Measuring Change in Sexual Behavior: Concordance Between Survey Measures.

The primary objective of AIDS/STD prevention programs and adolescent pregnancy prevention programs is to change people's sexual behavior. To evaluate whether such programs are successful, a reliable method of measuring how and/or if sexual behavior changes over time is essential. In this paper we investigate agreement between different methods for measuring change in sexual behavior using survey research. Several researchers have looked at the concordance between different survey measures of sexual behavior per se (Coates et al., 1988; Hornsby & Wilcox, 1989; Rotheram-Borus et al., 1994; for reviews, see Catania et al., 1995; Hospers & Kok, 1995); but few have looked at concordance between different measures of behavior change. Measuring change is the methodological question addressed here. To clarify this distinction, a question measuring sexual behavior would ask, "How often do you use condoms?" whereas a question measuring sexual behavior change would ask "Have you changed how often you use condoms?"

In evaluating AIDS and STD prevention programs, different methods have been used to measure sexual behavior change. The most commonly used method, the retrospective change method, asks respondents a question such as "Have you made any changes in your sexual behavior in the past [period of time], and if so, what?" This method is typically employed in studies with a cross-sectional design. Another method, the longitudinal or panel method, asks respondents at two different times about their sexual behavior and compares the two sets of answers. This method is generally used in studies with a prospective design. Studies that use both methods are rare, but these studies provide an opportunity to test the reliability of the two methods of measurement. Comparing respondents' answers on both types of measures can provide concordance data.

Retrospective change measures have been widely used in program assessment. In a review of AIDS prevention studies, Becket and Joseph (1988) cite seven studies using such measures. Several studies designed to assess the effectiveness of various interventions have been published since their review (Baker, Morrison, Gillmore, & Schock, 1995; Baldwin, Whiteley, & Baldwin, 1990; Cochran, Keidan, & Kalechstein, 1990; Keeter & Bradford, 1988; Hingson, Strunin, & Craven, 1989). The Chicago Sex Survey (Laumann, Gagnon, Michael, & Michaels, 1994), which received a substantial amount of media attention in 1994, asked respondents the following retrospective question to assess whether AIDS had any impact on their sexual behavior: "Have you made any changes in your sexual behavior because of AIDS? What have you changed?"

Longitudinal research is often expensive and difficult to carry out. Many prevention programs either do not have the resources available to do a longitudinal assessment or did not plan to do evaluation research when the program started. Even in longitudinal studies, survey researchers may realize in later waves that they want to assess change on a variable that was not asked about in earlier waves. The retrospective change method, therefore, is often the only method available for measuring behavior change. Thus, it is important to examine the concordance between the retrospective change method and other methods of measuring behavior change, such as the longitudinal/panel method.

There are inherent difficulties in measuring sexual behavior accurately, because it generally cannot be observed directly and biases can enter into people's self-reports of their own sexual behavior. Some people are embarrassed about sex or simply feel it is too personal to report on a survey, and therefore may underreport their sexual behavior. Others may be motivated to exaggerate their behavior when reporting on it. Prior research on measurement error in assessments of sexual behavior (item refusal, over- and underreporting, and test-retest reliability) has examined, to some extent, how respondents, instruments, interviewers, and mode of data collection influence measurement error/bias (Catania, Binson, Canchola, & Pollack, 1996; Catania, Binson, Van der Straten, & Stone, 1995; Catania, Gibson, Chitwood, & Coates, 1990; Catania, Gibson, Matin, Coates, & Greenblatt, 1990). For instance, item wording, gender of interviewer, respondent control, and interview mode (telephone vs. face-to-face) have all been found to have some level of impact on response bias to key behavioral assessments in sexological surveys (Catania et al., 1995; Catania et al., 1996; Catania, Gibson, Chitwood, & Coates, 1990; Catania, Gibson, Matin, & Coates, 1990). Where researchers have been able to estimate the bias in measures of sexual behavior, the biases are almost always in the direction that the respondents perceive to be socially desirable.

This past research has generally focused on methodological problems in obtaining point estimates of behavior (although prior investigations on reliability issues touch measurement of behavior change). Methodological issues in assessing changes in sexual behavior over time are key to understanding how populations are altering their sexual behavior in response to HIV/STD prevention programs, other types of sexual health initiatives, and societal and technological change (e.g., Viagra). Any method for measuring change in sexual behavior will have some bias. We have no objective data to test various methods against and, therefore, no way to assess absolute validity of change measures. Social desirability biases are likely to enter into measures of change in sexual behavior as well. Both the retrospective method and the longitudinal/panel method are subject to the problem of social desirability biases.

Retrospective change methods may have some additional sources of bias. In memory studies, when people are explicitly asked "How has your behavior changed?" they often exaggerate the degree to which they have changed compared to objective data when they hold an implicit theory about whether a certain personal attribute is likely to change (Ross, 1989). For example, if people hold implicit theories that their sexual behavior should be changing as a result of the AIDS epidemic or participation in a certain program, or think they will be viewed favorably for making such changes, retrospective self-reports of risk reduction may be inflated. Implicit theories about what kinds of sexual behavior would be likely to change in response to the AIDS epidemic could affect respondents' memory of how their own behavior actually changed. This kind of bias often involves remembering earlier behavior as being worse than it actually was (Conway & Ross, 1984; Linton & Melin, 1982; Ross, 1989; Woodruff & Birren, 1972). Only measures which ask explicitly about changes in behavior are subject to these biases from implicit theories. Because the retrospective change method asks about change in sexual behavior directly and explicitly, it would be subject to these types of memory biases.

Both the longitudinal/panel method and the retrospective change method are subject to social desirability biases, whereas only the retrospective change measure is subject to this kind of implicit theory bias. With somewhat different possible sources of bias, there is reason to think that the two methods might not give estimates of change that agree. We cannot conclude that one method would be more biased than the other, but if the sources of bias are different, then concordance between the two measures is an issue of concern in estimating behavior change.

Before addressing questions of validity, we can ask questions about agreement between these different measures of sexual behavior change. The AIDS in Multi-Ethnic Neighborhoods (AMEN) Study, conducted between 1988 and 1992, was a longitudinal population survey of adults aged 20 to 44 in several high-risk neighborhoods in San Francisco. The AMEN Study collected panel/longitudinal estimates of sexual behavior change, and also used retrospective change measures at Wave 2. This data set thus provides an opportunity to compare data from the panel/longitudinal method and the retrospective change method on changes in sexual behavior. We focused on the two measures of sexual behavior change that are most often measured in evaluation research of AIDS prevention programs: condom use and change in number of partners over one year. We examined concordance in two separate risk groups of interest to AIDS prevention researchers: heterosexual men and women, and gay men. Because the sexual behavior of these two groups is quite different, both in number of sexual partners and type of sexual activity engaged in (vaginal sex vs. anal sex), and because what is perceived as socially desirable may be different between these two groups, separate analyses seemed appropriate.

METHODS

Sample Design and Data Collection

The AMEN Study was designed to investigate AIDS risk behaviors and attitudes in a sample of unmarried, multiethnic adults in San Francisco, an urban area with a high incidence of AIDS. We obtained a household probability sample of unmarried men and women aged 20 to 44 in 16 census tracts of San Francisco. Wave 1 interviews were conducted between April 1988 and July 1989. Wave 2 interviews were conducted approximately one year later, between May 1989 and December 1990. Full methodological details of this study have been reported in Catania, Stone, Binson, and Dolcini (1995).

We analyzed the two measures for two separate samples of risk groups: male and female heterosexual respondents who were sexually active during at least one wave (N = 962; 449 men, 513 women), and gay male respondents (N = 596).

Measures Used

At Wave 1 of our study, we asked respondents about risky sexual behavior (vaginal sex and anal sex) in the past year. Specifically, respondents were asked how many people they had had sexual intercourse with in the past year. Female respondents were told, "For the purposes of this study, sexual intercourse is defined as follows: A man puts his penis in your mouth, vagina or rectum." Male respondents were told, "For the purposes of this study, sexual intercourse is defined as follows: You put your penis in your partner's mouth, vagina or rectum, or your male partner puts his penis in your mouth or rectum." Immediately after being asked how many partners they had had in the past year, respondents were asked to recall the partner they had had sex with most recently. They were asked a series of questions about that partner, including demographic information and questions about sexual activities. In particular, heterosexual respondents were asked, "How many times, if any, did you have vaginal intercourse with that person in the past year?" Male respondents were asked "How many times did you put your penis in that partner's rectum in the past year?" and, if the partner was male, "How many times, if any, did that partner put his penis in your rectum in the past year?" The condom use questions were "Thinking of all the times you (had vaginal intercourse with that partner/put your penis in that partner's rectum/had that partner put his penis in your rectum), how often did you and that partner use a condom?" Interviewers actually referred to particular partners by their initials rather than saying "that partner." After the respondent had answered all the questions about the partner they had had sex with most recently, they were asked the same questions about the partner they had had sex with next most recently. This process was continued for up to 10 partners.

At Wave 2, which occurred approximately one year after Wave 1, we asked respondents the same questions about risky sexual behavior in the past year: that is, the year between Wave 1 and Wave 2 of the study. Differences between the same measures at Wave 1 and Wave 2 provided us with our panel measure of behavior change. There is a memory component to this measure, because subjects are being asked to remember their sexual behavior for the previous year. Blake, Sharp, Temoshok, and Rundell (1992) noted that recall of sexual behavior over a one-year period is most closely matched to recall of recent sexual behavior if respondents are prompted to remember behavior by individual partners (i.e., recalling their names or faces); thus, the method we used attempted to maximize accuracy of recall. The longitudinal/panel measure does not ask explicitly about change.

At Wave 2 of the study we also asked respondents for their retrospective self-reports of whether their AIDS risk behavior changed between Wave 1 and Wave 2 and how their risk behavior changed. Specifically, the interviewer asked "Tell me whether this is something you have done in the past year, since your last interview," and then listed (a) had fewer sexual partners, (b) used condoms/rubbers more often, (c) been more careful choosing your sexual partners, (d) used spermicidal jelly, cream, foam or sponge more often, (e) stopped having sex altogether, (f) changed the kinds of sexual practices you engage in. Respondents answered yes or no after each item. This provided us with our retrospective change measure, in which we asked about change explicitly.

For condom use, responses were averaged over both vaginal and anal intercourse, over receptive and insertive anal intercourse, and over all partners. We averaged over these types of intercourse to include all possible sexual risk behaviors in the condom use measure. Change was measured as any change in the average between Waves 1 and 2.

RESULTS

Heterosexual Respondents

Condom use: The prevalence of condom use at both waves was quite low, with the majority of respondents indicating that they use them less than half the time. The panel measure and the retrospective measure of change in condom use did agree (see Table 1). We computed kappas to measure the precise degree of agreement. Kappa is a measure that corrects for chance agreement, used particularly for dichotomous variables. It ranges between -1 and +1, with negative 1 indicating complete disagreement, 0 indicating no agreement, and +1 indicating complete agreement (Woolson, 1987). Kappa for these two measures of condom use was 0.38, indicating agreement, and significantly different from 0 (z = 11.6, p [is less than] 0.0001). A smaller number of respondents (N = 880) were used in calculating agreement between these two measures, because we had panel measures of condom use at both waves only for respondents who were sexually active at both waves. Of these, 675, or 76.7% of the sample, were in complete agreement on the two measures. The pattern of results was quite similar for male and female respondents (see Table 1).

Table 1. Agreement Between Longitudinal/Panel Method and Retrospective Change Method for Measuring Change in Frequency of Condom Use for Heterosexual Respondents

                                        Longitudinal/panel method
                                            Same frequency as
Retrospective change method                at W1 or less often

                                            All heterosexual
                                             respondents(a)
Same frequency as at W1 or less often            559(**)
More often than at W1                            66(**)
                                           Heterosexual men(b)
Same frequency as at W1 or less often            248(*)
More often than at W1                             25(*)
                                          Heterosexual women(c)
Same frequency as at W1 or less often            311(*)
More often than at W1                             41(*)

Retrospective change method               More often than at W1

Same frequency as at W1 or less often            139(**)
More often than at W1                            116(**)

Same frequency as at W1 or less often             69(*)
More often than at W1                             60(*)

Same frequency as at W1 or less often             70(*)
More often than at W1                             56(*)

Note. Change of frequency of condom use measured between Wave 1 and Wave 2. (a) N = 880. kappa = 0.38, z = 11.6. (b) N = 402. kappa = 0.41, z = 8.6. (c) N = 478. kappa = 0.35, z = 7.8.

(*) p < 0.001.

(**) p < 0.0001.

Number of partners: In contrast, we found a small level of disagreement between the panel measure and the retrospective measure of change in number of partners (see Table 2). Kappa for these two measures was -0.14, which, though small, is significant (z = -6.31, p [is less than] 0.001). Again, the pattern of results was quite similar for men and for women (see Table 2).

Table 2. Agreement Between Longitudinal/Panel Method and Retrospective Change Method for Measuring Change in Number of Partners for Heterosexual Respondents

                                     Longitudinal/panel method
Retrospective change method          Fewer partners than at W1

                                         All heterosexual
                                          respondents(a)
Fewer partners than at W1                     177(*)
Same # or more partners than at W1            118(*)
                                        Heterosexual men(b)
Fewer partners than at W1                      79(*)
Same # or more partners than at W1             60(*)
                                       Heterosexual women(c)
Fewer partners than at W1                      98(*)
Same # or more partners than at W1             58(*)

Retrospective change method               Same # or more
                                        partners than at W1

Fewer partners than at W1                     530(*)
Same # or more partners than at W1            137(*)

Fewer partners than at W 1                    251(*)
Same # or more partners than at W1             59(*)

Fewer partners than at W1                     279(*)
Same # or more partners than at W1             78(*)

Note. Change in number of partners measured between Wave 1 and Wave 2.

(a) N = 962. kappa = -0.14, z = -6.31.

(b) N = 449. kappa = -0.18, z = -5.36.

(c) N = 513. kappa = -0.11, z = -3.62.

(*) p < 0.001.

Of those who disagree on these two measures of change in number of partners (N = 648), there are two groups: those whose retrospective change measure looks more risky, and those whose longitudinal/panel measure looks more risky. A total of 530 people claimed in the retrospective change measure that they had had fewer partners (a health-protective change), but had a longitudinal/panel measure indicating that they had had the same number of partners or more. A total of 118 people claimed in the retrospective change measure that they had had the same number of partners or more (a possible increase in STD risk), but had a longitudinal/ panel measure indicating that they had had fewer partners. A total of 314 people, 32.6% of the sample, were in complete agreement between the two measures.

Gay Male Respondents

Condom use: Again, overall condom use was quite low, with most respondents claiming to use condoms less than half the time. Similar to the heterosexual respondents, gay male respondents' retrospective change measure of how their condom use had changed agreed with the panel measure of change in condom use (see Table 3). Kappa was 0.35, significantly greater than 0 (z = 8.32, p [is less than] 0.0001). Again, a smaller number of respondents (N = 541) was used to calculate concordance between measures of condom use because we had condom use measures at both waves only for respondents who were active at both waves. A total of 419 respondents, or 77.4% of the sample, were in complete agreement on the two measures.

Table 3. Agreement Between Longitudinal/Panel Method and Retrospective Change Method for Measuring Change in Frequency of Condom Use for Gay Male Respondents

                                        Longitudinal/panel method
                                            Same frequency as
Retrospective change method                at W1 or less often

Same frequency as at W1 or less often            361(**)
More often than at W1                             40(**)

Retrospective change method               More often than at W1

Same frequency as at W1 or less often            82(**)
More often than at W1                            58(**)

Note. Change of frequency of condom use measured between Wave 1 and Wave 2. N = 541. kappa = 0.35. z = 8.32.

(**) p < 0.0001.

Number of partners: For gay men (N = 596), there was also a small level of disagreement between change in the longitudinal/panel measure of number of partners between waves and the retrospective change measure of how their number of partners had changed between waves (see Table 4). Kappa was -0.16, significantly different from 0 (z= -6.57, p [is less than] 0.001).

Table 4. Agreement Between Longitudinal/Panel Method and Retrospective Change Method for Measuring Change in Number of Partners for Gay Male Respondents

                                        Longitudinal/panel method
Retrospective change method             Fewer partners than at W1

Fewer partners than at W1                         103(*)
Same # or more partners than at W                  69(*)

Retrospective change method                   Same # or more
                                           partners than at W1

Fewer partners than at W1                         359(*)
Same # or more partners than at W                  65(*)

Note. Change in number of partners measured between Wave 1 and Wave 2. N = 596. kappa = -0.16, z = -6.57.

(*) p < 0.001.

Of the 489 respondents who did disagree on the two measures, 382 reported on the retrospective change measure that they had had fewer partners, while their longitudinal/panel measure showed that they had had the same number or more partners. There were 107 respondents who reported on the retrospective change measure that they had had the same number or more partners, while their longitudinal/panel measure showed that they had had fewer partners. A total of 107 men, 18.0% of the sample, were in complete agreement on the two measures.

DISCUSSION

This study investigated agreement between two different methods of measuring change in sexual behavior among a sample of heterosexual men and women and a sample of gay men. For both samples, we found concordance between the two measures of change in condom use, but disagreement between the two measures of change in number of partners.

The retrospective change method and the longitudinal/panel method of measuring change provide convergent evidence for changes in condom use for both heterosexual respondents and gay male respondents. For both groups, the two types of measure were in agreement (kappa = 0.38 for heterosexual respondents, kappa = 0.35 for gay male respondents), though not complete agreement. Levels of agreement for heterosexual men (kappa: 0.41) and heterosexual women (kappa = 0.35) were similar. Although other studies have not looked at agreement between measures of change in condom use, they have looked at agreement between different measures of condom use. McLaws, Oldenburg, Ross, and Cooper (1990) looked at the concordance between recall of various sexual practices over the past month and diary entries in the form of a checklist filled out every morning for a month. They found that recall was best for intercourse without a condom. For condom use, they found correlations ranging between -0.28 and +0.99, depending on the type of intercourse (insertive or receptive, vaginal or anal). Saltzman, McCusker, Moon, and Mayer (1987) looked at test-retest agreement between two identical measures of condom use over a 6-week period and found kappas ranging from 0.47 to 0.61. Because our retrospective change measure asked about condom use without specifying which type of intercourse, we could not analyze the agreement between longitudinal/panel measures and retrospective change measures of condom use for vaginal intercourse, receptive anal intercourse and insertive anal intercourse separately. In future studies, it would be valuable to compare concordance measures of change in condom use for each type of activity.

We can only speculate about why our two measures of change in condom use agreed. The most pessimistic possibility is that the agreement in the change measures may come from people who reported no change: specifically, those who never use condoms, who know that they have not changed in this respect, and who report this fact consistently across both measures. It may be that high risk behavior is memorable, as McLaws et al. (1990) found, leading to consistent reporting of high-risk behavior. Alternatively, interpersonal factors may affect how salient certain risk behaviors are. Use of condoms requires substantial negotiation between partners. Because of this negotiation, it may be that going from never using condoms to any condom use (even infrequent), or from any condom use to no condom use, is a salient enough change that both of our change measures could pick it up and show agreement.

The situation is quite different for estimates of change in the number of partners over the past year. There was a small amount of disagreement between the two measures of behavior change for both heterosexual respondents (kappa = -0.14) and gay male respondents (kappa: -0.16). For heterosexual respondents, the degree of disagreement was similar for men and for women. Again, other studies have not looked at agreement between different measures of change in number of partners, but some have looked at agreement between different measures of number of partners in a given time period. McLaws et al. (1990) found correlations from 0.61 to 0.66 between recall of number of partners over the past month and diary entries over the past month. We can only speculate about the cause of disagreement between the two measures we compared. Recall over one month, as in McLaws et al.'s (1990) study, should be more accurate than recall over a year. Furthermore, asking someone "Are you having fewer partners than you had a year ago?" as retrospective change measures do is asking a question that could be interpreted several ways. Respondents who had 10 sexual partners in the past year and 15 the year before that could answer yes, they were having fewer partners. On the other hand, respondents who had 10 partners 9 months ago but settled down with one partner shortly before the interview might answer that they were having fewer partners, even though in the past year they had had more partners than a year ago. Retrospective change measures that are quite specific about which time periods the respondent is to compare might produce more reliable measures.

One other possible reason for disagreement between the two measures might be that when respondents answer the retrospective change question, they remember previous behavior as worse than it was; that is, they remember having had more partners a year ago than they actually did. Previous studies in other fields have demonstrated that retrospective measures of change can inflate estimates of change relative to the panel/longitudinal method on a number of self-report measures, such as personal health and use of welfare (Finney, 1981), personality and adjustment over a 25-year period (Woodruff & Birren, 1972), or degree of pain experienced on admission to a hospital clinic (Linton & Melin, 1982). Conway and Ross (1984), Linton and Melin (1982), Ross (1989), and Woodruff and Birren (1972) all found that people had a tendency to exaggerate change on retrospective change measures by remembering previous states as being worse than they were. Where objective measures were obtainable, the panel/longitudinal method agreed most closely with the objective data; retrospective change measures agreed significantly less. Finney (1981) cautions researchers seeking to evaluate programs that, because of its potentially greater bias, the retrospective change method can overestimate the impact of a program, making a program appear effective when it is not. Conway and Ross (1984) advise similar caution in interpreting retrospective change measures. However, we do not know that the disagreement between our longitudinal/panel measure and our retrospective change measure results from the retrospective change measure being less accurate. Our data can tell us only about agreement, not about validity of the two measures.

Social desirability has been found to have a strong effect on responses to questions about sexual behavior. It seems likely that respondents think they will be viewed favorably for making changes that will reduce their AIDS risk. Respondents' comments during interviews, such as, "Oh, I know I should be using condoms more," and "I promise I'll try to be better after this," indicate that some people felt significant self-presentation pressures during the interview. However, social desirability does not provide a ready explanation for agreement or disagreement between our two measures of change, because we do not know which of our two measures would be more subject to social desirability influences. We also note that we do not know for certain which behaviors the populations in the AMEN study consider socially desirable, and that the direction of social desirability might be different than what has been found for, say, college students.

If two methods of measurement of change in sexual behavior agree, as is the case with our two measures of change in condom use, we still have the question of whether they are valid, accurate measures. When two methods of measurement disagree, it raises the issue of which measure is more valid. Methodological research on sexual behavior has made great progress by focusing on reliability and validity of self-reported measures of sexual behavior. Researchers seeking to accurately evaluate intervention programs would benefit from more research on reliable and valid ways of measuring change in sexual behavior.

We hope to see more studies in the future comparing different methods for measuring behavior change, and more studies focused on how to assess the validity of change measures. For now, those seeking to evaluate behavior change might want to rely on more than one method of assessing change. As it becomes more vital to design effective prevention programs, evaluation research to assess sexual behavior change in such programs should take these methodological considerations into account.

REFERENCES

Baker, S. A., Morrison, D. M., Gillmore, M. R., & Schock, M. D. (1995). Sexual behaviors, substance use, and condom use in a sexually transmitted disease clinic sample. The Journal of Sex Research, 32, 37-44.

Baldwin, J. I., Whiteley, S., & Baldwin, J. D. (1990). Changing AIDS and fertility-related behavior: The effectiveness of sexual education. The Journal of Sex Research, 27, 245-262.

Becker, M. H., & Joseph, J. G. (1988). AIDS and behavioral change to reduce risk: A review. American Journal of Public Health, 78. 394-410.

Blake, S. M., Sharp, E. S., Temoshok, L., & Rundell, J. R. (1992). Methodological considerations in developing measures of HIV risk-relevant behavior and attitudes. Psychology & Health, 6, 265-280.

Blalock, H. M. (1979). Social statistics. New York: McGraw-Hill.

Catania, J. A., Binson, D., Canchola, J., & Pollack, L. M. (1996). Effects of interviewer gender, interviewer choice, and item wording on responses to questions concerning sexual behavior. Public Opinion Quarterly, 60, 345-375.

Catania, J. A., Binson, D., van der Straten, A., & Stone, V. E. (1995). Methodological research on sexual behavior in the AIDS era. Annual Review of Sex Research, 6, 77-125.

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., Gibson, D. R., Marin, B. V., Coates, T. J., & Greenblatt, R. (1990). Response bias in assessing sexual behaviors relevant to HIV transmission. Evaluation & Program Planning, 13, 19-29.

Catania, J. A., Stone, V. E., Binson, D., & Dolcini, M. M. (1995). Changes in condom use among heterosexuals in wave 3 of the AMEN survey. The Journal of Sex Research, 32, 193-200.

Coates, R. A., Calzavara, L. M., Soskolne, C. L., Read, S. E., Fanning, M. M., Sheperd, F. A., Klein, M. H., & Johnson, J. K. (1988). Validity of sexual histories in a prospective study of male sexual contacts of men with AIDS or an AIDS-related condition. American Journal of Epidemiology, 128, 719-728.

Cochran, S. D., Keidan, J., & Kalechstein, A. (1990). Sexually transmitted diseases and Acquired Immunodeficiency Syndrome (AIDS): Changes in risk reduction behaviors among young adults. Sexually Transmitted Diseases, 17, 80-86.

Conway, M., & Ross, M. (1984). Getting what you want by revising what you had. Journal of Personality and Social Psychology, 47, 738-748.

Finney, H. C. (1981). Improving the reliability of retrospective survey measures: Results of a longitudinal field survey. Evaluation Review, 5, 207-229.

Hingson, R., Strunin, L., & Craven, D. E. (1989). Survey of AIDS knowledge and behavior changes among Massachusetts adults. Preventive Medicine, 18, 806-816.

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

Hospers, H. J., & Kok, G. (1995). Determinants of sate and risk-taking sexual behavior among gay men. AIDS Education and Prevention, 7, 74-94.

Keeter, S., & Bradford, J. B. (1988). Knowledge of AIDS and related behavior change among unmarried adults in a low-prevalence city. American Journal of Preventive Medicine, 4, 146-152.

Laumann, E. O., Gagnon, J. H., Michael, R. T., & Michaels, S. (1994). The social organization of sexuality: Sexual practices in the United States. Chicago: The University of Chicago Press.

Linton, S. J., & Melin, L. (1982). The accuracy of remembering chronic pain, Pain, 13, 281-285.

McLaws, M.-L., Oldenburg, 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.

Ross, M. (1989). Relation of implicit theories to the construction of personal histories. Psychological Review, 96, 341-357.

Rotheram-Borus, M. J., Rosario, M., Meyer-Bahlberg, F. L., Koopman, C., Dopkins, S. C., & Davies, M. (1994). Sexual and substance use acts of gay and bisexual male adolescents in New York City. The Journal of Sex Research, 31, 47-57.

Saltzman, S. P., Stoddard, A. M., McCusker, J., Moon, M. W., & Mayer, K. H. (1987). Reliability of self-reported sexual behavior risk factors for HIV infection in homosexual men. Public Health Reports, 102, 692-697.

Woodruff, D. S., & Birren, J. E. (1972). Age changes and cohort differences in personality. Developmental Psychology, 6, 252-259.

Woolson, R. F. (1987). Statistical methods for the analysis of biomedical data. New York: John Wiley & Sons.

Manuscript accepted November 11, 1998

Valerie E. Stone, Joseph A. Catania, and Diane Binson

University of California, San Francisco

Valerie Stone is now at the Department of Psychology, University of Denver. Preparation of this article was supported by NIMH/NIDA center grant #MH42459, NIMH/NIA grant #MH43892, and AMEN 3 grant #MH48638.

We wish to acknowledge the very helpful contributions of Jesse Canchola, Estie Hudes, Walter Hauck, Ross Gibson, Bonnie Faigeles, David Heilbron, and two anonymous reviewers.

Address correspondence to Joseph Catania, Center for AIDS Prevention Studies, 74 New Montgomery, Suite 500, San Francisco, CA 94105; e-mail: [e-mail]