether the respondent was doing anything else in addition to the primary activity being reported. Some time-use studies also need to collect information on respondents’ characteristics, how respondents felt during the activity, and other behaviors of the respondent. One session of the workshop was devoted to discussing the various methods used to collect data on time use and survey design issues surrounding these methods. This session focused on two methods of collecting time-use data, the time diary method and the experiential sampling method. Participants also discussed how the quality of data and the feasibility of these methods compare with other methods of collecting time-use data, such as stylized questions on surveys (questions that ask respondents to estimate how much time they spend in certain activities) and observational approaches to measuring time use. This section first briefly describes each of these methods and then discusses some sampling and questionnaire content issues that relate to the methods.

OCR for page 36
Time-Use Measurement and Research: Report of a Workshop METHODS FOR MEASURING TIME USE Time Diary Method The most widely used method for collecting time-use data for a large sample of persons is the time diary. This method was used for the four major time-use surveys in the United States, as well as for other large time-use surveys in the world (Australia, Canada, the 1965 Multi-national Time Use Survey, and the forthcoming Eurostat Harmonized Time Use Survey; see Table 1 in Chapter 4). The essence of the time diary method is that respondents are asked to make a complete record of their activities over a period of time, usually one day. Although it is not always the case, time diaries usually ask open-ended questions about the respondent’s amount of time spent in activities. In other words, respondents enter the time an activity starts and finishes on a free-form basis, rather than in time slots of (say) 15 minutes. Activities are then typically classified and coded first into broad groups, and then into more specific groups according to a set standard. The first set of such coding standards were developed by Szalai for the Multi-National Time Budget Study. (Horrigan et al., 1999, summarize several different coding standards currently being used.) Time diaries can be filled out during the day, or retrospectively. Sometimes, survey respondents are interviewed to orient the respondent to the survey, and then diaries are left behind with the respondent to be filled out for the next day. These are called leave-behind diaries, which were used in the University of Maryland and Australian time-use surveys. In contrast, a retrospective diary is one in which a respondent is asked to recall what he or she did for the “designated diary day”—the day for which the respondent has been asked to report his or her activities. Retrospective diaries were used in the 1975-1976 and 1981-1982 Michigan studies and the Canadian time-use studies. The choice of a leave-behind diary or a retrospective diary has cost and data quality implications. Using leave-behind diaries tends to be more expensive because an orientation interview for the study must usually be given to the respondent prior to leaving the diary. An interview may also be needed after the diary is completed to clarify respondents’ answers or to fill in missing information.1 For retrospective diaries, respondents are oriented to the interview and provide responses in one setting or telephone call and so are less expensive. However, retrospective diaries rely on respondent’s ability to recall how they spent their time, which may affect data quality (see below). Although time diaries may be targeted to specific groups, they are readily 1   This was a principal reason that the 1975-1976 Michigan study used a retrospective diary (Juster, 1999).

OCR for page 36
Time-Use Measurement and Research: Report of a Workshop adaptable on a large scale. Typically, studies are conducted for a random sample of households. They are sometimes further randomized across days of the week, so that each randomly selected respondent is randomly assigned to a designated day or days to account for their activities. This method makes the entire sample of diaries representative across days of the week, which is important because there are likely systematic differences in time use across weekdays and weekend days. For leave-behind diaries, respondents are contacted and asked to fill out a diary for the next 24 hours. For retrospective diaries, respondents are often called or contacted one day and asked to recall what happened on the previous day, the designated day. Sometimes diaries are collected for several days for each respondent: a common method is to collect diaries for a weekday and a weekend day. Diaries are usually collected for only one household member who is randomly selected from all household members who are in the age range of the survey. The Australian studies collected time diaries from all members of the household over the age of 15; for other surveys, the expense of doing so usually limits the number of respondents. In addition to collecting data on the activities and the time spans of the activities in which respondents engage, diaries may also collect information on whom the respondent was with, where the activity took place, and whether the respondent was doing anything else. The Canadian survey also asked whether the respondent was helping someone in or out of the household or helping an institution. Diaries might collect only very basic information about the respondent, such as age, race, and household size, or they may have extensive sets of questions on specific topics. As noted in Chapter 4, the 1998 Canadian time-use survey included questions on volunteer activities, educational activities, time spent in unpaid activities, and time spent in child care activities. Questions on how much time a respondent spent in a particular activity are called stylized questions. They are often used to supplement time diaries to gather information about activities that the regular diary may not capture (for example, another household member’s time use or time spent being “on call” for child care—not actively caring for the child, but simply being present in case of an emergency—which may not be recorded as the primary activity for the time period and hence, may not be easily identified in diaries); that respondents are unwilling to report in a diary (sexual activity or drug use, for example); or for a longer reference period, since they may be unlikely to occur on the specified day. (Stylized questions are discussed further below.) Recently, time-use diaries have been conducted over the telephone with Computer Assisted Telephone Interview (CATI) technology. The Canadian time-use survey used CATI. CATI is often less expensive than paper-and-pencil interviews. Using CATI can also help speed up interviews and allows validation of answers while an interview is ongoing (for example, interviewers

OCR for page 36
Time-Use Measurement and Research: Report of a Workshop may be notified when a value given by the respondent falls out of a valid range of answers), which can improve data quality. A problem in using CATI for time diaries is that interviewers are sometimes given considerable discretion to classify activities while the interview is in progress. Since different interviewers may classify similar activities differently, there may be variability in the classification of activities across interviewers. This means that special and careful attention to establishing coding procedures and to training interviewers about these procedures is needed. The time diary method does have limitations. Most often, time diaries rely on respondent recall of activities, which is a potential source of error (see discussion below). Time diaries have also been found to underestimate activities with short time spans (see Juster, 1985), such as trips to the bathroom or going to the refrigerator for a snack. Experiential Sampling Method Another method for collecting data on how people spend time is called the experiential sampling method (ESM), which was primarily developed by Mihalyi Csikszentmihalyi and associates (see Csikszentmihalyi and Csikszentmihalyi, 1988; Csikszentmihalyi and Larson, 1992; and Zuzanek, 1999). ESM studies have typically been conducted to understand experiential, cognitive, and motivational aspects of activities, although these studies have also been used to estimate time spent in different activities. The typical method used in ESM studies is to give survey respondents a pager, beeper, or programmable wrist watch that is randomly activated (beeped, vibrated or buzzed) throughout the day. When the respondent is beeped, he or she is asked to fill out a self-report of what he or she was doing and about various aspects of the activity. A respondent may be beeped many times within a day, and the study may cover a day, a week, or a month. The goal of these studies is to sample how people spend time, by randomly beeping them during the day and asking them to record what they are doing, who they are with, and how they feel during the activity, etc. In general, the self-reports that respondents fill out include a core set of questions (Zuzanek, 1999): What day and time were you beeped? Where were you when you were beeped? Who were you with, what were you doing, and what were your thoughts at the time of the beep? Typically, these studies then ask questions about the respondent’s experiential, motivational, and cognitive aspects of the activities. Box 1 shows a typical form that respondents are asked to fill out when they are beeped. Like time diaries, ESM studies allow respondents to specify the activity in which they are participating. This is in contrast to stylized questions about time use, which must prompt respondents about a particular activity (i.e., ask them how much time they spent doing a named activity instead of allowing respondents to name

OCR for page 36
Time-Use Measurement and Research: Report of a Workshop Box 1 Typical Experiential Sampling Method Response Form

OCR for page 36
Time-Use Measurement and Research: Report of a Workshop

OCR for page 36
Time-Use Measurement and Research: Report of a Workshop the activity themselves). This flexibility of reporting may make classifying activities more difficult. But it also means the data analyst can make his or her own classifications of activities for different purposes. ESM surveys usually do not have the goal of completely accounting for an individual’s time use. They are, rather, typically used to explore processes of daily behavior (Zuzanek, 1999). Consequently, they have advantages and limitations for measuring time use, depending on the purpose of the study. One advantage, relative to time diaries and stylized questions, is that since activities are recorded soon after the beeper signal is sent, recall error is not a concern. Responses may also be less susceptible to normative editing within the framework of the experiential sampling method because respondents are asked to immediately record what they were doing and have less time to construct an “acceptable” response. Furthermore, because the random beep method is more free form and respondents are often encouraged to express how they feel during the activity, respondents may feel less pressure to record only normatively sanctioned responses and, hence, may give more genuine responses. Experiential sampling methods are useful in assessing human behavior and subjective emotional states and in understanding interpersonal relationships. An example of such a study is one where couples, who were both given beepers, were beeped at the same time (and sometimes in the same place) and asked to record their emotional states (Larson and Richards, 1994, as described in Zuzanek, 1999). The study uncovered a phenomenon of “unmutual togetherness;” even though couples were spending time together, they were emotionally not together. Finally, with longitudinal data, this method could be used to make causal links between emotional states. There are limitations to experiential sampling time-use studies. First, they are more expensive than other methods, and therefore, may not be expandable to a large national survey. Second, as Jiri Zuzanek reported, while the response rates for beeps is good, typically, the studies are more burdensome on participants, and there may be a selection bias in that the people who agree to participate in the study are systematically different from the people who do not agree to participate. Gaining respondent compliance for larger representative samples is perhaps the biggest challenge facing these studies (Zuzanek, 1999). Another limitation is that they typically are not designed to fully account for all time in a day (or other time unit). ESM studies may also miss certain types of activities because respondents are not willing to carry the beeping, paging, or vibrating device with them while participating in certain activities because they do not want to interrupt what they are doing to fill out a survey. Juster and Stafford (1985) found that beeping respondents at random times recorded fewer activities outside of the home than time diary reports, presumably because respondents were less willing to carry the beeper

OCR for page 36
Time-Use Measurement and Research: Report of a Workshop device with them outside the home. However, as more and more people carry pagers and cellular telephones, this problem may be reduced. Workshop participants were enthusiastic about the possibilities of the experiential sampling technique for certain purposive studies. Workshop participants suggested that it could be useful for health-related research, since emotions, feelings of pain or stress, and levels of exertion may be associated with activities. Participants also suggested that the method would be useful for understanding the time crunch or stress from the time crunch. Understanding the emotional states associated with different activities may also help classify activities by whether they give intrinsic or extrinsic rewards. Several participants suggested that ESM could be extremely useful in uncovering how people spend their time at work. Time diaries are likely to be difficult to use in the context of work because they require a time commitment on the part of the respondent (if the respondent is filling out a diary as the day goes along). Further, if the goal of a study is to obtain detailed information on work activities (as opposed to broad categories of work activities) and if retrospective diaries are used, respondents may have difficulty recalling their activities because activities may be done for short intervals or because there may be interruptions so that the respondent must attend to another matter. For these reasons, obtaining a “sample” of the day’s activities using the experiential sampling method or the random hour technique may be more appropriate. Either of these methods is likely to be a less burdensome method of collecting detailed data on time use at work than a method that completely accounts for all the time at work during the day. Both of them are less likely to be subject to normative editing of responses. It was also suggested by workshop participants that experiential sampling studies could be used to cross-validate data produced from time diaries and stylized questions. A similar technique–the random-hour technique–has been used in the past to cross-validate data: time diary respondents are called randomly on the day they are filling out their diaries to cross-validate responses for the given hour (see Robinson, 1999). While the experiential sampling method is unlikely to be the primary method of collecting data for a national study of time use, participants said that further work towards integrating time diaries with such studies would be beneficial for understanding time use. Stylized Questions Stylized questions are another method to measure time use, asking respondents how much time they spend in certain activities. Some examples are: About how much time do you spend cooking in your home during the week? About how much time do you spend caring for you child on a daily basis? Questions can be open-ended, where respondents can fill in a number

OCR for page 36
Time-Use Measurement and Research: Report of a Workshop of hours, or they can have a range of answers, where respondents choose one answer from categories such as “never,” “once a week,” “several times a week,” or “every day.” Many surveys with goals other than measuring time use have used these types of questions, usually as indicators of behavior patterns. For example, a health survey may ask how many times a respondent exercises each week. A survey with the purpose of measuring child development may ask how often a parent reads to a toddler. Although stylized questions have the advantage of being the least expensive way to measure how people use their time, using this method as a way to estimate time spent in activities across the population is troublesome, mainly because the answers that respondents give have a high degree of error in them: that is, respondents underreport or overreport time spent in different activities. There are several reasons that stylized questions are prone to error. First, people may overreport activities that are socially “good” activities. For example, Sandra Hofferth reported on comparisons of stylized measures of time spent reading to children to time diary reports of time spent reading to children; she concluded that parents exaggerate the amount of time they spend reading to their children through stylized measures relative to the amount of time reported in a time diary. John Robinson also described a study in which stylized reports of church-going were much higher than time spent at church as measured by diary data. Similarly, respondents may underreport socially “bad” activities, such as time spent watching television. Another reason that stylized questions may be measured with error is because respondents have a difficult time recalling what they have done over the time period the question references, if the question asks how much time the respondent spent doing a certain activity over the past week (or day or month or year). Respondents may also have a difficult time recalling and conceptualizing what a “typical” or “average” week is like in responding to such questions about time use in the activity over the week. For activities that take place on a daily basis, such as time spent commuting to work, the respondent may be able to make a much better estimate of the average time spent in the activity over the week. However, for activities that take place on a more variable basis, such as time spent talking on the phone, respondents may have a more difficult time recalling the amount of time spent in the activity. (These recall issues are discussed further below.) Third, stylized measures of time use do not take into account any activities that occur simultaneously. This may be important for measuring passive activities, like watching television. The television may be turned on for many hours a day, but respondents may be doing many other things while the television is on. When asked how many hours of television were watched each day, respondents may not know whether to report the time spent solely watching the television screen or the total time they spent passively “watching” television while doing other activities. Depending on whether the re-

OCR for page 36
Time-Use Measurement and Research: Report of a Workshop spondent judges time spent passively watching television as an activity worth reporting, the amount of time spent watching television may be over- or underreported. Another disadvantage to stylized questions is that the questions must be worded so that the respondent understands the types of activities for which the respondent is to report time use. That means that the activities need to be defined and classified within the question; in contrast, in a time diary, the activities are coded after the respondent has completed the diary. Despite the problems with stylized questions, workshop participants agreed that there is a role for this method. Stylized questions can effectively be used to measure incidence of certain activities, especially those activities that occur infrequently, such as how much time was spent on vacations or how many days were spent in the hospital over the past year. Some workshop participants suggested that stylized questions are better for a specific and short time period (such as whether the respondent did a particular activity yesterday) than questions about usual activities over a day or week (such as whether the respondent usually does a particular activity on a weekly or daily basis). Cognitive testing of survey questions that ask for stylized reporting of time spent in activities can enhance the abilities of these types of questions to obtain valid measures of time use. Some previous research shows that some stylized questions are not measured with as much error as others (e.g., time spent at work, traveling, and shopping [Juster and Stafford, 1985]). As Francisco Samaniego suggested, well-designed stylized questions could be selectively used to obtain very specific information. Well-designed stylized questions cannot substitute for a complete account of time spent in all activities, but, they may be suitable for counting the time spent in a very specific activity. Observational Approaches On some occasions, direct observation of an individual’s daily activities may be possible. In observational studies, an “interviewer” records what the respondent does during the day as opposed to the self-reports used in diaries, ESM studies, and stylized questions. For example, anthropologists have long used this approach in studying different cultures, and some child development studies use cameras or observational rooms to record how children spend their time in a controlled setting. John Robinson reported that he has recently trained students to “shadow” people they know throughout a day and record their activities, which are later validated against the trackee’s own diary report of activities for the day. Use of electronic tracking devices might also be included in the category of observational studies. Robinson (1999) gives the example of media rating services that use electronic badges to record

OCR for page 36
Time-Use Measurement and Research: Report of a Workshop when the participants are near operating televisions or radios as a way of understanding time spent watching television. The key advantage of observational studies is that they are very accurate. Their biggest drawbacks are that they are intrusive, may contain little useful information, and are expensive. Furthermore, since consent is usually needed from participants, participants know that they are being watched which means that they may change their behavior for the camera or the observer. However, in some settings, observational studies can be very useful, both in their own right and as a way to validate data collected through other means. For example, given parental consent, observational studies of children in day care settings or even school settings may provide a good source of data for studying child development. Workshop participants were very supportive of the use of multiple methods in a single study. As Norman Bradburn noted, the advantages and limitations of each method are known. Further understanding of how these methods can be used in tandem to get to the information that is needed would be valuable in understanding time use. For example, it will be useful to know which stylized questions can be used in conjunction with diary studies to save survey costs. Or, a single study may find it useful to measure some activities through experiential sampling and others through a time diary. Understanding the methodological underpinnings of using these methods in tandem is an important area for future research. In addition, if certain methods are known to produce biases in reporting, research could be conducted to assess the extent of the bias. If the bias can be determined, then less expensive methods of collecting data can be used, despite their biases, because adjustments can be made to correct the bias. SAMPLING ISSUES Respondent Recall in Diary Surveys One problem with retrospective diary studies is that respondents are asked to recall what they did, usually over the past 24 hours, but respondents may not be able to recall accurately what they have done. This measurement problem also relates to sampling issues for diary studies in which, it is common to obtain diaries across all the days of the week. It is important to sample across days of the week because time use is likely to be different for weekends and weekdays, and perhaps even between weekdays, (Mondays and Fridays may not be the same as Tuesdays, Wednesdays, and Thursdays). Typically, once a household is randomly chosen for the sample, it is then randomly assigned to provide a diary for a day of the week (or a random weekday and weekend day). Using this “designated” day method has implications for recall error, because it is often difficult to contact a respondent soon after the respondent’s designated day. At issue is how long after the designated day

OCR for page 36
Time-Use Measurement and Research: Report of a Workshop respondents can be contacted and expected to provide high-quality recalls of their activities for the designated day. Should respondents who are not contacted the day after be contacted two or more days after (and still report on the designated day), or should they be counted as nonresponders, or, perhaps should they be contacted the following week on the same day of the week? There has been some research on this topic. In her discussion of the Statistics Canada survey, Lorna Bailie explained that it allows for a 48-hour recall period without significant deterioration of data quality. In contrast, Juster and Stafford (1985) found that recall error rates increased if the recall period for a designated day that was a weekday was more than 24 hours (e.g., if a respondent was contacted on Wednesday or later and asked to recall what happened on the preceding Monday). However, this study also found that the recall period for weekend designated days could be extended for up to a week with little increase in error rates. Echoing Juster and Stafford’s findings, Norman Bradburn cautioned against more than a 24-hour recall, as memory deterioration speeds up after that, especially for weekdays. Workshop participants discussed some ways that recall error could be reduced. Using CATI methods is one way, where cross-validation of answers can take place as respondents can be asked to clarify their answers, if needed, during the interview. With further developments in cognitive research in surveys, questions could be better designed to enhance respondent recall. Participants also suggested that with longitudinal surveys, where diaries are kept for each respondent more than once, respondents become better at recalling their activities as they fill out more diaries and become familiar with the diary processes. Data on Multiple Days of the Week for Each Observation Individuals’ time allocations to different activities can vary greatly across the days of the week and across seasons of the year. The activities that occur on weekdays are likely to be quite different than the activities that occur on weekend days. Furthermore, a one-day diary might represent an atypical day for the respondent. Over a large national sample where days of the week are equally represented (or if there is a controlled sample of days), this may not be as great an issue, because any atypical days would wash out in the aggregate. However, in using microdata for examining individual behavior, it may be crucial to obtain accounts of time use for multiple days for each individual. The ESM technique typically collects data for each day of the week for each respondent in the sample. Variation in time use across different days of the week could be captured using this method. Time diaries are typically not collected for every day of the week, although previous time diary surveys have collected data for a couple of days of the week for each respondent. The 1975-1976 and 1981-1982 Michigan time diary studies collected data on four different days of the week for each sample member (over

OCR for page 36
Time-Use Measurement and Research: Report of a Workshop the course of a year). The Australian surveys collected diaries on two consecutive days for each respondent. Collecting multiple diaries from each sample member may increase the cost of a survey, and it may have implications for response rates. Respondents may be reluctant to agree to participate in a diary study if they are expected to fill out more than one diary. In a methodological test conducted in conjunction with the Canadian time-use survey, adults in single-person household who were asked to complete diaries for two days had an 88 percent response rate, but married couples’ response rates (where both spouses were asked to fill out diaries for two days) were only 46 percent. Having multiple diaries from each respondent can reduce sampling error and can be done in a cost-effective manner. As noted above, some evidence indicates that recall for time use on a weekend deteriorates at a slower rate than for weekdays. If a study collects a weekday and weekend day diary for each respondent, it may be possible to get diaries for both days during one interview, a day after the designated weekday. For example, a respondent with designated days of Saturday and Tuesday would be interviewed on Wednesday, the day after the designated weekday. This method may be more cost-efficient and also less harmful to response rates if both diaries can be completed during one, instead of two, telephone calls. Data on Multiple Household Members Another sampling issue raised at the workshop was whether diaries should be collected from more than one person in a household. From a conceptual standpoint, one argument for collecting time diaries from multiple persons in a household is to better understand labor force participation of household members, intrahousehold resource and time allocation, and who delivers family care (for children or other relatives). Child development researchers may also be interested in time use by children and adults in a household. In the public policy arena, it is also useful to have time diaries for multiple persons: for example, to understand the effects of a tax credit for households that provide care for elderly relatives, it is important to have data on the time use of all household members who could be providing the care. Collecting diaries for multiple persons within a household may be difficult because interviewing more than one member of the household means that the interviews will take longer. Statistics Canada believed that this was one reason for the low response rate in the test study where diaries were obtained from multiple household members of the Canadian time-use survey. It may also be difficult to find a time to interview each household member who is part of the study. If one household member is interviewed, but other household members are not available for interview, the question arises about which day the second household member should be interviewed. Again,

OCR for page 36
Time-Use Measurement and Research: Report of a Workshop recall is a key issue. If a wife is interviewed on a Wednesday about her Tuesday activities, but the husband is not available for an interview on Wednesday, should the husband be interviewed on Thursday for Tuesday’s activities (with some recall error) or for Wednesday’s activities (which means activities for husbands and wives are recorded for different days)? Many workshop participants agreed that there are some tricky issues to resolve in collecting diaries for multiple household members. However, the availability of data on multiple persons would greatly enhance the value of such data for understanding household behavior. Furthermore, as Juster (1999) argues, there are benefits for minimizing sampling error and statistical noise to collecting information on multiple household members and at multiple times for each household member (see also Kalton, 1985). Many workshop participants argued that data on multiple household members should be collected. MATCHING DATA COLLECTION TO DATA USE In choosing a method for measuring time use, the analytical purpose of the study should be the guiding principle (Bittman, 1999). While time diaries are probably the best method for collecting data on time use on a large scale, most workshop participants agreed that the other methods clearly have roles to play in collecting time-use data. For example, time diaries and stylized questions often do not provide much detail on respondents’ use of time while at work. ESM studies could be useful in such a setting, however, as they are less subject to normative editing, require less time by respondents than do time diaries, and can be relatively easily adapted to diverse work settings. For other situations, it may be possible to use stylized questions to measure some activities, for example, to measure events that happen quite infrequently, making them more easily recalled by respondents, such as time spent on family vacations. Which covariates are collected with time-use measures—that is, the supplementary information about respondents and their behavior—should also be guided by the uses of the data. For example, to assess subjective well-being during activities, an experiential sampling study may need to collect data on a person’s emotional state and surroundings. To understand household labor force participation decisions, it is important to have data on wages, past work experience, and income of household members. If the goal of a time-use survey is to better understand household production, then data are needed on the technology and capital stock available to the household, as well as on inputs to the household production process. It is likely that no single survey is going to be able to collect all the covariates that researchers will want or need. Therefore, a time-use survey may need to be linked to other data sets with a wider range of covariates or

OCR for page 36
Time-Use Measurement and Research: Report of a Workshop modules to the survey could be added as needed. Workshop participants emphasized the need to carefully consider which covariates are collected and how they are collected in developing a time-use survey. THE 24-HOUR CONSTRAINT AND STYLIZED QUESTIONS As has been noted by Robinson (1985) and others, data collected in stylized time-use surveys often violate the 24 hours per day (or 168 hours per week) constraint. There are several suggested ways of dealing with this effect. For example, in the pilot BLS time-use survey carried out by Westat in 1997, the working rule of shrinking reported time toward the constraint was advocated: the recommended strategy would shrink 30 reported hours in a given day to 24 by proportional shrinkage for all reported categories (from 10 hours of sleep to 8, for instance). It appears that approaches to the treatment of such data are largely ad hoc, and there is considerable room for additional research on this issue and for more comprehensive guidance on the analysis of such data. Some ways of dealing with the 24-hour problem are covered in Chapter 3 (Conceptual Issues) under “Simultaneous Activities.” Other ways are discussed here. In considering a simplified version of a (stylized) time-use survey in which each respondent reports the number of hours spent on two mutually exclusive and exhaustive categories (e.g., work and nonwork), the number of hours of each activity reported in a day should clearly total 24. However, there might be a number a perfectly plausible reasons why the number of hours do not add to 24, most likely because some simultaneous activities (e.g., time spent grading papers while providing child care) are counted twice. It is also possible that the tally in each category is done with relative independence and that the separate reports simply do not obey the constraint. In his presentation at the workshop, Samaniego illustrated modeling these data as observations from a mixture of two bivariate normal distributions, pointing out that there may well be two types of respondents—those who obey the constraint and those who do not. He derived the maximum likelihood estimate of the bivariate mean. This exercise was devised to demonstrate that the “right approach” to estimating the mean time spent in each of the two activities depends crucially on the model assumed to govern the available data. In a particular example, Samaniego demonstrated that the maximum likelihood approach led to a markedly different prescription for estimating mean time use than the “shrink toward the constraint” strategy advocated in other studies. Samaniego suggested that the modeling of time-use survey data merited more research and greater care and that mathematical statistics might be usefully brought to bear on some on the thorny questions posed by the constrained estimation problems that often accompany such surveys.