We calculated the frequency of each behavioral occurrence at each wave, including the number of adolescents who (1) did not engage in oral or vaginal sex, (2) engaged in oral but not vaginal sex, (3) engaged in vaginal but not oral sex, and (4) engaged in both behaviors. These data were used to identify the temporal order of oral and vaginal sex, with analyses conducted on all 627 participants. Once the temporal order was established, we used discrete-time survival analysis to test whether one behavior (eg, oral sex) increased or decreased the likelihood of the other behavior (eg, vaginal sex) at each period of assessment. Since analyses were designed to describe whether and/or when initiation occurred, analyses were conducted on the 560 participants who did not engage in vaginal sex prior to the first assessment. Discrete-time survival analysis is a statistical technique based on logistic regression.17- 20 Logistic regression was conducted on data that had been restructured into a person-period format, with the data amounting to 2364 records for the 560 participants. This technique is commonly used in longitudinal analyses to investigate risks of onset, since once onset has occurred, the participant should no longer be included in the base rate assessments in subsequent waves. This data analytic strategy allowed us to focus on the first report of each behavior across periods. Participants remained in the analysis for each time wave until (1) they reported the behavior, (2) they dropped out of the study (censored), or (3) the study concluded. Therefore, once a participant reported oral sex, information about oral or vaginal sex was not needed for subsequent waves. The advantage of using this statistical technique is that all 560 participants were retained for analyses. The vast majority of participants remained until they reported sexual behaviors; only 2 participants dropped out of the study before the last assessment period and before reporting sexual behavior. The results were reported in 2 ways. First, we reported hazard rates, which can be interpreted as probabilities of initiating vaginal sex at that particular point, given initiation had not occurred previously. Second, we reported survival rates, which can be interpreted as the probability that an individual will have abstained from vaginal sex up to that interval. The hazard function is an interval-specific estimate; the survival function is based on cumulative survival rates from previous intervals.