Recalling sexual behavior: a methodological analysis of memory recall bias via interview using the diary as the gold standard.

Retrospective self-reported assessments of sexual behavior are a mainstay of research on human sexuality. Systematic bias in self-reports of sexual behavior is generally associated with two factors: self-disclosure and recall bias. Research on self-disclosure bias (i.e., deliberate misrepresentation of the truth about one's self) in sex research has been reviewed previously (Catania, 1999). Researchers have less often studied problems of recall bias associated with retrospective reports of sexual behavior (Catania et al., 1993; Croyle & Loftus, 1993). One reason for this is the lack of a recall bias assessment method that might be applied to behaviors that, except for rare circumstances, are private and unavailable to investigators for direct study.

To a large extent, surveys of sexual behavior require respondents to tap episodic memories of prior sexual experiences. Recall of episodic memories is subject to errors caused by telescoping (i.e., assigning events to the recall period that occurred prior to or after the recall period) and forgetting (Croyle & Loftus, 1993). Daily diaries provide a means of reducing these errors, because daily diaries can be used to record behaviors occurring on the same day (or within 24 hours).

In an earlier attempt to assess recall bias in sexual assessments, Berk, Abramson, and Okami (1995) used a diary-interview method to study recall error in college students' self-reports of sexual behavior. The design required respondents to keep daily diaries of their sexual behavior for 2 weeks. The diary phase was then followed by a time interval (2 weeks), and respondents were subsequently interviewed and asked to recall their sexual behavior during the time period covered by the diaries (recall interview).

This "diary-retrospective interview" design assumes that under the appropriate conditions, daily diaries are capable of providing accurate counts of sexual behavior. Diaries have been found to produce improvements in accuracy in general health surveys, have relatively high completion rates even over several months of recording time, and be useful with respondents with little formal education (Sudman & Lannom, 1980; Verbrugge & Depner, 1980). Compared to face-to-face interviews, research has shown diaries to produce better detailed responses to health questions, greater accuracy, and more complete and detailed information (Sudman & Lannom, 1980).

Although the use of diaries in the diary-recall model helps address the problem of "near real-time" assessments of sexual behavior, the diaries and the subsequent recall interview may still be contaminated by self-disclosure bias, since both are essentially self-report measures and therefore subject to deliberate misreporting of events. Particularly problematic would be the case in which the diary and interview phases produce different effects related to self-disclosure bias (e.g., over vs. underreporting). Researchers might use various strategies to address self-disclosure bias in the diary and retrospective interviews. In the Berk et al. (1995) study, the investigators used a self-administered questionnaire (SAQ) in the recall interview that might have reduced self-disclosure bias, but the diary phase could still be contaminated. Further, because respondents in this study might have believed the investigators would see the questionnaire and therefore might know who they were, the SAQ may not have significantly reduced self-disclosure bias. Thus, differences in observed frequency counts between the diary and recall-interview in the Berk et al. study (see below) may be related to recall bias, disclosure bias, or both.

Another approach to dealing with the disclosure bias problem would be to directly assess self-disclosure biases and control statistically or in the design of the study for such bias. In the present study, we sought to reduce selfdisclosure bias through preselection of participants who were very forthcoming and open about discussing sexual matters as measured on a standardized sexual self-disclosure instrument (Catania, 1999).

Despite limitations, Berk et al.'s (1995) study suggests that the diary-interview model may be useful for studying recall error in the sexual interview. They provided evidence showing that respondents overreported vaginal intercourse and oral sex in the recall interview versus diary. Conversely, another study, also using a diary-retrospective interview design, found few diary-interview differences; however, the retrospective interview in that study followed immediately after the 4-month diary phase, and the brevity of the recall period may have substantially attenuated recall errors (Leigh, Gillmore, & Morrison, 1998). Fortenberry, Cecil, Zimet, and On" (1997) also used a diary-proximal recall interview, albeit with a longer diary phase, and found significant differences between diary and recall interview on coital events and events involving substance use but not on numbers of sexual partners or condom use. In this latter study, "memory errors" may have occurred because the longer diary phase increased memory burden even with a proximal recall interview. Although past research suggests that there may be merit in considering the diary-interview model as a tool for studying recall error in sexological interviews, prior studies do not offer clear validation that in fact previously observed diary-interview differences are due to memory error. (1)

Berk et al. (1995) made a valiant attempt to validate the diary-retrospective interview model of studying error in self-reported sexual behavior by attempting to manipulate respondents' ability to recall past events through a cognitive survey recall enhancement technique. This technique did not, however, have any effect on recall. Consequently, they concluded that the observed discrepancies between diary and interview might not have been the result of memory errors. Thus, it remains to be shown if the diary-interview model is a valid model for testing recall errors that may occur with respect to retrospective reports of sexual behavior.

A less complicated test of the diary-retrospective interview model might be to examine the difference between diary and recall reports as a function of the time interval between the diary and recall interview phases. The present study, based on the diary-interview model, examined recall error for three different such intervals (1, 2, and 3 months respectively). Participants who completed the diary were randomly allocated to one of three subgroups corresponding to the three different intervals. We hypothesized that recall error would increase with the length of the recall period and expected that our data would provide a test of the fundamental research model for studying recall problems in self-reports of sexual behavior. A finding of no difference in discrepancies between diary and interview across time intervals might suggest that the research design is flawed as a model of recall error.

In addition to the evaluation of recall error depending on the time lag between the diary and recall assessments of sexual behavior, we examined several hypotheses concerning variables that might be related to recall error. First, we examined whether the frequency of the behavior influences recall ability. For instance, behaviors that are performed infrequently may be cognitively more salient and, therefore, remembered more easily (for review, see Catania, Gibson, Chitwood, & Coates, 1990). In accord, prior research has found that test-retest correlations are higher for sexual behaviors that are performed less frequently (Catania et al., 1990). Second, the use of alcohol or other psychotropic substances during sexual encounters might inhibit the person's ability to recall particular events that occurred during those encounters. Further, sexual encounters that have a positive emotional valance (e.g., those that are associated with orgasm) may be more readily recalled than encounters that are more neutral or negative. Prior research suggests that self-events such as sexual behavior are more easily recalled when the events are pleasant as opposed to unpleasant or neutral (Skowronski, Betz, Thompson, & Shannon, 1991). In brief, we expected the discrepancy between the diary and the recall interview assessment to be smaller for respondents who did not use alcohol or drugs during sex, who had more positive sexual encounters, and who had less frequent sex.

METHODS

Design

Each participant completed a daily diary of his or her sexual activities for 4 consecutive weeks. To examine how recall accuracy would be influenced by length of the recall period, we randomly assigned participants to one of three groups: 1-, 2-, or 3-month recall according to the time interval between the completion of the diary phase and the interview.

Recruitment and Initial Screening

We mailed 920 recruitment letters to graduate students randomly selected from a large Midwestern university student directory, inviting them to participate in a study investigating "methods of assessing the frequency of different types of sexual behavior." Those students between 21 and 35 years of age who were interested in participating were asked to send their names and contact telephone numbers in a provided, preaddressed campus mail envelope. We then telephoned respondents and made appointments for them to come into the lab. One hundred thirty-nine students (74 females, 59 males, mean age = 27.2 years) attended the initial screening visit and completed the Sexual Self-Disclosure Scale (SSD; see Measures section). Following this, an interview established whether individuals met the following selection criteria: (a) currently in an active heterosexual relationship; (b) has had three or more occasions of sexual activity in the previous month; and (c) scored below the national median value on the SSD (chosen to screen out individuals who were low in self-disclosure).

Of the 139 students who were screened, 64 (33 females, 31 males) were ineligible for the following reasons: no active current heterosexual relationship (32); current sexual partner not living in town (14); sexual activity occurred on fewer than three occasions during the previous month (8); moving out of town in the next 4 months (9); SSD scores above the required cut-off score (1). After screening, 75 participants (44 females, 31 males) were entered into the study.

Measures

Sexual Self-Disclosure Scale (SSD). The SSD (Catania, McDermott, & Pollack, 1986; Catania, 1999) is a self-report 7-item Likert-rated scale that measures the degree of discomfort associated with self-disclosure of sexual matters. The items assess respondents' self-reported ease or difficulty with disclosing sexuality information in different contexts and interpersonal situations. Scores are produced by summing across items; the range of possible scores is 7 to 28. The SSD has good internal (or > .80) and test-retest (.92) reliability and good construct validity, and has been normed on a nationally representative sample of adults (Catania, 1999). The SSD was used in the present study as a screening tool to control for self-presentation and disclosure bias. By controlling in the design of the study for this source of error (i.e., by recruiting only people comfortable with disclosing personal sexual information), we hoped to obtain a clearer assessment of recall accuracy on questions pertaining to sexual behavior and outcomes. Individuals whose SSD scores were above the median of the national normative range (above 22; i.e., more discomforted by sex questions) were excluded from participating in the study.

Sexual Behavior Diary. Participants completed the Sexual Behavior Diary, developed for the present study, daily for 4 consecutive weeks. Each day's ratings were made on a separate page. Participants recorded the occurrence of the following sexual activities and experiences with a partner: vaginal intercourse, oral sex, anal intercourse, mutual masturbation or genital touching, and orgasm. If either vaginal or anal intercourse occurred, condom use was assessed; if no condom was used, participants indicated whether ejaculation had occurred. To assess the number of sexual partners, participants were asked to identify each sexual partner with a single initial. If more than one occasion of sexual activity occurred during a 24-hour period, the details and partners' initials for these were also provided on the diary.

In addition to the above partnered activities, participants recorded whether or not they had masturbated and also rated the degree of sexual interest they had experienced (on a 0-5 scale, from none at all to very strong). Lastly, participants answered questions on alcohol and substance use in the 4 hours leading up to any sexual encounter. The Sexual Behavior Diary took approximately 5 minutes each day to complete.

Four-week follow-up questionnaire. This questionnaire was used to assess the degree of difficulty participants experienced in completion and return of the daily diaries. Participants gave separate ratings, on a 7-point scale from extremely easy to extremely difficult, for completing the Sexual Behavior Diary and for mailing back the diaries on a daily basis.

Follow-up interview. The retrospective interview asked about participants' sexual behavior during the 4 weeks of diary completion and during each subsequent month since the end of the diary phase (between 1-3 months, depending on group). The sexual behavior items included all of those on the daily diary. Questions were asked on a month-by-month basis, with the interviewer first establishing dates for the time period under consideration. In addition to information about the frequency of sexual behavior, interviewers obtained information on the number of sexual partners; participants identified each sexual partner by initial to permit comparison between diary and interview data. Lastly, we collected demographic information as well as relationship details, such as duration of relationship and contraceptive method(s) used.

Procedure

We obtained study approval from the Indiana University Human Subjects Committee. After providing written informed consent, participants were instructed in the use of the Sexual Behavior Diary by the researcher or one of two trained research assistants (one male and one female). We asked participants to complete the diary at a regular time each day and, once they had made their ratings, to place the diary in a campus mail envelope and return this to us within the next 24 hours. If diaries were not received on a regular basis, we contacted the participant and reminded her or him to send in the diaries on a daily basis (only 13% of respondents needed reminder calls). None of the diaries or envelopes contained individual names; all participants were assigned a study number.

After the 4-week period of daily monitoring, participants returned to the lab and completed the 4-week follow-up questionnaire, providing feedback about the degree of difficulty they had experienced in completion and return of the daily diaries. Compensation for the diary phase was paid at this visit, as follows: $1.00 for each diary completed and mailed back on a daily basis, for a total of $28.00 maximum at the end of the 4 weeks. Appointments for the final interview were not made at this visit; instead participants were only reminded that we would be contacting them to schedule their interview sometime during the next 3 months.

We randomly assigned participants to one of the three groups corresponding to the interval between the diary period and retrospective interview: Group 1 (1 month), Group 2 (2 months), or Group 3 (3 months). The interview took approximately 30 to 45 minutes to complete. Of the 69 participants who were interviewed, 61 (88.4%) were interviewed by a female researcher and 8 (11.6%) were interviewed by a male research assistant who was trained to conduct interviews. Upon completion of the interview, we paid participants an additional $20.00 and debriefed them regarding the specific purposes and hypotheses of the study.

RESULTS

Statistical Analysis

To assess the extent to which recall bias (mean recall error) differed depending on the lag time to recall (1, 2, or 3 months from diary completion to interview) and the other hypothesized predictors of recall error, we conducted a series of multiple regression analyses for each of the sexual behaviors assessed. We used a plot of the residuals (observed recall error-fitted recall bias) to check the OLS model, and this showed the residuals were reasonably centered around zero. We also ran a formal test of normality, which gave no indication that the distribution of the residuals was not normal. Except for the behaviors with the lowest frequency counts (e.g., anal intercourse), the proposed models seemed appropriate for this particular data sample.

We considered whether the high correlations between some of the sexual behavior variables (e.g., the correlation between orgasm frequency and frequency of vaginal intercourse was .75) might raise a possible multicollinearity issue. Multicollinearity occurs when linear regression model variables are highly correlated (0.90 and above) and can cause both statistical as well as logical problems. Logically, redundant variables weaken the regression analyses through reduction of degrees of freedom error. A rule of thumb would be to avoid using pairs of variables with a correlation of 0.80 or higher in the same analysis. However, though the variables in this analysis were correlated, the correlations were not high enough to warrant a multicollinearity diagnosis.

Independence of observations is violated if longitudinal data are analyzed as if they are cross-sectional data (i.e., independent observations). In this study, there were no per se violations of the independence assumptions since the data analyses were formulated in a way that created independent and identically distributed (IID) observations. For example, whereas the diary observations and corresponding post period observations by themselves (i.e., separately analyzed) would be correlated, their difference (e.g., for recall bias) creates an IID observation.

Participant Characteristics

There were 6 dropouts (3 females, 3 males); 4 of these were participants who did not complete the 4 weeks of daily diaries and did not respond to repeated attempts to make contact. The remaining 2 individuals completed the diary phase but did not attend an interview. Sixty-nine participants (41 females, 28 males) completed the study; their mean age was 27.0 years (range: 19-35 years, SD = 3.7). Sixty participants submitted complete data. Of the 9 (6 males, 3 females) with missing diary data, 8 had one missing diary and one participant had two missing diary days. The majority of participants (58; 84.1%) were childless, and 61(88.4%) were currently using some method of birth control. The average duration of their current relationships was 3.2 years (SD = 3.3 years). As expected (and designed), the sample was highly educated; the mean number of years of post-high-school education was 7.2 years (SD = 2.7). In terms of ethnicity, the majority of the sample (60; 87.0%) was White. The mean SSD score of the completers was 11.0 (SD = 2.8).

Comparison of the 6 dropouts with the participants who completed the study revealed few differences. In particular, the mean SSD score of the individuals who withdrew from the study was 11.8; thus, there was no indication that the dropouts had greater difficulty disclosing sexuality information than those who completed the study. The mean age of the dropouts was 25.3 years, slightly younger than the completers (27.0 years).

Comparison of the Three Groups

The numbers of participants in each of the 3 groups were as follows: Group 1, 24; Group 2, 23; Group 3, 22. There were no significant differences among the 3 groups in terms of gender ([chi square] = .33, p [is less than or equal to] 0.85). There were also no between-group differences in age or scores on the SSD.

Four-Week Follow-Up Questionnaire

Results from this questionnaire, which assessed the degree of difficulty participants experienced with the daily diary procedure, suggested that mailing back the diaries on a daily basis was somewhat more difficult than diary completion. On a 7-point scale from extremely easy to extremely difficult, the mean rating for completing the daily diary was 1.4 (SD = 0.72), whereas for diary return the mean rating was 2.1 (SD = 1.4).

Diary Reports of Sexual Behaviors

The mean frequencies of each of the behaviors assessed during the diary month are shown in Table 1. The frequencies for three variables--anal intercourse, condom use with anal intercourse, and drug use before vaginal intercourse--were extremely low and therefore will not be considered further. Only one participant (a woman) reported more than one sexual partner during the diary month, and a small number of participants (8) reported more than one sexual encounter per day. Using one-way ANOVA, we compared the mean diary counts for each of the sexual behaviors (i.e., vaginal intercourse, oral sex received, etc.) across the three groups. We found no significant between-group differences (see Table 1), indicating that the randomization was successful and the groups were essentially similar at the diary phase of the study on frequency of sexual activities.

Total Recall Error and Directionality of Error

Total recall error was defined as the difference between the interview value and the diary value for a particular behavior (I-D, I = interview value and D = diary value). To remove any directionality, we used the absolute value of the total recall error in all regression models. We indexed overreporting by estimates of recall error that had positive values, indicating overestimation of the behavior at interview compared to diary.

Table presents a summary of the descriptive data on recall error, including the percentage of individuals in each group who showed no discrepancy between their diary and interview counts, the mean discrepancy between diary and interview, and the percentage of individuals who overestimated or underestimated the frequency of sexual behavior at interview compared with the diary month.

We examined hypothesized correlates of total recall error in a series of multivariate regression models (see Table 2). Each model examined the effects of time lag (2 and 3 months vs. 1 month), frequency of alcohol use just prior to sex for the diary period, frequency of orgasm during the diary period, frequency of the behavior (for which recall error was being modeled) during the diary period, and the interaction between frequency of the behavior and time lag.

Looking across models, time since completing the diary phase had little impact on total recall error independent of the other independent variables, with the notable exception of vaginal intercourse. For vaginal intercourse, total recall error increased significantly at 3 months (p < 0.01), suggesting that by 3 months participants were making substantial errors in reporting frequency of vaginal intercourse compared to that observed at 1 month postdiary. Across all other outcomes examined, however, the 2- and 3-month time intervals did not produce a significant increase in total recall error, suggesting that memory is reasonably stable over these shorter time periods.

The models also tested hypotheses concerning total recall error and frequency of orgasm and alcohol use prior to sexual activity. Our expectation was that orgasm would lead to greater total recall error by producing greater overestimation. Alcohol consumption prior to sexual activity was expected to increase recall error, but we did not expect this distortion to affect directionality of the error (see following section). We also examined the more conventional hypothesis that recall error would increase with frequency of the behavior.

Alcohol was significant at the point of entry in three of the six models (more alcohol use = greater recall error), but these effects remained significant once all other variables were entered only in the orgasm model. Orgasm frequency was significant at point of entry in all five models (note in the orgasm recall error model the effects of orgasm are not separable from those of orgasm frequency), but was reduced to nonsignificance or a trend (p < .10) once behavior frequency was entered.

Behavioral frequency had significant effects on recall error (higher frequencies of the behavior or outcome associated with greater recall error) at point of entry in three of the six models, but after entering the interaction term this reduced to statistical nonsignificance in two models (condom use during vaginal intercourse and mutual masturbation) and only a significant trend for one other model (orgasm recall error).

These multivariate analyses are, however, somewhat inconclusive because of the high correlations between orgasm and other sexual behaviors (i.e., orgasm often is the result of other sexual activities such as vaginal intercourse, oral sex, and mutual masturbation). The correlation, for instance, between orgasm frequency and frequency of vaginal intercourse was .75. Furthermore, among college-age students there may be a tendency to have sex after consuming alcohol or drugs. For instance, the correlation between frequency of alcohol or substance use before sex was .52 with frequency of vaginal intercourse. These strong correlations make it difficult to separate the effects of behavioral frequency, orgasm, and alcohol or substance use on recall of sexual behavior.

We also examined the total recall error model correlates in logistic regression models designed to test directionality. We limited our testing to overreporting since our primary directional prediction was for overestimation [dependent variable = 1 if (interview - diary reports) > 0, or 0 if (interview - diary reports) [is less than or equal to] 0]. Specifically, we had hypothesized that when sexual encounters are associated with orgasm, individuals may come to overreport the events or encounters that are associated with this form of physical pleasure. In examining the zero-order correlations between orgasm and overreporting, we did in fact find significant positive correlations between the number of orgasms and overestimation of vaginal intercourse (r = 0.34, p = 0.003) and condom use during vaginal intercourse (r = 0.27, p = 0.03), but not mutual masturbation or oral sex. However, in the multivariate models orgasm was not a significant correlate of condom use recall error independent of alcohol consumption or behavioral frequency (not shown).

In the context of the broader array of correlates examined for total recall error, we observed only a small number of significant relationships to overreporting any one of the examined behaviors or outcomes [because of the small number of significant effects, we present only significant odds ratios (ORs) in text]. Alcohol use was a significant correlate of overreporting at point of entry (order of entry as in Table 2) in 3 of the 6 models (orgasm OR = 1.55, p = .01; condom use OR = 1.40, p = .05; mutual masturbation OR = 1.48, p = .01), but remained significant in only two models once all other variables were entered (orgasm frequency OR = 1.54, p = .01; mutual masturbation OR = 1.93, p = .05). In addition, greater orgasm frequency was significantly correlated with overreporting vaginal intercourse (OR = 1.18, p = .01), and frequency of condom use was significantly correlated with overreporting condom use (OR = 1.51, p = .001). In no instance were the time or Time x Behavior frequency variables significant correlates of overreporting.

DISCUSSION

This paper examined a daily diary-recall paradigm useful for studying problems related to recall of sexual behavior. To validate the basic model, we examined the degree of error in respondents' ability to recall sexual behavior recorded during the diary phase of the study in a randomized group design with groups varying by the length of the time interval between diary and recall interview (1, 2, and 3 months). For a high-frequency heterosexual behavior, vaginal intercourse, there was a significant increase in recall error by 3 months (vs. 1 month). However, many other commonly measured outcomes (condom use, oral sex, orgasm) showed relatively stable recall over the largest window of 3 months. Because of the low frequency of some behaviors, it may have been difficult to detect small increases in recall error, or, as has been observed previously, 3 months may provide a relatively reliable recall window (Kauth, St. Lawrence, & Kelly, 1991). Studies with larger samples and longer recall periods (e.g., 6 and 12 months) may be needed to show associations between time and error. An incremental study would be useful for identifying the limits of recall for specific behaviors (e.g., by 1 month increments).

We also examined several hypothesized correlates of total recall error for each behavior-related outcome (oral sex, vaginal intercourse, condom use, alcohol with sex, mutual masturbation, and orgasm). Higher frequency of the behavior or outcome was significantly associated with recall error in some (condom use, mutual masturbation; strong trend in orgasm model), but not all models (vaginal intercourse, oral sex). These inconsistencies were not expected based on past work examining test-retest correlations and diary type studies (Catania et al., 1990; Fortenberry et al., 1997; Kauth et al., 1991; Reading, 1983). Alcohol use and orgasm were also associated with total recall error and overestimates in some but not all behaviors and outcomes. Despite inconsistencies across models, the results do provide a partial validation of the diary-interview recall model as a method for studying recall error of sexual behavior. These inconsistencies are not easy to explain, however. In particular, a challenge to interpretation arises in that frequencies of sexual behavior, orgasm, and alcohol use prior to sex may often be associated with each other (as observed). Thus, it is not a simple matter to examine these alternative explanations for recall error as independent explanations. Using large population-based samples as a starting point, it may be possible to disentangle these effects, given, for example, that there exist people who may have high frequencies of some sexual behavior but low frequencies of orgasm and alcohol use.

The results also have public health significance. Vaginal intercourse and condom use, for example, are both key behaviors affecting transmission of a wide range of sexually transmitted infections (STIs). Factors that influence overreporting or total recall error (e.g., alcohol use, behavior frequency) might need to be taken into consideration in studies that examine, for instance, the effects of condom use on STI transmission or of an intervention on condom use.

Limitations of this study include the fact that the sample was limited to young adults who were predominantly White and female. In addition, low frequency counts for anal intercourse, condom use with anal intercourse, and drug use before vaginal intercourse precluded analysis of these variables. However, although this was a small, exploratory study, some aspects of the study design may have served to minimize some types of measurement error. For example, we controlled for self-presentation bias by screening volunteers on sexual self-disclosure, with the aim of obtaining a more accurate estimate of recall error. One of the objections often raised about the written daily diary method is that there is no independent verification of the time when the diary was completed, making it possible for a respondent to fill out diaries for several days retrospectively (Morrison, Leigh, & Fillmore, 1999). In the present study, participants were requested to return their daily diaries within 24 hours, and if diaries were not received on a regular basis, the participant was contacted and given a reminder to send her or his diaries in. We also instructed participants that when they forgot to complete a day's ratings, they should draw a line through that day's ratings to indicate this. In previous studies, a range of collection periods have been used (Morrison et al., 1999), but as far as we are aware, daily return of diaries has seldom been required. All of the above procedures would have served to minimize the likelihood that participants would complete diary entries retrospectively.

In brief, the findings from the present study suggest that the daily diary-recall paradigm may be a useful tool with which to explore issues related to recall bias and possible correlates of recall error. This paradigm might also be used to evaluate how different methods--such as different length response windows, number of questions, and memory aids--impact recall error. Lastly, although the use of daily diaries clearly involves an investment of time and effort on the part of the participants (and the researchers!), we found that participants rated the tasks of diary completion and return as not unduly difficult and the attrition rate was low.

Table 1. Diary Results and Interview Versus Diary Discrepancy, by
Time Lag From Diary to Interview Assessment

                                 Diary result        Interview versus
                                                     diary discrepancy

                        Mean frequency                       %
Behavior                     (SD)       p value (a)  overestimating (b)

Vaginal
  intercourse
    Group 1               6.3 (4.3)        0.08            50.0
    Group 2               9.1 (5.1)                        56.5
    Group 3               7.2 (3.5)                        63.6
Condom use w/ vaginal
  intercourse
    Group 1               2.5 (4.0)        0.90            29.2
    Group 2               3.0 (5.4)                        17.4
    Group 3               2.9 (4.2)                        18.2
Oral sex received
    Group 1               3.3 (3.1)        0.96            33.3
    Group 2               3.1 (2.7)                        52.2
    Group 3               3.3 (3.8)                        40.9
Oral sex performed
    Group 1               2.6 (3.9)        0.78            20.8
    Group 2               3.1 (2.1)                        34.8
    Group 3               3.3 (4.1)                        27.3
Mutual masturbation
    Group 1               1.5 (1.18)       0.27             8.3
    Group 2               0.78 (1.1)                       17.4
    Group 3               1.7 (2.6)                        13.6
Masturbation alone
    Group 1               4.6 (4.4)        0.89            33.3
    Group 2               4.0 (4.4)                        39.1
    Group 3               4.2 (4.4)                        27.3
Orgasm
    Group 1               6.3 (5.0)        0.50            54.2
    Group 2               6.9 (2.2)                        47.8
    Group 3               8.0 (4.3                         59.1
Alcohol prior to sexual
  activity
    Group 1               1.6 (2.1)        0.85            20.8
    Group 2               1.6 (2.2)                        30.4
    Group 3               1.3 (l.6)                        22.7

                             Interview versus diary discrepancy

                           % no              %             Estimated
Behavior                discrepancy  underestimating    recall bias (d)
                                            (c)

Vaginal
  intercourse
    Group 1                33.3            16.8            0.9 (2.5)
    Group 2                13.0            30.4            1.1 (3.1)
    Group 3                 9.1            27.3            2.5 (3.7)
Condom use w/ vaginal
  intercourse
    Group 1                62.5             8.3            1.1 (2.2)
    Group 2                69.6            13.0            0.3 (1.5)
    Group 3                68.2            13.6            0.7 (2.0)
Oral sex received
    Group 1                29.2            37.5           -0.5 (2.0)
    Group 2                26.1            21.7            1.4 (3.5)
    Group 3                22.7            36.4            0.6 (3.1)
Oral sex performed
    Group 1                45.8            33.3           -0.2 (l.4)
    Group 2                30.4            34.8            0.5 (3.3)
    Group 3                36.4            36.4           -0.2 (l.8)
Mutual masturbation
    Group 1                54.2            37.5           -0.5 (1.5)
    Group 2                60.9            21.7            0.0 (1.1)
    Group 3                40.9            45.5           -0.5 (1.3)
Masturbation alone
    Group 1                25.0            41.7            0.0 (3.3)
    Group 2                30.4            30.4            1.3 (4.4)
    Group 3                40.9            31.8           -0.05 (4.5)
Orgasm
    Group 1                29.2            16.7            1.0 (2-8)
    Group 2                26.1            26.1            1.5 (3.8)
    Group 3                18.2            22.7            2.4 (3.9)
Alcohol prior to sexual
  activity
    Group 1                62.5            16.7            0.4 (l.4)
    Group 2                56.5            13.0            1.2 (3.0)
    Group 3                54.5            22.7            0.5 (2.4)

Note. Group 1 (N = 24) = 1 mo. lag-time; Group 2 (N = 23) = 2 mo.
lag-time; Group 3 (N = 22) = 3 mo. lag-time.

(a) p value for F[2,66] test a one-way ANOVA: dependent variable =
diary frequency; independent categorical variable = group.
(b) Interview > Diary. (c) Interview < Diary. (d) Recall bias = mean
recall error (i.e., mean of I - D difference).

Table 2. Correlations of Total Recall Error: Multiple Regression Models

                                          Dependent
                                          variables

Independent                         Vaginal
variables                         intercourse
                                                            Orgasm
                         Entry [beta]    Final [beta]    Entry [beta]

Time lag
  2 months                   .65             .84             .70
  3 months                  1.50 *          2.69 *          1.28
Alcohol                      .34 **          .15             .56 ***
Orgasm                       .15 **          .11              --
Behavior
  frequency (BF)             .03             .13             .12 [dag-
                                                               ger]
BF x Time                     --              ns              --
Final model [R.sup.2]         --             .25 **           --

                                          Dependent
                                          variables

Independent                                         Condom
variables                   Orgasm                   use

                         Final [beta]    Entry [beta]    Final [beta]

Time lag
  2 months                   1.44            -.59            -.14
  3 months                   1.97            -.42            -.18
Alcohol                       .39 *           .32 **          .02
Orgasm                        --              .08 *           .01
Behavior
  frequency (BF)              .20 ([dagger])  .24 ***         .40 ***
  BF x Time                    ns              --              ns
Final model [R.sup.2]         .25 **           --             .51 ***

                                          Dependent
                                          variables

Independent                         Mutual               Oral sex on
variables                        masturbation              partner

                         Entry [beta]    Final [beta]    Entry [beta]

Time lag
  2 months                   -.26            .20            1.16 *
  3 months                    .35            .51             .26
Alcohol                       .06            .07             .16
Orgasm                        .06 *          .03             .16 ***
Behavior
  frequency (BF)              .35 ***        .48 ***         .11 ([dag-
                                                               ger])
  BF x Time                   --              ns              --
Final model [R.sup.2]         --             .41 ***          --

                                          Dependent
                                          variables

Independent              Oral sex on              Oral sex on
variables                  partner                respondent

                         Final [beta]    Entry [beta]    Final [beta]

Time lag
  2 months                   -.22        1.23 ([dagger])     .21
  3 months                   -.07            .67             .26
Alcohol                      -.01            .19             .01
Orgasm                        .15            .19 ***         .10 ([dag-
                                                               ger])
Behavior
  frequency (BF)             .04             .25 **          .10
  BF x Time                ns               --                ns
Final model [R.sup.2]        .36 ***        --               .37 ***

Note. Dependent variable = interview - diary count|; Orgasm = orgasm
experienced during a sexual encounter; Condom use = during vaginal
intercourse; Alcohol = alcohol during sexual encounters; Behavior
frequency (BF) = frequency of behavior over diary period; time lag
reference = 1 month; BF x Time = interaction between time lag and
behavioral frequency.

* p = .05. ** p = .01. *** p = .001. ([dagger]) p < .10.

U.S. POSTAL SERVICE STATEMENT OF OWNERSHIP, MANAGEMENT, AND CIRCULATION
 (Required by 39 US.C. 3685)

1. Title of publication: The Journal of Sex Research
2. Publication number: 0022-4499
3. Filing date: September 26, 2003
4. Issue frequency: Quarterly
5. No. of issues published annually: 4
6. Annual subscription rate: $95.00
7. Complete mailing address of known office of publication: 1248
Warwick P1ace, Allentown, PA 18104.
8. Complete mailing address of headquarters or general business office
of publisher: 1248 Warwick Place, Allentown, PA 18104.
9. Full names and complete mailing addresses of publisher, editor, and
managing editor: Publisher--The Society for the Scientific Study of
Sexuality, 1248 Warwick Place, Allentown, PA 18104; Editor--John
DeLamater, Ph.D., Sociology Dept., University of Wisconsin-Madison:
Madison, WI 53706; Managing Editor--Kelly Mella,
University of Wisconsin-Madison, Madison, WI 53706.
10. Owner: The Journal of Sex Research & The Society for the
Scientific Study of Sexuality; Attn. Dave Fleming, 1248 Warwick Place
Allentown, PA 18104.
11. Known bondholders, mortgagees, and other security holders: None
12. Nonprofit organization: Status has not changed during preceding 12
months
13. Publication name: The Journal of Sex Research
14. Issue date for circulation data: 4th issue
15. The following circulation figures are provided for (a) average
number of copies of each issue during the preceding 12 months and (b)
actual number of copies of single issue published nearest to filing
date:

Total number of copies printed      (a)    1,698    (b)    1,685

Paid circulation
  Sales through dealers,
    carriers, etc.                  (a)       85    (b)       84
  Mail subscriptions                (a)    1,401    (b)    1,415
Total paid and/or requested
  circulation                       (a)    1,486    (b)    1,499
Free distribution by mail           (a)      --     (b)      --
Free distribution outside the
  mail                              (a)      --     (b)      --
Total free distribution             (a)      --     (b)      --
Total distribution                  (a)    1,486    (b)    1,499
Copies not distributed
  Office use, leftovers, spoiled    (a)      213    (b)      186
  Return from news agents           (a)      --     (b)      --
Total                               (a)    1,698    (b)    1,685
Percent paid and/or requested
  circulation                       (a)      100    (b)      100

16. This statement of ownership will be printed in the 4th issue of
this publication.
17. I certify that all information furnished on this form is true and
complete. (Signed) Dave Fleming, Executive Director, October 1, 2003.

(1) Other studies have employed the diary-interview design but were not designed to address the central recall error problems; for example, they used a different time period for the two assessments (McLaws, Oldenburg, Ross. & Cooper, 1990). or used aggregate level analyses (Hornsby & Wilcox, 1989), which limits conclusions regarding recall bias.

REFERENCES

Berk, R., Abramson, P. R., & Okami, E (1995). Sexual activities as told in surveys. In P. R. Abramson & S. O. Pinkerton (Eds.), Sexual nature, sexual culture (pp. 371-386). Chicago: University of Chicago Press.

Catania, J. A. (1999). A framework for conceptualizing reporting bias and its antecedents in interviews assessing human sexuality. The Journal of Sex Research, 36, 25-28.

Catania, J. A., Gibson, D. R., Chitwood, D. D., & Coates, T. J. (1990). Methodological problems in AIDS behavioral research: Influences on measurement error and participation bias in studies of sexual behavior. Psychological Bulletin, 108, 339-362.

Catania, J. A., McDermott, L. J., & Pollack, L. M. (1986). Questionnaire response bias and face-to-face interview sample bias in sexuality research. The Journal of Sex Research, 22, 52-72.

Catania, J. A., Turner, H., Pierce, R. C., Golden, E., Stocking, C., Binson, D., et al. (1993). Response bias in surveys of AIDS-related sexual behavior. In D. Ostrow & R. Kessler (Eds.), Methodological issues in AIDS behavioral research (pp. 133-162). New York: Plenum.

Croyle, R., & Loftus, E. (1993). Recollection in the kingdom of AIDS. In D. Ostrow & R. Kessler (Eds.), Methodological issues in AIDS behavioral research (pp. 163-180). New York: Plenum.

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

Hornsby, P. P., & Wilcox, A. J. (1989). Validity of questionnaire information on frequency of coitus. American Journal of Epidemiology, 130, 94-99.

Kauth, M. R., St. Lawrence, J. S., & Kelly, J. A. (1991). Reliability of retrospective assessments of sexual HIV risk behavior: A comparison of biweekly, three-month, and twelve-month reports. AIDS Education and Prevention, 3, 207-214.

Leigh, B. C., Gillmore, M. R., & Morrison, D. M. (1998). Comparison of diary and retrospective measures for recording alcohol consumption and sexual activity. Journal of Clinical Epidemiology, 51, 199-127.

McLaws, M.- L, Oldenburgh, B., Ross, M. W., & Cooper, D. A. (1990). Sexual behavior in AIDS-related research: Reliability and validity of recall and diary measures. The Journal of Sex Research, 27, 265-281.

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.

Reading, A. E. (1983). A comparison of the accuracy and reactivity of methods of monitoring mate sexual behavior. Journal of Behavioral Assessment, 5, 11-23.

Skowronski, J. J., Betz, A. L., Thompson, C. P, & Shannon, L. (1991). Social memory in everyday life: The recall of self-events and other-events. Journal of Personality and Social Psychology, 60, 831-843.

Sudman, S., & Lannom, L. (1980). Health care surveys using diaries (DHHS Publication No. PHS 80-3279). Washington, DC: National Center for Health Survey Research.

Verbrugge, L., & Depner, C. (1980). Methodological analyses of Detroit health diaries (DHHS Publication No. PHS 80-3279). Washington, DC: National Center for Health Survey Research.

Manuscript accepted May 20, 2003

Cynthia A. Graham

Indiana University

Joseph A. Catania, Richard Brand, Tu Duong, and Jesse A. Canchola

University of California, San Francisco

This research was supported by NIMH Grants MH54320-06AI, MH51523, and MH52022 to Dr. Joseph Catania. Drs. Catania, Brand, Duong, and Canchola are all in the Center for AIDS Prevention Studies, Department of Medicine and Health Survey Research Unit, UCSE We thank Nicole Flory and Jason Hatton for assistance with data collection.

Address correspondence to Dr. Cynthia Graham, The Kinsey Institute for Research in Sex, Gender, and Reproduction, Morrison Hall 313, Indiana University. Bloomington, IN 47405; e-mail: [e-mail].