Daily Data Collection of Sexual and Other Health-related Behaviors.

Studies of sexual behavior in the U.S. typically have relied on self-reported behaviors that are recalled retrospectively, sometimes over a rather long time, or have asked respondents to report about their most recent sexual event. Shiffman and Stone (1998) point out that the former gives rise to inaccuracies not only because of forgetfulness, but because the task is cognitively complex. They point to a body of research that indicates that recall is not a simple process of retrieving facts from memory, but rather a reconstruction of events that can lead to considerable bias in the data. Although asking about a specific event, such as the most recent sexual occasion, reduces the cognitive complexity of the task, Shiffman and Stone argue that even in such cases respondents often invoke an "heuristic reconstruction" that can introduce bias (p. 121). Moreover, it is not known to what extent a specific event is a typical event in the life of a given respondent. Because they are less subject to recall bias, yet provide data on more than one event of interest, some researchers have favored the use of daily diaries to study sexual behaviors (Fortenberry, Orr, Zimet, & Blythe, 1997; Gold & Rosenthal, 1995; Leigh, 1993; Leigh, Gillmore, & Morrison, 1998).

Although diaries have not been used widely to study sexual behaviors, their use in studies of other social and health phenomena is neither new nor rare. A review of research published in the past 5 years revealed over 300 citations to diary studies involving daily data collection in the psychological literature alone. In these studies, diaries have been used for two primary purposes. First, they have been used in nonexperimental research to obtain data on diverse behaviors such as sexual behavior (Fortenberry et al., 1997; Gold & Rosenthal, 1995; Leigh, 1993; Leigh et al., 1998), smoking (Thomsson, 1997), drug and alcohol use (Carney, Tennen, Affleck, del-Boca, & Kranzler, 1998; Leigh et al., 1998; Rabow & Duncan, 1995; Webb, Redman, Sanson-Fisher, & Gibberd, 1990), mood states (Campbell, Chew, & Scratchley, 1991; Edelmann & Connolly, 1998; Vittengl & Holt, 1998), and health behaviors (Berg, Dunbar, & Rohay, 1998). Second, diaries have been used as a method of self-monitoring, or treatment, to help motivate changes in behaviors such as drinking (Cronin, 1996), smoking (Abueg, Colletti, & Kopel, 1985; Moss, Prue, Lomax, & Martin, 1982; Singh & Leung, 1988), and inappropriate social behaviors (Pope & Jones, 1996).

When diaries are used for assessment, rather than for treatment, it is necessary to assume that the method of data collection does not alter the behaviors under observation. In view of demonstrated treatment effects for self-monitoring, this assumption has been questioned. Some time ago, Fremouw and Brown (1980) argued that "self-monitoring studies almost invariably include some implicit or explicit demands to change behavior" (p. 212). About the same time, and citing evidence from several studies, Verbrugge (1980) pointed out that frequent data collection also can produce fatigue effects (i.e., a decline over time in the amount or frequency of behavior reported). Indeed, several diary studies have shown evidence of response decay over time (e.g., Gerstel, Harford, & Pautler, 1980; Leigh, 1993; Verbrugge, 1980), but others have not (Lemmens, Knibbe, & Tan, 1988; Persky, Strauss, Leif, Miller, & O'Brien, 1981; Searles, Perrine, Mundt, & Helzer, 1995). Stone, Kessler, and Haythornthwaite (1991) suggested that, in addition to reflecting possible fatigue effects, response decay also might be the result of other types of reactivity (changes in behavior as a result of its being measured), such as socially desirable responding. Some diary studies have shown evidence of such reactivity (Berg et al., 1998; Mulheim, Allison, Heshka, & Heymsfield, 1998), yet others have not (Cruise, Broderick, Porter, Kaell, & Stone, 1996; Samo, Tucker, & Vuchinich, 1989; Sobell, Bogardis, Schuller, Leo, & Sobell, 1989). It is generally believed that studies of sensitive behaviors, like sexual behavior, are especially prone to reactivity.

The motivations of study participants in the different types of diary studies (e.g., assessment or treatment) may be one factor affecting the likelihood of reactivity. It may not be frequent self-recordings per se that produce the effect, but heightened sensitivity to one's behavior when one is consciously trying, or senses a demand, to alter that behavior, as might occur in a treatment study. Reactivity effects also may dissipate over time as people become more habituated to the self-recordings. Kopp (1988), in a review of self-monitoring studies, noted that many studies followed participants for no more than 3 weeks, making it difficult to tell if such effects would wear off over time. If effects of frequent self-recording are of limited duration, then frequent self-recording may produce valid data when the period of data collection is long enough for some types of reactivity effects to wear off. However, longer periods of data collection may increase the likelihood of other types of reactivity, such as fatigue effects.

The inconsistencies in the literature regarding the reactivity of self-recordings have led researchers to try to identify conditions under which such effects are more likely to occur (for a review, see Fremouw & Brown, 1980), explicate the underlying mechanisms accounting for such effects (Kazdin, 1982; Kirby, Fowler, & Baer, 1991), and suggest ways to adjust for these effects (e.g., Stone et al., 1991). The reactive effects of frequent self-reporting are presumed to be caused by heightened evaluation apprehension, socially desirable responding, self-regulation, discriminative function effects, and fatigue (Kazdin, 1982; Kirby et al., 1991; Verbrugge, 1980), many of which might be exacerbated when measuring sensitive behaviors like sexual activities. It has been argued that these effects can be minimized by using discrete, rather than continuous, recordings; recording more than one behavior; and recording behaviors that study participants express no desire to change or experience no demand to change.

The present study is an attempt to extend the findings on effects of frequent self-recordings of sexual behavior, using daily data collected over a period of 8 weeks. We addressed three questions: (a) Is there evidence of fatigue effects over the 8-week period of daily data collection, (b) is there evidence of socially desirable responding, and (c) do the trends over time in reports of behaviors differ by method of data collection or gender of the study participant? We assumed that fatigue effects were evident if there was response decay over time, although we recognize that for some behaviors socially desirable responding could lead to the same pattern of responses. For that reason, we examined not only sexual behaviors but also less sensitive socially desirable behaviors such as eating fruits and vegetables, as well as other types of socially undesirable behaviors such as smoking and drinking. Socially desirable behaviors should increase over time and socially undesirable behaviors should decrease over time if there is socially desirable responding. In contrast, fatigue effects would result in response decay for both socially desirable and socially undesirable behaviors. In addition, daily data collection occurred over a sufficiently long period (8 weeks) to allow some types of initial reactivity, if it occurs, to wear off. To systematically examine whether different methods of data collection are more prone to fatigue and social desirability effects, study participants were randomly assigned to either a self-administered diary condition or a telephone-administered interview condition; question wording and format were identical in both conditions.

The data come from a larger study of the relationship between drinking/drug use and sexual risk-taking. We employed an analysis technique (hierarchical linear modeling, HLM: Bryk & Raudenbush, 1992) in which changes in individual rates of behavior over time are analyzed, in contrast to prior studies that typically have focused on changes in group means (i.e., a within-subjects as opposed to between-subjects analysis). We also employed a larger sample than that typically reported for studies using diary methods. Explicit steps were taken to reduce reactivity, including (a) using discrete rather than continuous data collection, (b) collecting data on more than the target behaviors and embedding these questions among other health-related questions, and (c) describing the study as one in which "health habits" (as opposed to the relationships between drinking/ drug use and sexual behavior) were being examined.

METHOD

Study Participants

Study participants were 90 men and 87 women (N = 177) between the ages of 18 and 23 (mean age = 20.51, SD = 1.31) who were randomly sampled from student registration records at a large northwestern university. They were asked to participate in a study of "health habits" thats purpose was described as examining daily patterns of a variety of health-related behaviors. Potential study participants were screened and deemed eligible if they were (a) unmarried, (b) not in a steady monogamous relationship of longer than a 6 month duration (these criteria ensured that participants were at higher risk of acquiring STDs), (c) currently sexually active (defined as having had sex on at least four occasions in the prior 2 months), (d) drank at least occasionally (defined as having drunk alcohol on at least four occasions in the past 2 months), (e) sometimes drank in the context of sexual activities (but not always or never), and (f) had used a condom at least once in the past year. The latter four criteria helped ensure that we would have adequate variation on the variables of interest to achieve sufficient statistical power to examine the relationships of interest for the main study: the relationship between drinking and risky sexual behaviors. The majority (60%) reported that they had had vaginal sex once a week or more in the 2 months preceding the study; the modal frequency of vaginal sex during this period was 2 to 3 times per week.

The majority (76%) of participants identified themselves as White; the remainder identified themselves as African Americans (1.7%), Asian Americans (11.9%), Hispanics (4%), and Native Americans (1.7%). A small number (2.3%) identified themselves as multi-racial or other (2.3%). Just over a quarter of the study participants lived in dorms (5.1%) or fraternities or sororities (22%); the rest lived in off-campus housing with roommates (35%), with parents or other relatives (22%), or lived alone (15.8%). Most participants were third- (29.2%) or fourth- (31.6%) year students, but 8.2% were first-year students, 19.9% second-year students, and 11.1% were fifth-year students.

Procedure

Study participants were randomly assigned to either a written diary condition or a telephone interview condition. Participants in both conditions were sent initial questionnaires (entry surveys) to complete and return before daily data collection began. Daily data were then collected for a period of 8 weeks (56 consecutive days). In the diary condition, participants were mailed packets each week that contained instructions, seven daily diaries, and seven self-addressed stamped envelopes with which to mail back, daily, each day's diary. They were also telephoned weekly to offer encouragement, answer questions, and remind them to mail back each day's survey. In the telephone condition, trained interviewers telephoned participants daily. Participants were mailed copies of the daily survey form, which was identical to that used by those in the diary condition, to follow as the interviewer asked the questions. At the conclusion of each interview, an appointment was made for the next day's interview. After the 8 weeks of daily data collection, participants in both conditions were mailed a brief questionnaire (exit survey) to complete and return. Participants were each paid $15 for the entry survey, then weekly for the daily diaries ($2 for each daily report, a $3 bonus for each week with no missing days), and $10 for the exit survey. All procedures were approved by the university's human subjects review committee.

Materials

Entry survey. The entry survey included questions on demographic information, sexual history and current sexual activities, alcohol and drug use history and current use patterns, health-related information, and measures of individual difference traits such as impulsivity.

Daily report form. The daily report form included questions on sexual behaviors, smoking, diet, dental care, exercise, seatbelt use, sleeping patterns, and drug and alcohol use. Participants reported activities that occurred during the 24 hours prior to data collection the first day, and the previous day's data for subsequent days of data collection. To reduce demand characteristics, questions on sensitive behaviors were separated by other items. Once participants became accustomed to the questions and their flow, it typically took no more than about 5 minutes to respond to all the questions each day.

Exit survey. The exit survey included questions about drinking, drug use, and sexual activities that had occurred during the period of daily data collection. Participants' reactions to study participation also were measured.

Measures

Study participants were asked each day whether they had had vaginal sex and, if so, whether they used a condom; smoked cigarettes; drunk alcohol; eaten two or more servings of fresh, canned, or frozen vegetables; or eaten two or more servings of fresh, canned, or frozen fruit since the previous day's data collection (or past 24 hours if the first day of data collection). We also included questions about oral and anal sex; however, the former rarely occurred in the absence of vaginal or anal sex (90% of all sex occasions included vaginal sex; oral sex, in the absence of vaginal or anal sex, occurred on only 9% of the occasions). Therefore oral sex could not be separately analyzed. Anal sex occurred too infrequently to analyze (only 1.8% of sex occasions were anal sex). Therefore, we focused on vaginal sex in the analyses reported in this paper.

We computed the proportion of valid days (the number of days each week that data were provided) each week that respondents reported having engaged in each of five behaviors: vaginal sex, drinking alcohol, smoking cigarettes, eating fruit, and eating vegetables. Condom use was computed as the proportion of vaginal sex occasions each week for which condoms were used. For each of the 8 weeks, each of these behaviors could take on values that varied from 0 to 1. For example, if a respondent completed 6 days of data collection in a given week, and reported having sex on 2 of those days, his or her value for sex is 2/6 = 0.33 for that week. Over the 8-week period, study participants reported having vaginal sex on average . 19 (s.d. = .20) valid days/week, drinking alcohol on .34 (s.d. = .22) valid days/week, smoking cigarettes on .32 (s.d. = .39) valid days/week, eating fruits on .42 (s.d. = .28) valid days/week, and eating vegetables on .55 (s.d. = .28) valid days/week. Table 1 displays the average weekly proportions for each behavior.

Table 1. Mean Proportions and Standard Deviations for Reported
Behaviors

Proportion valid        Had vaginal    Used      Drank
 days reported              sex       condoms   alcohol

Week 1
                   M       .2332       .4132     .3968
                   SD      .2500       .4571     .2712

Week 2
                   M       .1955       .4663     .3562
                   SD      .2304       .4820     .2803

Week 3
                   M       .1995       .3895     .3494
                   SD      .2518       .4557     .2694

Week 4
                   M       .1743       .4187     .3096
                   SD      .2277       .4778     .2553

Week 5
                   M       .2056       .4302     .3320
                   SD      .2451       .4876     .2665

Week 6
                   M       .1823       .4122     .3313
                   SD      .2292       .4736     .2839

Week 7
                   M       .1610       .4239     .2873
                   SD      .2146       .4817     .2567

Week 8
                   M       .1367       .3675     .2923
                   SD      .2069       .4747     .2603

Proportion valid          Smoked      Ate       Ate
 days reported          cigarettes   fruit   vegetables

Week 1
                   M      .3414      .3986     .5109
                   SD     .4193      .3159     .3266

Week 2
                   M      .3222      .3971     .5679
                   SD     .4065      .3206     .3202

Week 3
                   M      .3110      .4411     .5613
                   SD     .4048      .3369     .3427

Week 4
                   M      .3061      .4260     .5532
                   SD     .4040      .3302     .3298

Week 5
                   M      .3206      .3397     .5743
                   SD     .4019      .3329     .3281

Week 6
                   M      .2856      .4510     .5444
                   SD     .3977      .3399     .3297

Week 7
                   M      .2786      .4395     .5387
                   SD     .3939      .3403     .3339

Week 8
                   M      .2963      .4336     .5822
                   SD     .3994      .3515     .3193

Of our sample of 177 participants, 4 participated in the study for 1 week only and so were eliminated because a trend cannot be computed on a single data point, leaving 173 respondents in these analyses. This is the sample size for each behavior analyzed except condom use. Participants who had no vaginal sex during the reporting period (n = 32), had vaginal sex only once, or had vaginal sex during only one week (n = 29) are not included because sex is a necessary condition for condom use and more than one data point is needed to analyze a trend; the sample size for this analysis is therefore 116. It should be noted that 95% of the 173 participants included in all but the condom use analyses provided data for 6 or more weeks.

RESULTS

Analytic method.

To determine whether there is a trend over time in the data, and whether the method of data collection (self-administered diary vs. telephone-administered interview) or gender influenced the trend, the data were analyzed using hierarchical linear modeling (HLM). HLM is a generalization of multiple regression, designed for analyzing data that have an inherent hierarchical or "nested" structure. Examples of nested structures include individuals nested in larger units (e.g., students in classrooms) and multiple observations of individuals (e.g., as occurs in longitudinal studies). In both cases, there is an intraclass correlation; that is, observations within each unit or individual do not meet the usual regression assumption of independence, leading to standard errors that are too small and therefore to inflated Type I error. HLM accounts for the intraclass correlation and yields unbiased estimates of the standard errors. Another advantage of HLM is that cases with missing data can be included in the analysis, reducing the bias due to missing data. Thus, data from all respondents, even those who missed 1 or more (but fewer than 7) weeks of data collection were included in the analysis.

In HLM analyses, we can separate the variance into two distinct, but related, linear models or "levels." Level 1 is the within-unit (here, within-person) model; Level 2 is the between-unit (here, between-person) model.

In these analyses, the Level 1, or within-person, model is:

Y = [B.sub.0] + [B.sub.1]* (week) + e,

where Y is the individual's weekly rate of the behavior (vaginal sex, condom use, etc.), [B.sub.0] is an intercept, [B.sub.1] is a random slope, and e is a normally distributed error term with a mean zero and variance [[Sigma].sup.2]. Thus, each individual has an intercept representing his or her initial rate of the behavior (week 1), and a slope that expresses his or her change in that rate over the 8 weeks.

The Level 2, or between-person, model is:

[B.sub.0] = [G.sub.00] + [U.sub.0] and

[B.sub.1] = [G.sub.10] + [G.sub.11] * (condition) + [G.sub.12] * (gender) + [U.sub.1].

[G.sub.00] is the group average intercept, and [U.sub.0] is the average individual deviation from the intercept. In the second equation, [G.sub.10] is the group slope, [G.sub.11] the effect of data collection condition (written diaries vs. phone interviews), [G.sub.12] the effect of gender, and [U.sub.1] is the average individual deviation from the group slope.

Separate models were estimated for each of the six behaviors: average weekly frequency of sex, condom use, smoking, drinking, vegetable consumption, and fruit consumption. Time (weeks 1-8) was entered in the models as an uncentered predictor.

Trends Over Time in Self-reported Behaviors

Table 2 shows the estimates of coefficients, with associated standard errors, t-ratios, and p-levels, for the intercept, slope over time, condition, and gender effects, for each of the six behaviors. Three behaviors showed significant decreases over time, and three showed no significant changes. Over the 8 weeks the proportion of days per week on which respondents reported engaging in vaginal sex, drinking alcohol, and smoking cigarettes decreased significantly. The proportion of days on which respondents reported engaging in vaginal sex declined 41% over the 8 weeks. Similarly, the proportion of days on which respondents reported drinking alcohol and smoking cigarettes declined 26% and 13% respectively (Table 1). There were no significant differences in the proportion of vaginal intercourse occasions for which condoms were used, nor for the proportion of days per week on which respondents reported eating three or more servings of fresh, canned, or frozen fruit or eating three or more servings of fresh, canned, or frozen vegetables. Gender and data collection method, the Level 2 predictors, did not prove to be significant in any of the models, and thus do not help to explain the differences in the proportions of behaviors that changed over the 8 weeks.

Table 2. HLM Estimates of Coefficients for Intercepts and Slopes, by
Behavior

Reported behavior        Coefficient   Standard error     t        p

Had vaginal sex
  Intercept [G.sub.00]      0.2317         0.0191       12.092   0.000
  Slope [G.sub.10]         -0.0107         0.0035       -3.099   0.002
  Condition [G.sub.11]      0.0012         0.0036        0.342   0.732
  Gender [G.sub.12]         0.0016         0.0036        0.444   0.656

Used condoms
  Intercept [G.sub.00]      0.4835         0.0439       11.016   0.000
  Slope [G.sub.10]         -0.0142         0.0105       -1.343   0.180
  Condition [G.sub.11]      0.0014         0.0109        0.127   0.899
  Gender [G.sub.12]         0.0141         0.0110        1.290   0.197

Drank alcohol
  Intercept [G.sub.00]      0.3917         0.0197       19.835   0.000
  Slope [G.sub.10]         -0.0081         0.0037       -2.186   0.029
  Condition [G.sub.11]     -0.0034         0.0041       -0.828   0.408
  Gender [G.sub.12]        -0.0062         0.0041       -1.526   0.127

Smoked cigarettes
  Intercept [G.sub.00]      0.3379         0.0316       10.703   0.000
  Slope [G.sub.10]         -0.0091         0.0039       -2.349   0.019
  Condition [G.sub.11]      0.0055         0.0045        1.220   0.223
  Gender [G.sub.12]         0.0027         0.0045        0.614   0.539

Ate fruit
  Intercept [G.sub.00]      0.3964         0.0238       16.652   0.000
  Slope [G.sub.10]          0.0019         0.0049        0.379   0.705
  Condition [G.sub.11]      0.0103         0.0056        1.859   0.063
  Gender [G.sub.12]        -0.0024         0.0056       -0.430   0.667

Ate vegetables
  Intercept [G.sub.00]      0.5322         0.0243       21.859   0.000
  Slope [G.sub.10]          0.0032         0.0043        0.742   0.458
  Condition [G.sub.11]      0.0014         0.0049        0.297   0.767
  Gender [G.sub.12]         0.0021         0.0049        0.429   0.668

Note. Ns for all analyses = 173, except condom use, which was based on
only those who reported vaginal sex in more than one week (n = 116).

DISCUSSION

Researchers have relied on a variety of methods, such as daily diaries, to assess sexual behaviors, experiences, and cognitive states at the time of, or close to, their actual occurrence. These methods are purported to be more ecologically valid than retrospective interviews or questionnaires because the assessment occurs in the natural environment close in time to the actual experience, and so are less prone to errors of recall that are likely to affect retrospection (Shiffman & Stone, 1998). However, researchers also are concerned that, relative to other data collection methods, frequent recordings may be more reactive. This study examined whether there were trends over time in daily reports of sexual intercourse, condom use, drinking, smoking, vegetable consumption, and fruit consumption, and whether the method of data collection or gender of the respondent influenced these trends. The trends were explored for evidence of two types of reactivity: socially desirable responding and fatigue effects.

The results indicate that there was a statistically significant decline in reports of sexual intercourse over time, while reports of condom use remained stable over time. There were also significant downward trends in reports of drinking and smoking cigarettes. In all cases the slopes were relatively small; however, the percent changes from the first to the last week were more substantial. Conversely, reports of the socially desirable behaviors, vegetable and fruit consumption, increased slightly, but not significantly, over time.

For each of the six behaviors, there was significant individual variance about the slopes to allow us to test whether the method of data collection or participant gender was related to the slopes. Neither the method of data collection (self-administered or telephone-administered daily data collection) nor the gender of the participants was significantly related to the trends over time in reports of the six behaviors. The latter finding is encouraging because reports of sexual behavior, in particular, might be expected to be influenced by gender. Moreover, these findings suggest that neither method of daily data collection is more likely than the other to produce socially desirably responding or fatigue effects; thus, other considerations should determine which method is better (see Hoppe et al., 2000, for a cost-benefit analysis of these two methods).

Taken together, the trends over time in reporting these behaviors are not consistent with a fatigue effects explanation; fatigue should have affected reports of all behaviors in the direction of decreased reporting over time, not just the socially undesirable ones. Thus, daily data collection on sexual and other behaviors over a 2-month period does not appear to produce fatigue effects. Although there was a decrease in the number of occasions of sexual intercourse reported over time, this may be the result of our sampling criteria. All were selected on the basis of being currently sexually active and thus had partners at the outset of the study. For most participants, then, a change in relationships would mean the dissolution of a current relationship. While new relationships may have developed, the overall effect would be a decline in opportunities for sexual intercourse and, hence, a decline in the number of occasions of sexual intercourse. Reports of condom use, on the other hand, which is a socially desirable behavior, did not show evidence of any trend over time. These findings thus suggest that frequent collection of information about condom use is not likely to lead to fatigue or socially desirable responding effects.

With regard to smoking and drinking, however, the data suggest that daily data collection may produce either genuine behavior change (i.e., serve as an unintended intervention) in the direction of healthier behaviors, or are reactive in producing fatigue effects or socially desirable responding (behavior does not change, but reports do). Fatigue effects seem unlikely given that they were not evident across all behaviors. A study of self-reported medication adherence that compared daily diaries with an electronic monitoring device found that self-reported compliance rates were higher than monitored compliance rates (Berg et al., 1998). Similarly, a study using the bogus pipeline paradigm, in which food intake was reported by unsuccessful dieters in 1-week diaries, found that those who were told that their diaries were being verified reported a significantly greater intake of food than those who were not told this (Mulheim et al., 1998). Both studies show evidence of socially desirable responding rather than true behavior change. If socially desirable responding can account for these results, we have to question why reports of fruit and vegetable consumption also did not increase significantly over time. One possibility may be that the message about cigarette smoking and binge drinking among college students has been heard. Faced with evidence of engaging in such behaviors through daily data collection, the respondents (college students) may have been attempting to intentionally alter their behaviors.

Kubey, Larson, and Csikzentmihalyi (1996) have argued that the Experience Sampling Method (ESM), in which study participants are "beeped" at random times throughout the day to report on their behaviors, is less reactive and produces more accurate reporting than diary methods. They argue that because participants do not know in advance exactly when data will be collected, they are less likely to experience evaluation apprehension or to change their behaviors just before data are collected. Participants are more apt to recall immediate events or experiences (e.g., a mood state) more accurately. This method appears especially promising for recalling subjective experiences like mood states, that may be easily forgotten or reinterpreted after the fact, or for sampling behaviors, like smoking, that are likely to occur throughout the day. However, ESM is more intrusive than diary or prescheduled telephone interviews and therefore would not work well for persons whose work or school schedules do not permit such frequent interruptions, or for whom privacy might be a concern. Moreover, the method is less useful for behaviors like sexual activities that do not typically occur frequently throughout the day or even daily; for behaviors that occur at fairly regular times each day, like eating; or for behaviors that vary systematically with day of week (e.g., binge drinking among college students is more likely to occur on weekends).

Some researchers raise concerns that respondents using diary methods may fill out more than 1 day's dairy at a time, thus defeating the purpose and possibly introducing greater error of recall. We observed this in a previous pilot study of diary methods that we conducted. In that study, participants were instructed to fill out a diary each day, but then mail them to us only weekly. Indeed, some respondents reported afterwards that they had filled out more than 1 day's diary at a time occasionally. Telephone interviews clearly remove this possibility, but are a more expensive mode of daily data collection (see Hoppe et al., 2000). In our current study, we instructed respondents to return their completed diaries each day in pre-addressed and stamped envelopes we provided. Although this increases respondent burden somewhat, respondents appeared to comply (e.g., they returned each day's form in separate envelopes). Most importantly, the method of daily data collection (diaries or telephone interview) was not related to trends over time for any of the six behaviors, suggesting that returning diaries each day may have effectively reduced the problem of filling out several days' diaries at one time.

There are study limitations that should be mentioned. First, we do not know to what extent our findings might apply to other populations. Although random sampling was employed, which represents an improvement over many studies sampling college students, university students are not representative of the U.S. population. Moreover, because the primary goal of the study was to examine the relationship between drinking and condom use, it was necessary to screen out very low frequency and non-drinkers, those who were never or rarely sexually active, and those in long-term monogamous relationships. Second, the study's primary purpose was substantive and not designed to further elucidate the measurement factors. The random assignment to data collection conditions allowed us to make inferences about the relative utility of these methods (see Leigh et al., 1998, and Morrison, Leigh, & Gillmore, 1999, for related studies); however, there are other data collection techniques, such as the ESM, that would be useful to compare as well.

Choosing a method of data collection that reduces both random and nonrandom (systematic) sources of error should be the goal of any study of sexual behavior. The choice of method will depend in part on the type of constructs to be measured, subject burden, and project resources. There is evidence that self-administered computerized questionnaires may increase reporting of socially sensitive behaviors like sex and illicit drug use, presumably because of the greater privacy they afford (Gribble, Miller, Rogers, & Turner, 1999). The higher costs associated with this method of data collection (e.g., hardware and programming) are partially offset by elimination of data entry costs. However, this method may not be feasible when it is necessary to collect data on a daily basis over a period of time. Fortenberry (in Bancroft, 1997, p. 252) suggests that retrospective interviews or questionnaires in which respondents must sum their behaviors over a period of time are inadequate methods for understanding event-related phenomena. Indeed, some studies suggest that event-specific variables, such as drinking with sex, will be more accurately reported in diaries than with retrospective reporting (Fortenberry, Cecil, Zimet, & Orr, 1997; Graham & Bancroft, 1997). Diaries are thought to increase respondent burden, but our results show no evidence of fatigue effects over an 8-week period of daily data collection. However, as Graham (cited in Bancroft, 1997) points out, the number of variables being measured will makes a difference in the extent to which fatigue effects are likely to occur. Our study required only about 5 minutes a day to respond to the questions, but this was adequate to obtain information on nine variables, some of which included branching questions.

Given the cost of frequent data collection, the increased demands on respondents, and the possibility that it may be reactive for some sensitive behaviors, the question that must be addressed is whether the greater cost associated with daily (or frequent) data collection is worth it. In 1996, the Kinsey Institute hosted an international conference on methods used to study sexual behavior, proceedings from which were published by Bancroft (1997). The discussants concluded that, despite these drawbacks, event-related phenomena are probably better studied with diary methods than with retrospective interviews or questionnaires. However, they also noted that researchers could do a much better job of contextualizing behaviors in retrospective data collection. For example, rather than asking global questions like how frequently one has sex, researchers instead could ask what respondents do sexually and with whom they do it. Such contextualizing may not produce data as refined as that obtained through diaries, but it may improve precision of estimates based on retrospection without entailing substantially greater cost. Whether retrospection is more or less reactive than daily diaries remains to be seen. With some types of event-related phenomena, there is no viable alternative to frequent data collection. Future research determining how to reduce reactivity that can, but does not necessarily, occur with frequent data collection would be a valuable asset to our understanding and choice of methods when studying sexual behaviors.

REFERENCES

Abueg, F. R., Colletti, G., & Kopel, S. A. (1985). A study of reactivity: The effects of increased relevance and saliency of self-monitored smoking through enhanced carbon monoxide feedback. Cognitive Therapy and Research, 9, 321-333.

Bancroft, J. (1997). Researching sexual behavior. Bloomington, IN: Indiana University Press.

Berg, J., Dunbar, J. J., & Rohay, J. M. (1998). Compliance with inhaled medications: The relationship between diary and electronic monitor. Annals of Behavioral Medicine, 20, 36-38.

Bryk, A. S., & Raudenbush, S. W. (1992). Hierarchical linear models: Applications and data analysis methods. Newbury Park, CA: Sage Publications.

Campbell, J. D., Chew, B., & Scratchley, L. S. (1991). Cognitive and emotional reactions to daily events: The effects of self-esteem and self-complexity. Journal of Personality, 59, 473-505.

Carney, M. A., Tennen, H., Affleck, G., del-Boca, F. K., & Kranzler, H. R. (1998). Levels and patterns of alcohol consumption using timeline follow-back, daily diaries and real-time "electronic interviews." Journal of Studies on Alcohol, 59, 447-454.

Cronin, C. (1996). Harm reduction for alcohol-use-related problems among college students. Substance Use and Misuse, 31, 2029-2037.

Cruise, C. E., Broderick, J., Porter, L., Kaell, A., & Stone, A. (1996). Reactive effects of diary self-assessment in chronic pain patients. Pain, 67, 253-258.

Edelmann, R. J., & Connolly, K. J. (1998). Psychological state and psychological strain in relation to infertility. Journal of Community and Applied Social Psychology, 8, 303-31 I.

Fortenberry, J. D., Cecil, H., Zimet, G. D., & Orr, D. P. (1997). Concordance between self-report questionnaires and coital diaries for sexual behaviors of adolescent women with sexually transmitted infections. In J. Bancroft (Ed.), Researching sexual behavior (pp. 237-249). Bloomington, IN: Indiana University Press.

Fortenberry, J. D., Orr, D. P., Zimet, G. D., & Blythe, M. J. (1997). Weekly and seasonal variation in sexual behaviors among adolescent women with sexually transmitted diseases. Journal of Adolescent Health, 20, 420-425.

Fremouw, W. J., & Brown, Jr., J. P. (1980). The reactivity of addictive behaviors to self-monitoring: A functional analysis. Addictive Behaviors, 5, 209-217.

Gerstel, E. K., Harford, T. C., & Pautler, C. (1980). The reliability of drinking estimates obtained with two data collection methods. Journal of Studies on Alcohol, 41, 89-94.

Gold, R. S., & Rosenthal, D. A. (1995). Preventing unprotected anal intercourse in gay men: A comparison of two intervention techniques. International Journal of STD & AIDS, 6, 89-94.

Graham, C., & Bancroft, J. (1997). A comparison of retrospective interview assessment versus daily ratings of sexual interest and activity in women. In J. Bancroft (Ed.), Researching sexual behavior (pp. 227-236). Bloomington, IN: Indiana University Press.

Gribble, J. N., Miller, H. G., Rogers, S. M., & Turner, C. F. (1999). Interview mode and measurement of sexual behaviors: Methodological issues. The Journal of Sex Research 36, 16-24.

Hoppe, M. J., Gillmore, M. R., Valadez, D. L., Civic, D., Hartway, J., 8,: Morrison, D. M. (2000). The relative costs and benefits of telephone interviews vs. self-administered diaries for daily data collection. Evaluation Review, 24, 102-116.

Kazdin, A. E. (1982). Observer effects: Reactivity of direct observation. New Directions for Methodology of Social & Behavioral Science, 14, 5-19.

Kirby, K. C., Fowler, S. A., & Baer, D. M. (1991). Reactivity in self-recording: Obtrusiveness of recording procedure and peer comments. Journal of Applied Behavior Analysis, 24, 487-498.

Kopp, J. (1988). Self-monitoring: A literature review of research and practice. Social Work Research and Abstracts, 24, 8-20.

Kubey, R., Larson, R., & Csikzentmihalyi, M. (1996). Experience sampling method applications to communication research questions. Journal of Communication, 46, 99-118.

Leigh, B. C. (1993). Alcohol consumption and sexual activity as reported with a diary technique. Journal of Abnormal Psychology, 102, 490-493.

Leigh, B. C., Gillmore, M. R., 8,: Morrison, D. M. (1998). Comparison of diary and retrospective measures for recording alcohol consumption and sexual activity. Journal of Clinical Epidemiology, 51, 119-127.

Lemmens, P., Knibbe, R. A., & Tan, E (1988). Weekly recall and diary estimates of alcohol consumption in a general population survey. Journal of on Alcohol, 49, 13 l- 135.

Morrison, D. M., Leigh, B. C., & Gillmore, M. R. (1999). Daily data collection: A comparison of three methods. The Journal of Sex Research, 36, 76-81.

Moss, R. A., Prue, D. M., Lomax, B. D., 8,: Martin, J. E. (1982). Implications of self-monitoring for smoking treatment: Effects on adherence and session attendance. Addictive Behaviors, 7, 381-385.

Mulheim, L. S., Allison, D. B., Heshka, S., & Heymsfield, S. B. (1998). Do unsuccessful dieters intentionally underreport food intake? International Journal of Eating Disorders, 24, 259-266.

Persky, H., Strauss, D., Lief, H. I., Miller, W. R., & O'Brien, C. P. (1981). Effect of the research process on human sexual behavior. Journal of Psychiatric Research, 16, 41-52.

Pope, S. T., & Jones, R. S. P. (1996). The therapeutic effect of reactive self-monitoring on the reduction of inappropriate social and stereotypic behaviours. British Journal of Clinical Psychology, 35, 585-594.

Rabow, J., & Duncan, M. S. (1995). Drinking among college students. Journal of Alcohol and Drug Education, 40, 52-64.

Samo, J. A., Tucker, J. A., & Vuchinich, R. E. (1989). Agreement between self-monitoring, recall, and collateral observation measures of alcohol consumption in older adults. Behavioral Assessment, 11, 391-409.

Searles, J. S., Perrine, M. W., Mundt, J. C., & Helzer, J. E. (1995). Self-report of drinking using touch-tone telephone: Extending the limits of reliable daily contact. Journal of Studies on Alcohol, 56, 375-382.

Shiffman, S., & Stone, A. A. (1998). Ecological momentary assessment: A new tool for behavioral medicine research. In D. S. Krantz & A. Baum (Eds.), Technology and methods in behavioral medicine (pp. 117-131). Mahwah, NJ: Lawrence Erlbaum Associates.

Singh, N. N., & Leung, J. P. (1988). Smoking cessation through cigarette-fading, self-recording, and contracting: Treatment, maintenance and long-term follow-up. Addictive Behaviors, 13, 101-105.

Sobell, M. B., Bogardis, J., Schuller, R., Leo, G. I., & Sobell, L. C. (1998). Is self-monitoring of alcohol consumption reactive? Behavioral Assessment, 11, 447-458.

Stone, A. A., Kessler, R. C., & Haythornthwaite, J. A. (1991). Measuring daily events and experiences: Decisions for the researcher. Journal of Personality, 59, 575-607.

Thomsson, H. (1997). Women's smoking behaviour--Caught by a cigarette diary. Health Education Research, 12, 237-245.

Verbrugge, L. M. (1980). Health diaries. Medical Care, 18, 73-95.

Vittengl J. R., & Holt, C. S. (1998). Positive and negative affect in social interactions as a function of partner familiarity, quality of communication, and social anxiety. Journal of Social & Clinical Psychology, 17, 196-208.

Webb, G. R., Redman, S. S., Sanson-Fisher, R. W., & Gibberd, R. W. (1990). Comparison of a quantity-frequency method and a diary method of measuring alcohol consumption. Journal of Studies on Alcohol, 51, 271-277.

Manuscript accepted October 24, 2000

This work was supported by a grant from the National Institute on Alcohol Abuse and Alcoholism (#AA09701). The authors wish to thank W. W. Washburn, Executive Director of Admissions and Records at the University of Washington; the interviewers; and the student participants for their invaluable assistance with this study.

Address correspondence to Mary R. Gillmore, School of Social Work, University of Washington, 4101 15th Avenue NE, Seattle, WA 98105-6299; e-mail: [e-mail].