Inconsistencies in reporting the occurrence and timing of first intercourse among adolescents.

Surveys of adolescent sexual behavior rely on the information obtained from self-report. These data are used to develop research and policy agendas for adolescent reproductive health issues. Age at first (heterosexual) intercourse is commonly used to classify adolescents who are at risk for sexually transmitted infections (STIs), unintended pregnancy, and other social and health consequences of sexual activity. The accuracy of self-reported behaviors is an enduring concern among researchers. Findings from numerous methodological studies suggest that reporting inconsistencies are not random, but rather are associated with a variety of factors. These factors include the sociodemographic attributes of the respondent (Alexander, Somerfield, Ensminger, Johnson, & Kim, 1993; Catania, 1999; Lauritsen & Swicegood, 1997; Wu, Martin, & Long, 1999), the ability and motivation of the respondent (Catania, 1999; Weinhardt, Forsyth, Carey, Jarworski, & Durant, 1998), survey design, interview modality, and setting (Catania, Gibson, Chitwood, & Coates, 1990; Gribble, Miller, Rogers, & Turner, 1999; Turner et al., 1998; Wu et al., 1999). Much of this research, however, investigates reporting problems using adult samples or adolescent samples reinterviewed during adulthood. Fewer studies have specifically examined reports by adolescent samples at two points in time (for exceptions see Alexander et al., 1993; Capaldi, 1996; McFarlane & St. Lawrence, 1999; Mott, 1985; Newcomer & Udry, 1988; Rogers, Billy, & Udry, 1982).

Self-reports of sexual behavior cannot be easily checked against more objective data--that is, externally validated against a gold standard. Consequently, research has focused on assessing the consistency of self-reports. In cross-sectional studies, consistency can be examined by comparing responses across different questions that ask for similar sensitive information (Rogers et al., 1982; Wislar & Fendrich, 2000). Longitudinal studies provide the additional opportunity to identify logical inconsistencies in reporting behavior over time: specifically, adolescents who report they are sexually experienced at one point in time and subsequently deny being experienced at a later point in time (Alexander et al., 1993; McFarlane & St. Lawrence, 1999; Rogers et al., 1982). Longitudinal studies also have the ability to identify inconsistencies in age (or date) of sexual experiences across panels of data when event-history data are collected (Lauritsen & Swicegood, 1997; Wu et al., 1999).

In this paper, we use information obtained from both waves of the National Longitudinal Study of Adolescent Health (Add Health) to assess the consistency in the reporting of virginity status and the date of first sexual intercourse among adolescents. The work contributes to the evaluation of self-reported sexual behaviors using sample surveys in several ways. First, we examine the reporting problems in a nationally representative sample of a contemporary cohort of both boys and girls, using information obtained from two panels of data. Second, we expand on the race and ethnic groups that have been previously studied and provide some of the first national estimates of reporting problems for Hispanic and Asian American youth. Third, we investigate two separate types of inconsistent reporting: (a) those pertaining to inconsistent event occurrence (i.e., reclaimed virginity status), and (b) those pertaining to inconsistent timing of the event (i.e., different dates of first intercourse). Lastly, we develop seven strategies for addressing inconsistent reporting in longitudinal data and present findings for each strategy.

MEASURING ADOLESCENT SEXUAL BEHAVIOR

Much of what is known about adolescent (1) sexual behavior is understood in the context of heterosexual vaginal intercourse. This emphasis reflects the long-standing public health concerns of the social and health consequences of that behavior: unintended pregnancy, adolescent childbearing, and STI infection and sequelae (U.S. Department of Health & Human Services, 2000). Conceptually, first heterosexual intercourse has been viewed as a highly salient event for adolescents and used as one marker in the transition to adulthood (Elder, 1997). Empirically, younger ages of first intercourse are associated with increased risks for these social and health consequences (Alan Guttmacher Institute, 1994). The current investigation draws from this overall approach. We recognize, however, that adolescent sexual expression is not limited to heterosexual intercourse or heterosexual contact alone. We argue that the assessment of reporting problems specific to heterosexual intercourse is a necessary first step, and that the behavior itself (although limited in scope) has important public health implications.

National prevalence estimates of sexual experience among teens are derived from surveillance data such as the Youth Risk Behavior Survey (YRBS), cross-sectional survey data such as the National Survey of Family Growth (NSFG), and longitudinal survey data such as the National Survey of Adolescent Males (NSAM) and, more recently, the National Longitudinal Study of Adolescent Health (Add Health). All of these databases (minimally) define adolescents as sexually experienced if they have ever had vaginal intercourse. In 1999, 49.4% of adolescents were sexually experienced, and the percentage varied by gender and race/ethnicity (Centers for Disease Control [CDC], 2000). Over half (52.2%) of boys and 47.7% of girls have had heterosexual intercourse; 45.1% of White, 54.1% of Hispanic, and 71.2% of Black teens have had intercourse (CDC, 2000). These findings, along with those from other studies, are influential because they inform researchers and policy makers in setting priorities for health-related concerns for adolescents. Consequently, the degree of accuracy in reporting sexual behaviors impacts how risk groups are identified and how they are characterized.

When measuring sexual behavior, two types of information are often collected. Event-history type data require that the adolescent provide information on whether or not the event of interest occurred (i.e., ever had sexual intercourse) as well as the timing or date when the event occurred (i.e., month and year had sexual intercourse for the first time). Both of these measures are subject to reporting problems. Longitudinal studies allow one to assess reporting consistency across waves of data; in the case of the Add Health study, two waves of data are currently available. There are two possible transitions between Wave I and Wave II, but only one is logically inconsistent (i.e., sexually experienced at Wave I, not sexually experienced at Wave II). This subsample is the focus of the first analysis. We note, however, that these teens represent only a portion of potential respondents providing unreliable information. Even among teens providing consistent reports, some may be consistent but false. Similarly, some who state that they transitioned to becoming sexually active between the two waves may be giving false information. The second type of inconsistency we examine pertains to the timing of first intercourse. In this case, the respondent is consistent with his/her sexual experience report, but inconsistent about when the first sexual intercourse experience occurred. This subsample is the focus of the second analysis.

Researchers utilizing survey data are commonly faced with problems of inconsistent responses that they must resolve prior to analysis. These decision rules are often based on standard data management procedures (e.g., imputation) but are also based on how the researcher conceptualizes the underlying behavioral phenomenon she or he is investigating. In the case of sexual experience and age at first intercourse, one must decide which observations are to be counted as virgins versus nonvirgins. Also, among those observations that have multiple reports of the date of first intercourse, decisions must be made regarding which of these dates is the "truth." The current work allows for an assessment of the substantive implications of different strategies for resolving these problems.

Reporting inconsistencies have been found to systematically vary with the sociodemographic characteristics of respondents (Alexander et al., 1993; Catania, 1999; Lauritsen & Swicegood, 1997; Wu et al., 1999), their abilities and motivation (Catania, 1999), survey design, interview modality, and setting (Catania, et al., 1990; Gribble et al., 1999; Turner et al., 1998; Weinhardt et al., 1998; Wu et al., 1999). Studies investigating logical inconsistencies using adolescent samples have found between 5% and 12% of teens rescind reports of sexual intercourse, and the percentage varies by gender and race/ethnicity (Alexander, et al., 1993; McFarlane & St. Lawrence, 1999; Mott, 1985; Rogers et al., 1982). In general, African American boys are the most likely to give inconsistent reports and White girls the least; however, studies have been somewhat conflicting. Studies investigating reporting problems in age at first intercourse have found that between 25% to 66% of respondents provide inconsistent ages at first intercourse (Alexander et al., 1993; Capaldi, 1996; Lauritsen & Swicegood, 1997; Wu et al., 1999). Again, the findings have been conflicting; Some studies find that African American males have the greatest discrepancies, others African American females, and still others White females. In addition, very little is known about the reporting patterns of Hispanic and Asian American teens, and few studies analyze samples of teens that are nationally representative.

In the current study, we empirically investigate potential gender and race/ethnicity differentials for two types of inconsistent reporting, extending the number of groups to include Hispanic and Asian American adolescents. In addition to gender and race/ethnicity, we include the respondent's age and measures of family socioeconomic status (SES). The focus of our work is to assess the variability in reporting problems over a limited set of sociodemographic variables rather than testing why these differences are observed. We recognize that sociodemographic attributes serve as proxies for numerous underlying social and cultural differences pertaining to gender roles, sexual norms, and beliefs of social desirability and agreeability (Catania, 1999; Catania et al., 1990; DeGaston, Weed, & Jensen, 1996; Franco, Malloy, & Gonzalez, 1984; Furstenberg, Morgan, Moore, & Peterson, 1987; Sabogal, Perez-Stable, Otero-Sabogal, & Hiatt, 1995; Sprecher, McKinney, & Orbach, 1987), but do not explicitly test this in the current work.

In addition, the respondent's degree of interest, level of motivation, and abilities are also associated with reporting problems. Highly motivated respondents may apply greater effort to understanding and answering questions, whereas less motivated ones may skim or skip questions, or give less thoughtful answers (Catania et al., 1990). Also, since reporting tasks can be complex, respondents who demonstrate greater abilities may be more consistent reporters (Lauritsen & Swicegood, 1997; Wu et al., 1999). To investigate the potential association between motivation and reporting problems, we include two measures of the interviewer's assessment of the teen's exhibited boredom and candor. Also, to assess the potential contribution of vocabulary comprehension on reporting problems, we include the teen's score for a shortened version of the Peabody Picture Vocabulary Test.

Our analyses focus on a set of interrelated questions examining two forms of inconsistency in reporting: event occurrence and timing. We first examine the frequency of rescinded reports of sexual experience and the personal characteristics associated with them. Then, we investigate the frequency and magnitude of inconsistent reports of date of first intercourse and the personal characteristics associated with them. Finally, we develop seven strategies for resolving these inconsistencies and remark on the substantive implications of these different "fixes."

METHOD

The data were from the National Longitudinal Study of Adolescent Health (Add Health), a survey designed to assess the health status of adolescents in America (Bearman, Jones, & Udry, 1998). The sampling frame consists of all U.S. high schools; younger students were sampled from one feeder school that included a 7th grade and sent graduates to the sampled high school. Each participating school provided a student roster, which constituted the student-level sampling frame. From that listing, a baseline sample was drawn consisting of a core sample and several special oversamples. The core sample is a probability sample of 12,105 students and is nationally representative of adolescents enrolled in grades 7 to 12 during the 1994 to 1995 academic year. The oversamples include specific ethnic groups, disabled teens, and biologically related adolescents. The combined Wave I sample size is 20,745. The Wave II sample is all adolescents interviewed at Wave I except for the deletion of teens who were 12th graders at Wave I and who were not part of the genetic oversample, and the deletion of the disabled sample. Wave II interviews were completed in 1996 and resulted in a sample size of 14,738. We dropped 41 cases from the original 20,745 sample because of very poor data quality. Also, there were 403 adolescents who self-identified as Native American or Other who were dropped because of small sample size and heterogeneity. These deletions resulted in a baseline sample of 20,301.

In this paper, three distinct analyses were performed using different analytic strategies. For the first analysis, we examined logical inconsistencies in reporting of sexual status (i.e., reclamation of virginity status). This analysis was limited to the 5,049 teens who reported they had sexual intercourse at Wave I and were reinterviewed at Wave II. For the second analysis, we examined inconsistencies in reported date of first intercourse. This analysis was limited to the 3,855 adolescents who reported they had sexual intercourse at Wave I and Wave II and provided complete dates (month and year) of first intercourse in both interviews. (2) For the third analysis, we presented coding options for reconciling logical inconsistencies and dating inconsistencies. To the extent possible, we utilized information from the entire Wave I sample. For these analyses, the sample size varied depending on the strategy employed, and only cases with complete date of first intercourse were included. Cases with only partial or missing date information will be considered in subsequent work.

Measures and Variable Description

Measuring first heterosexual intercourse. Approximately midway through the survey, respondents were asked about their sexual and contraceptive histories. This portion of the interview was self-administered by audio-CASI (Computer Assisted Self Interview) because of the sensitive nature of the questions. Questions were read to respondents by means of audio headphones. Respondents were given instructions on how to complete their answers on the computer. They were first asked: "Have you ever had sexual intercourse? When we say sexual intercourse, we mean when a male inserts his penis into a female's vagina." This question was asked of all interviewed respondents at Wave I and Wave II. If the respondent answered yes, she or he was then asked: "In what month and year did you have sexual intercourse for the very first time?" The response given by the respondent at Wave I was not taken into consideration at the Wave II interview. Consequently, it is possible for teens to rescind their report and change their date of first sexual intercourse. Logistic regression was used to examine the correlates of logically inconsistent reports (1 = inconsistent report, 0 = consistent report). The second analysis examined differences in reported dates using two different measures. The first outcome was the difference, in months, between Wave I and Wave II reports of the date of first intercourse (Wave I minus Wave II). Ordinary least squares (OLS) regression was used. The second outcome was the squared values of the residuals obtained from the previous OLS analysis; it is a measure of the variability in the two reports. OLS was also used for this analysis.

Resolving inconsistencies in virginity status and reported dates of first intercourse. To resolve reporting inconsistencies, seven alternative coding schemes were developed and each option is described below. Survival analysis was used, so information for all respondents regardless of their re-interview status or sexual experience was utilized. (3)

Option 1: Wave I only. The Wave I date of first intercourse was used as the correct date and additional information provided in Wave II was ignored. Results based on this option provided a cross-sectional snap shot; thus, this option does not address any of the inconsistencies considered in this analysis.

Option 2 and option 3: Lower and upper boundaries. For respondents with two reports of date of intercourse, Option 2 selected the earlier (younger) of the two dates and Option 3 selected the later (older) of the two dates. If only one date was available, it was used. Both options treated logically inconsistent reports as nonvirgins, since any mention of sexual experience was used as correct.

Option 4: Wave I dominant. The Wave I information was used as correct except when the Wave II information represented incident cases. That is, we used the Wave II date of first sex only if it occurred between the interview dates for respondents who reported they were not sexually active at Wave I. Note that respondents who reported no for intercourse at Wave I and yes to intercourse at Wave II were counted as virgins if the date of intercourse was prior to the Wave I interview date. Respondents with logically inconsistent reports were treated as nonvirgins.

Option 5: Wave II dominant. The Wave II report was used as the correct report for respondents, regardless of the information provided at Wave I. Note that this option allowed respondents to transition to first intercourse even if the date provided at Wave II was prior to the Wave I interview date. Also, respondents with logically inconsistent reports were treated as virgins.

Option 6: More recent yes dominant. For respondents who provided two reports of date of first intercourse, the Wave II report was taken as correct. Respondents with logically inconsistent reports were treated as nonvirgins.

Option 7: Average. For respondents with two reports of first intercourse, the average of the two was used as the correct date. If there was only one report of date of first intercourse, that was used; thus, logically inconsistent reports were treated as nonvirgins.

This analysis consisted of two parts. First, to investigate the extent to which estimated ages at first intercourse vary across the seven options, Kaplan-Meier life tables were computed for each gender-by-ethnicity group. Then, Cox regression was used to explore if the different options produced substantive differences in the estimated effects of the independent variables on the time to first intercourse (hazard of first sex).

Independent Variables

The independent variables we investigated can be broadly classified into three groups: (a) sociodemographic attributes; (b) ability (vocabulary comprehension); and (c) motivation, which also characterizes aspects, of interview conditions. Sociodemographic characteristics included gender, race/ethnicity, age, family structure, and family socioeconomic status (SES). Gender and race/ethnicity were interacted, resulting in eight gender-by-ethnicity groups: White, African American, Hispanic, and Asian American females and the same categories for males. White female was used as the reference category. Family structure was categorized into four living situations: living with both biological parents, living in a stepfamily, living with one biological parent only, and all other living situations. We used two-biological-parent families as the reference category. Family SES consisted of three measures: mother's education, father's education, and household income. Missing values for parents' education were imputed using the weighted mean education, stratified by gender and race/ethnicity. Both are continuous variables and interpretable only for family situations in which the relevant parent is present. The annual household income for 1994 was taken from the parent's questionnaire and was imputed for cases with missing data. The final measure was the logarithm of income, which was then collapsed into quartiles, using the middle two (25-75) as the reference category.

The Add Health Picture Vocabulary Test (AHPVT) is a computerized and abridged version of the Peabody Picture Vocabulary Test and was used to assess vocabulary comprehension (Bearman et al., 1997). The AHPVT is a hearing vocabulary test where the interviewer reads a word and the adolescent selects an illustration that best fits the word, choosing from a multiple-choice format of items. There are 78 items on the AHPVT. We standardized the raw test score and coded it into quartiles using the middle two quartiles as the reference category. (In additional analysis not presented here, age standardized AHPVT scores were used; none of the findings differed from those presented in this paper.) Motivation and candor during the interview were assessed with information obtained from the interviewer; two measures from this battery were used. The first asked the interviewer to rate the candidness of the respondent in answering questions. This measure is a four-category variable ranging from not candid to very candid; the latter was used as the reference. The second question asked the interviewer to evaluate if the respondent appeared bored or impatient during the interview and was scored as a dichotomy.

All multivariate analyses presented in the paper were performed using STATA software (StataCorp, 1999). Robust variance estimates were computed to adjust for the complex clustered sample design using the Huber correction.

RESULTS

The first table shows the weighted distribution of the characteristics of the full sample. The average age of the respondent at Wave I was slightly older than 15 years. About two thirds of males and females were White, about 16% were African American, another 12% Hispanic, and the remainder were Asian American. Over one half of teens lived with both biological parents, approximately one quarter lived with one biological parent, and the remainder lived in stepfamilies and other situations. The mean years of schooling achieved were similar for mothers and fathers (about 13 years), and the mean household income (1994) was $42,8.64. The mean score of the raw AHPVT was 64.8. Over 80% of respondents were considered very or moderately candid during their interview, and only one in eight appeared bored during the interview (as observed by the interviewer).

Table 2 presents summary statistics for the two measures of inconsistent reports for each of the gender-by-ethnicity groups. White females had the lowest percentage of logically inconsistent responses (Column 1), at about 6%, and Asian American males had the highest percentage of inconsistent reports, at about 22%. In each of the four ethnic groups, males had higher percentages of inconsistent reporting than their female counterparts. As can be seen in the second column, teens on average revised their reported date of first sex to older ages, varying from 2.5 months (White girls) to almost 7 months (African American and Asian American boys). (4) Also, variability (SD) of these means differed across groups and was generally larger for males than for females. Thus, the descriptive findings demonstrated substantial variability in the extent of inconsistent reports across categories of gender-by-ethnicity. We now turn to the multivariate results.

Logically Inconsistent Reports of First Intercourse

The logistic regression results for the eight gender-by-ethnicity categories mirror the bivariate results in Table 2. That is, all groups had higher odds of inconsistent reporting than White girls, although not significantly so for Hispanic and Asian American girls. African American girls were 1.5 times more likely than White girls to provide logically inconsistent information. Most striking, however, were the findings for boys in every ethnic group. White, African American, and Hispanic boys were more than two times more likely and Asian American boys almost 5 times more likely than White girls to be inconsistent reporters. Also, older teens were less likely than younger teens to change their response at Wave II.

Family background effects were modest. Controlling for parents' education, teens living in other situations were less likely to provide logically inconsistent reports (see Table 3). Higher levels of mother's education were significantly associated with lower odds, among teens living with mothers. In contrast, neither father's education nor household income had an impact on reporting. In addition, teens who scored in the first quartile (lowest) on the AHPVT were significantly more likely than those scoring in the middle two quartiles to provide inconsistent reports. However, neither the teen's degree of candor nor level of boredom was associated with reporting behavior.

Variability in Reported Dates of First Intercourse

In the next table (Table 4), we investigated factors that may be associated with inconsistent reports of the date of first heterosexual intercourse. The dependent variable in the first model (Column 1) was the difference in dates, and the dependent variable in the second model (Column 2) was the squared residuals obtained from the first model. A notable finding for both outcomes is that the model [R.sup.2] explained very little of the total variability in reported dates, suggesting small systematic variation in the reports for these model specifications. Looking first at Model 1 results, we found that all gender-by-ethnicity groups revised their dates to older ages (except for Asian American girls), and only two of the contrasts were significant. Compared to White girls, White and African American boys revised their dates significantly upward to older ages (by almost 4 months). There was a modest and somewhat unexpected effect for teens scoring in the top quartile of the AHPVT, who revised their reports to younger ages at Wave II. Family background and interview characteristics were not significant.

Model 2 shows the extent of variability in reported dates. When compared to White girls, African American boys had significantly greater variability, followed by White and Hispanic boys. None of the other gender-by-ethnicity contrasts were significant. Again, family background characteristics were only modestly important; we found that teens living in other family situations had significantly greater variability and those living in higher income households had significantly less. Lastly, teens scoring lower on the AHPVT had significantly greater variability in reported dates than those scoring in the middle two quartiles.

Resolving Logical and Date Inconsistencies: Seven Options

For the last analysis, we investigated the following questions: To what extent do inconsistencies in reporting alter estimates of age at first sex? And, to what extent do these inconsistencies alter the estimates of the covariates on the risk of first sexual intercourse? To answer these questions, we examined the results obtained from seven different options for reconciling inconsistencies. Note that for all but one of the options (Option 5), teens were considered to be nonvirgins if there is any mention of first intercourse. Option 5 allowed teens to regain their virginity status, and they were counted as virgins. Table 5 shows the Kaplan-Meier estimated median ages of first intercourse for girls and boys, respectively. What is most notable in the table is the overall similarity in the estimates, regardless of the option used for computation. For girls, African American girls generally reported the youngest age, followed by White, Hispanic, and, lastly, Asian American girls (for whom ages could not be computed for most options because so few had become sexually active by the last interview). The last column shows the greatest difference (in fraction of years) between the two most disparate estimates of median age at first sex. For girls, the largest difference was 0.4 years. For boys, African American boys reported the youngest ages of first sex overall, followed by Hispanic, White, and, lastly, Asian American boys. There was somewhat greater variability in the estimated ages among boys. The largest discrepancies in age were among African American boys (0.8 years) and Asian American boys (0.6 years).

For our final analysis, we estimated multivariate models of the time to first sex using Cox proportional hazard regression (Table 6). We were interested in (qualitatively) assessing variation in the parameter estimates across the options, highlighting any changes in substantive findings. Overall, differences in substantive results were small, with a few notable exceptions. Compared to White girls, Hispanic and Asian American girls had significantly lower risk (older ages) of first intercourse, as did White and Asian American boys. African American boys had significantly higher risk (younger ages) than White girls, and Hispanic boys did not differ from White girls. Note, however, that the results for African American girls depended on the option selected. In two cases, we found that African American girls had significantly higher risk than White girls. In five cases, the differences between African American and White girls were not statistically significant. Thus, although the effects for African American girls were substantively the same across the options, the precision of the estimates differed. It is also possible to see the impact of logically inconsistent reports in the results from Option 5. Because boys were more likely to provide logically inconsistent reports (and were thus coded as virgins in this case), the higher relative risk for African American boys was somewhat suppressed, and the lower risk among White and Asian American boys was even lower when compared to the other options.

Because family background characteristics were not strongly predictive of inconsistent reporting, their effects were strikingly similar across the options. Controlling for parents' education, teens living in stepfamilies had higher risk of sex than those living with both biological parents. Higher levels of both mothers' and fathers' education reduced the risk of sex among those teens living with a mother and/or father. Across alt options, household income was not significant. Lastly, compared to the middle two quartiles, teens scoring in the top quartile of the AHPVT had significantly lower risk of intercourse; Option 5 results show that teens scoring, in the first quartile also had significantly lower risk.

DISCUSSION

Our findings suggest that although adolescents may have difficulty providing consistent information about their sexual experience, these reporting problems (i.e., reclaiming virginity status and changing the date of first sex) do not substantially impact the substantive conclusions regarding age at first sex. Taken together, the results presented in Tables 3 and 4 suggest that the overall variability is only modestly explained by the models. Or, inconsistencies in reporting first heterosexual intercourse are largely random (unspecified), with a few notable exceptions. Boys, especially Asian American boys, are more likely than White girls to provide logically inconsistent reports, as are teens with lower AHPVT scores; older teens are less likely to do so. Among sexually active teens who are not logically inconsistent, boys, especially White and African American boys, report greater differences than White girls in dates of first intercourse. With respect to the degree of variability in reported dates, boys, in particular African American boys, have much greater dispersion in the dates they provide than White girls, as do teens scoring in the lowest quartile of the AHPVT. In addition, family background factors, the degree of candor, and the level of boredom during the interview do not influence reporting.

Although we presented gender-by-ethnicity interaction models, our findings suggest that gender differences in inconsistent reports are greater than race/ethnic differences. Overall, boys are more inconsistent reporters than girls, results found by others analyzing earlier cohorts of adolescents (Alexander et al., 1993; Lauritsen & Swicegood, 1997; Mott, 1985; Rogers et al., 1982). Nevertheless, the gender-by-ethnicity comparisons provide some additional insights. When compared to White girls, Asian American boys are almost five times more likely to provide logically inconsistent reports. When compared to White girls, both White and African American boys significantly revise their ages at first intercourse to older ages. Although we did not attempt to explain why these differences are observed, other studies have suggested that honesty in reporting may be a factor (Newcomer & Udry, 1988; Siegel, Aten, & Roghmann, 1998). When students were asked directly about the honesty of their reports of sex, gender, race/ethnicity, and nonintercourse sexual experiences were associated with teens' propensity to be honest (Newcomer & Udry, 1988). And, the youngest boys were the least likely to provide honest reports about their sexual experiences (Siegel et al., 1998). Problems with recall may also be a contributing factor, although the recall period for most of the teens in the sample is of relatively short duration and onset of sexual activity is most often considered a salient experience. Nevertheless, bias due to memory problems are particularly relevant for recollection of dates, and many people may be poor reporters of the timing of even relatively recent events (Sudman, Bradburn, & Schwarz, 1996).

Teens scoring in the lowest quartile of AHPVT were more likely to rescind their report and more likely to give inconsistent dates of first intercourse than teens with higher scores. Teens with lower scores may have more difficulty understanding the content of the questions being asked and thus more inclined to provide faulty answers. Also, they may have greater difficulty performing the computations needed to convert their age at first intercourse into a calendar date (Wu et al., 1999). In other work analyzing the overall quality of the Add Health data, teens with lower AHPVT scores were more likely to give out-of-range, missing, or "don't know" responses throughout the survey (Joyce Tabor, personal communication, 1999).

An important limitation of our study is our narrow definition of sexual behavior--heterosexual vaginal intercourse--and our focus on the age when this first occurred. Adolescent sexual expression encompasses a multiplicity of behaviors, and there are forms of sexual initiation other than heterosexual vaginal intercourse. A recent commentary (Remez, 2000) discusses the difficulties in characterizing these behaviors, especially in relation to identifying teens "at risk." Because of the narrow focus on sexual intercourse in past research, a major component of early sexual activity, namely noncoital behaviors, has historically been overlooked. Anecdotal reports suggest that teens may be engaging in behaviors (e.g., oral sex) that avoid pregnancy but leave them at risk for STIs, including HIV (Remez, 2000). Similarly, this perspective also overlooks same-sex sexual behaviors, which also place teens at risk of STIs (in some cases, substantially increased risk). Failure to measure and estimate the prevalence of these behaviors will underestimate adolescents' risk for STIs and hinder our ability to elaborate the Complexities of adolescent sexual behaviors. Nonetheless, we argue that assessing the reliability of reporting of heterosexual vaginal intercourse is useful because it is an important marker of risk. One of the innovations of the Add Health data set is that it also contains information regarding same-sex and other noncoital behaviors (Bearman et al., 1997).

The results show that inconsistency in reporting has only limited influence on the estimated ages at first intercourse or on the effects of standard sociodemographic determinants on the risk of first sex. Despite different assumptions about virginity status and about which reported date of first intercourse was considered true, the seven options we developed for resolving inconsistencies produced substantively similar results. Although others also find that inconsistent reporting does not affect results for age at first intercourse (Lauritsen & Swicegood, 1997; Wu et al., 1999), these studies did not present a range of options for resolving inconsistencies. In particular, we found that the gender-by-ethnicity differentials in risk did not markedly differ across the options, with the exception of the results for African American girls. Family background effects also were remarkably similar across the fixes. The substantive results are in agreement with other studies using the Add Health data set as well as other nationally representative data sets such as the YRBS (Kolbe, Kann, & Collins, 1993; Resnick et al., 1997). The large sample size of the Add Health data set undoubtedly contributed to the robustness of these findings. Nevertheless, our findings also show that the prevalence estimates of sexual experience are influenced by these reporting problems; the estimates vary from 38.9% (Option 1) to 48.3% (Options 2 and 5) (data not shown). Thus, these coding options do not substantially impact the effects of the explanatory variables but do affect the point estimates of the prevalence of sexually active adolescents.

In sum, although teens may have difficulty providing consistent information regarding their sexual experience, it appears that these reporting problems are largely random and that they have little impact on the substantive conclusions regarding age at first intercourse. Our work shows, however, that there are still moderate levels of inconsistent reporting of these behaviors, and points to the importance of continuing to explore the ways in which adolescents understand sexuality, sexual behaviors, and the personal meanings they attach to these phenomena.

Table 1. Distribution of Sample Characteristics, Add
Health, Wave I

Characteristics                Percentage or mean

Age                                    15.4
Gender by race/ethnicity
   White females                       33.4%
   African American females             8.1
   Hispanic females                     6.0
   Asian American females               1.7
   White males                         34.5
   African American males               8.0
   Hispanic males                       6.3
   Asian American males                 1.9
Family structure
   Two biological parents              54.7%
   Stepfamily                          10.3
   One biological parent               28.6
   Other situations                     6.4
Family SES
   Mother's education                  13.1
   Father's education                  13.3
   Household income                42,864
AHPVT                                  64.79
Respondent candor
   Very candid                         51.3%
   Moderately candid                   33.9
   Somewhat candid                     10.5
   Not candid                           4.3
Respondent boredom
   Bored (vs. not bored)               12.6%
N                                  20,301

Note. Weighted percentages and means; unweighted N.

Table 2. Summary Statistics of Logically Inconsistent
Reports and Inconsistent Dates of Reported First
Intercourse, by Gender-by-Ethnicity Groups

                    Percent logically          Mean
                      inconsistent        differences in
                       (reclaimed         reported dates
                    virginity status)    (Wave I-Wave II)      (SD)

White girls                5.94               -2.54          (14.13)
African American
   girls                   9.78               -4.22          (21.93)
Hispanic girls             9.33               -3.67          (16.51)
Asian American
   girls                  12.26               -2.91          (15.16)
White boys                12.85               -5.73          (26.79)
African American
   boys                   14.75               -6.94          (39.29)
Hispanic boys             13.96               -4.99          (29.12)
Asian American
   boys                   22.22               -6.87          (26.62)
Total                     11.2                -4.5
N                      5,049               3,855

Table 3. Logistic Regression Coefficients for Rescinded
Sexual Intercourse at Wave II, Add Health

                                         Parameter      Odds
Characteristics                          estimate       ratio

Deviation from mean age                   -0.252 ***    0.777
Gender by race/ethnicity
    African American females               0.382 *      1.465
    Hispanic females                       0.296        1.345
    Asian American females                 0.853        2.347
    White males                            0.883 ***    2.418
    African American males                 0.839 ***    2.314
    Hispanic males                         0.799 ***    2.223
    Asian American males                   1.564 ***    4.778
Family structure
    Stepfamily                            -0.161        0.851
    One biological parent                 -0.416        0.659
    Other situations                      -0.839 *      0.432
Family SES
    Mother's education                    -0.037 *      0.964
    Father's education                    -0.013        0.987
    Household income-first quartile        0.067        1.069
    Household income-fourth quartile      -0.023        0.977
Peabody Picture Vocabulary Test
    First quartile                         0.337 ***    1.400
    Fourth quartile                       -0.138        0.871
Interview characteristics
    Moderately candid                      0.137        1.147
    Somewhat candid                        0.152        1.164
    Not candid                             0.383        1.468
    Bored                                 -0.043        0.957
    Constant                              -2.052         --
N                                      5,049
a.f                                       21
Log-likelihood                         -1673.41 ***

Note. Characteristics limited to those teens who were sexually active at
the Wave I interview. Huber correction for SE.

* p < 0.05. ** p < 0.01. *** p < 0.001.

Table 4. OLS Regression Coefficients for the Difference (in
Months) in Reports of Date of First Intercourse
(Model 1) and for Squared Residuals of
Differences (Model 2)

                                          Model--      Model 2--squared
Characteristics                         differences       residuals

Deviation from mean age                    0.690           2.744
Gender by race/ethnicity
    African American females              -1.021          62.79
    Hispanic females                      -0.565         -86.46
    Asian American females                 0.033         -98.85
    White males                           -3.340 **      535.24 ***
    African American males                -3.888 **     1141.79 ***
    Hispanic males                        -2.229         526.67 **
    Asian American males                  -4.182         437.95
Family structure
    Stepfamily                             0.295          45.89
    One biological parent                  1.333         156.72
    Other situations                      -2.036         789.98 *
Family SES
    Mother's education                    -0.044          17.18
    Father's education                     0.062          16.86
    Household income-first quartile        1.852          91.99
    Household income-fourth quartile       0.769        -243.46 *
Peabody Picture Vocabulary Test
    First quartile                        -1.251         403.58 ***
    Fourth quartile                        1.759 *       -92.32
Interview characteristics
    Moderately candid                     -0.437         106.28
    Somewhat candid                       -0.154         -15.79
    Not candid                             3.308          28.79
    Bored                                 -2.239          52.23
    Intercept                             -3.866        -281.84
N                                      3,855           3,855
d.f.                                      21              21
F value                                    3.53 ***       11.77 ***
R2                                         0.01            0.04

Note. Characteristics limited to adolescents who reported they were
sexually active at both Wave I and Wave II interviews. Huber
correction for SE.

* p < 0.05. ** p < 0.01. *** p < 0.001.

Table 5. Kaplan Meier Life Table Results for Median Age at
First Sex, Based on Seven Different Options

Ethnic Groups      Option 1   Option 2   Option 3    Option 4

                                      Girls

White                16.9       16.8       16.9        17.0
African American     16.3       16.2       16.5        16.4
Hispanic             18.0       17.6       17.6        17.9
Asian American                             18.8

                                      Boys

White                17.5       17.3       17.4        17.6
African American     15.4       15.0       15.5        15.4
Hispanic             17.l       16.7       17.0        17.2
Asian American       19.5       19.5       19.8        19.5

Ethnic Groups      Option 5   Option 6   Option 7   Greatest [DELTA]

                                      Girls

White                17.0       16.9       16.8         0.2
African American     16.6       16.4       16.3         0.4
Hispanic             17.8       17.6       17.6         0.4
Asian American

                                      Boys

White                17.6       17.4       17.3         0.3
African American     15.8       15.3       15.2         0.8
Hispanic            17. l       16.9       16.8         0.5
Asian American       19.5       19.2       19.5         0.6

Note. See text for description of coding options.

Table 6. Cox Proportional Hazard Results for Risk of First Intercourse,
Using Seven Different Coding Options

Characteristics                     Option 1            Option 2

Gender by race/ethnicity
   African American girls              0.148 *            0.116 *
   Hispanic girls                     -0.433 ***         -0.413 ***
   Asian American girls               -0.734 ***         -0.701 ***
   White boys                         -0.137 ***         -0.125 ***
   African American boys               0.609 ***          0.614 ***
   Hispanic boys                      -0.034              0.014
   Asian American boys                -0.724 ***         -0.703 ***
Family structure
   Stepfamily                          0.403 ***          0.407 ***
   One biological parent               0.076              0.051
   Other situations                   -0.079             -0.176
Family SES
   Mother's education                 -0.026 **          -0.031 ***
   Father's education                 -0.030 ***         -0.032 ***
   HH income--first quartile          -0.028             -0.016
   HH income--fourth quartile          0.018              0.016
AHPVT
   First quartile                     -0.075             -0.052
   Fourth quartile                    -0.233 ***         -0.229 ***
N                                 19,232             19,467
Log-likelihood                   -67,058.927        -85,285.925

Characteristics                     Option 3            Option 4

Gender by race/ethnicity
   African American girls              0.092              0.108
   Hispanic girls                     -0.421 ***         -0.401 ***
   Asian American girls               -0.729 ***         -0.752 ***
   White boys                         -0.156 ***         -0.167 ***
   African American boys               0.538 ***          0.549 ***
   Hispanic boys                      -0.044             -0.060
   Asian American boys                -0.744 ***         -0.756 ***
Family structure
   Stepfamily                          0.389 ***          0.393 ***
   One biological parent               0.077              0.075
   Other situations                   -0.172             -0.118
Family SES
   Mother's education                 -0.028 ***         -0.029 ***
   Father's education                 -0.029 ***         -0.030 ***
   HH income--first quartile          -0.023             -0.023
   HH income--fourth quartile          0.015              0.041
AHPVT
   First quartile                     -0.069             -0.077
   Fourth quartile                    -0.227 ***         -0.213 ***
N                                 19,479             19,232
Log-likelihood                   -84,818.191        -78,089.854

Characteristics                     Option 5            Option 6

Gender by race/ethnicity
   African American girls              0.073              0.098
   Hispanic girls                     -0.408 ***         -0.421 ***
   Asian American girls               -0.730 ***         -0.719 ***
   White boys                         -0.194 ***         -0.147 ***
   African American boys               0.457 ***          0.552 ***
   Hispanic boys                      -0.077             -0.029
   Asian American boys                -0.813 ***         -0.731 ***
Family structure
   Stepfamily                          0.387 ***          0.397 ***
   One biological parent               0.079              0.068
   Other situations                   -0.152             -0.192
Family SES
   Mother's education                 -0.028 **          -0.029 ***
   Father's education                 -0.029 ***         -0.030 ***
   HH income--first quartile          -0.025             -0.019
   HH income--fourth quartile          0.022              0.018
AHPVT
   First quartile                     -0.111 **          -0.065
   Fourth quartile                    -0.198 ***         -0.223 ***
N                                 19,475             19,474
Log-likelihood                   -80,801.983        -84,923.082

Characteristics                     Option 7

Gender by race/ethnicity
   African American girls              0.105
   Hispanic girls                     -0.421 ***
   Asian American girls               -0.717 ***
   White boys                         -0.135 ***
   African American boys               0.602 ***
   Hispanic boys                      -0.008
   Asian American boys                -0.721 ***
Family structure
   Stepfamily                          0.402 ***
   One biological parent               0.064
   Other situations                   -0.163
Family SES
   Mother's education                 -0.029 ***
   Father's education                 -0.030 ***
   HH income--first quartile          -0.017
   HH income--fourth quartile          0.014
AHPVT
   First quartile                     -0.052
   Fourth quartile                    -0.233 ***
N                                 19,479
Log-likelihood                   -85,169.316

Note. Huber correction for SE.

* p < 0.05. ** p < 0.01. *** p < 0.001.

This research is supported by grant R01 HD36993 from the National Institute of Child Health and Human Development. An earlier version of this paper was presented at the Annual Meeting of the Population Association of America, March 25-27, 1999, New York, NY. The authors acknowledge Yasamin Kusunoki and Rachel Oakes for their invaluable assistance in the preparation of the manuscript.

(1) Adolescent is defined broadly and includes individuals between the ages of 11 and 20.

(2) Because our inconsistency analysis uses information from both waves of data and because the oldest teens (high school seniors) were not reinterviewed, we can only make statements about reporting problems for younger teens (i.e., between 12 and 18).

(3) In each of the following options, for respondents who were not reinterviewed at Wave II, we used the information available from the Wave I interview. In addition, where it was determined that the respondent was a virgin, the observation was censored at the appropriate interview date.

(4) We also note that only 22.2% of teens reported the same date of first intercourse, and only one third reported dates within 6 months of each other (data not shown).

REFERENCES

Alan Guttmacher Institute. (1994). Sex and America's teenagers. New York: Author.

Alexander, C. S., Somerfield, M. R., Ensminger, M. E., Johnson, K. E., & Kim, Y. J. (1993). Consistency of adolescents' self-report of sexual behavior in a longitudinal study. Journal of Youth and Adolescence, 22, 455-471.

Bearman, P. S., Jones, J., & Udry, J. R. (1998). The national longitudinal study of adolescent health: Research design. Carolina Population Center, University of North Carolina, Chapel Hill, NC. Retrieved 1998 from the World Wide Web: http://www.cpc.unc.edu/projects/addhealth/design.html.

Capaldi, D. (1996). The reliability of retrospective report of timing of first sexual intercourse for adolescent males. Journal of Adolescent Research, 11, 375-387.

Catania, J. A. (1999). A framework for conceptualizing reporting bias and its antecedents in interviews assessing human sexuality. The Journal of Sex Research, 36, 25-38.

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 the studies of sexual behavior. Psychological Bulletin, 108, 339-362.

Centers for Disease Control and Prevention. (2000). Sexually transmitted disease surveillance 1999. Atlanta, GA: Division of STD Prevention, Department of Health and Human Services.

DeGaston, J. F., Weed, S., & Jensen, L. (1996). Understanding gender differences in adolescent sexuality. Adolescence, 31, 217-231.

Elder, G. H., Jr. (1997). The life course and human development. In R. M. Lerner (Ed.), Handbook of child psychology: Vol. 1. Theoretical models of human development (pp. 939-991). New York: Wiley.

Franco, J., Malloy, T., & Gonzalez, R. (1984). Ethnic and acculturation differences in self-disclosure. Journal of Social Psychology, 122, 21-32.

Furstenberg, F. F., Jr., Morgan, S. P., Moore, K. A., & Peterson, J. L. (1987). Race differences in the timing of adolescent intercourse. American Sociological Review, 52, 511-518.

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.

Kolbe, L., Kann, L., & Collins, J. (1993). Overview of the Youth Risk Behavior Surveillance System. Public Health Report, 108 (Suppl. 1), 2-10.

Lauritsen, J. L., & Swicegood, C. G. (1997). The consistency of self-reported initiation of sexual activity. Family Planning Perspectives, 29, 215-221.

McFarlane, M., & St. Lawrence, J. S. (1999). Adolescents' recall of sexual behavior: Consistency of self-report and effect of variations in recall duration. Journal of Adolescent Health, 25, 199-206.

Mott, F. (1985). Evaluation of fertility data and preliminary analytic results from the 1983 (5th round) survey of the National Longitudinal Surveys of Work Experiences of Youth. Technical Report, Ohio State University, Center for Human Resource Research.

Newcomer, S., & Udry, J. R. (1988). Adolescents' honesty in a survey of sexual behavior. Journal of Adolescent Research, 3, 419-413.

Remez, L. (2000). Oral sex among adolescents: Is it sex or is it abstinence? Family Planning Perspectives, 32, 298-304.

Resnick, M. D., Bearman, P. S., Blum, R. W., Bauman, K. E., Harris, K. M., Jones, J., et al. (1997). Protecting adolescents from harm: Findings from the National Longitudinal Study on Adolescent Health. Journal of the American Medical Association, 278, 823-832.

Rogers, J. L., Billy, J. O. G., & Udry, J. R. (1982). The rescission of behaviors: Inconsistent responses in adolescent survey data. Social Science Research, 11, 280-296.

Sabogal, F., Perez-Stable, E. J., Otero-Sabogal, R., & Hiatt, R. A. (1995). Gender, ethnic, and acculturation differences in sexual behaviors: Hispanic and non-Hispanic White adults. Hispanic Journal of Behavioral Sciences, 17, 139-159.

Siegel, D. M., Aten, M. J., & Roghmann, K. J. (1998). Self-reported honesty among middle and high school students responding to a sexual behavior questionnaire. Journal of Adolescent Health, 23, 20-28.

Sprecher, S., McKinney, K., & Orbuch, T. L. (1987). Has the double standard disappeared? An experimental test. Social Psychology Quarterly, 50, 24-31.

StataCorp. (1999). Stata statistical software: Release 6.0. College Station, TX: Stata Corporation.

Sudman, S., Bradbum, N. M., & Schwarz, N. (1996). Thinking about answers: The application of cognitive processes to survey methodology. San Francisco: Jossey-Bass.

Turner, C. F., Ku, L., Roger, S. M., Lindberg, L. D., Pleck, J. H., & Sonenstein, F. L. (1998). Adolescent sexual behavior, drug use, and violence: Increased reporting with computer technology. Science, 280, 867-873.

U.S. Department of Health and Human Services (2000). Healthy people 2010. Washington, DC: Author.

Weinhardt, L. S., Forsyth, A. D., Carey, M. P., Jaworski, B. C., & Durant, L. E. (1998). Reliability and validity of self-report measures of HIV-related sexual behavior: Progress since 1990 and recommendations for research and practice. Archives of Sexual Behavior, 27, 155-180.

Wislar, J. S., & Fendrich, M. (2000). Can self-reported drug use data be used to assess sex risk behavior in adolescents? Archives of Sexual Behavior, 29, 77-89.

Wu, L. L., Martin, S. P., & Long, D. A. (1999, October). Comparing data quality of fertility and first sexual intercourse histories. Paper presented at the Data Quality in Longitudinal Surveys conference, Hyattsville, MD.

Manuscript accepted February 18, 2002

Dawn M. Upchurch University of California, Los Angeles

Lee A. Lillard University of Michigan (deceased)

Carol S. Aneshensel University of California, Los Angeles

Nicole Fang Li Genentech, CA

Address correspondence to Dawn M. Upchurch, UCLA School of Public Health, 650 Charles Young Drive South, Los Angeles, CA 90095-1772; email: [e-mail].