Family Instability and Early Initiation of Sexual Activity in Western Kenya

Consistent with standard practice in the international literature, orphans are defined in this article as children who have lost one or both parents (UNAIDS, UNICEF, and USAID 2004).

I also do not attempt to investigate whether particular sources of family change are more highly associated with the outcome than others. Rather, I aim to broaden the scope of family instability considered and measured.

Coming as they do from a cross-sectional study, the RHC data represent only those alive and residing in Kisumu in the year 2007, and not of all those in Kisumu at the start of the calendar in 1997.

For example, it is not uncommon for young people in rural Kenya to live in extended-family compounds that comprise multiple closely clustered residences. It is possible that a primary caregiver lives nearby, although not in the same household.

“Parent” is used rather than “mother” or “father” because some respondents did not think they could choose between their mother and father, prompting interviewers to place a response of “both parents” under “other” in the questionnaire. Separate analyses separated out responses of “mother,” “father,” and “both parents,” but there were no differences in the substantive results of the study.

In separate analyses, I used indicators that distinguished between maternal and paternal relatives. Based on these categories, the results for caregiver type were identical to those shown in Table 3.

I tested several additional model specifications that isolated rural/urban moves in particular. The results for residential change were the same as those presented in Table 3.

Age is started at 8 rather than the mean to allow for easier interpretation of the results by referencing the first age represented in the calendar. Various polynomials were tested, but none improved model fit compared with the linear treatment of age.

Because household economic status was elicited only for the time of survey, it might not reflect the status preceding sexual debut. Measures of wealth based on accumulated assets and housing characteristics, however, generally change less readily over time than employment and income (Mberu 2006). In addition, under an assumption that an effect between sexual debut and later economic status would most likely occur through pregnancy, I conducted an ordered logit analysis with household economic status as the outcome, sexual debut as the independent variable, and ever pregnant as the mediating variable. When the pregnancy variable was added, there was virtually no attenuation of the association between sexual debut and economic status for males and females, suggesting that reverse causality is unlikely. Like most research on family instability, I cannot address with this time-invariant measure whether change in economic status is a mechanism for any observed association between caregiver change and the outcome. However, changes in economic status have not been found in empirical work in the United States to explain observed family instability effects (Osborne and McLanahan 2007; Wu 1996).

Twenty-two individuals did not provide information on their timing of sexual debut, 12 were missing the timing of parental death, five were missing their date of birth, and two were missing data on economic status. Where possible, checks indicate that individuals missing data did not have significantly different ages of sexual debut than those with no missing data.

Quadratic and cubic terms for the number of caregiver changes were not statistically significantly related to the outcome, nor did their inclusion improve model fit.

A model including both the cumulative and recency variables was tested but did not significantly improve model fit over Models 2 and 4. In addition, there was no evidence of interactive effects between the two sets of variables.

The interactions with female in Model 2 were also not jointly significant. Running separate regressions for young women and men for the models in Tables 2 and 3 also suggested minimal gender differences in the relationship between caregiver instability and the outcome. The most substantial difference was that, for young men, although the association with sexual debut was of greatest magnitude and significance in the first 1–6 months after caregiver change, there was also a significant association 7–12 months after change that did not exist for women.

None of the variance inflation factors (VIFs) for these variables exceed approximately 2, suggesting that the observed results do not simply reflect multicollinearity.

For these two tests, I used two different specifications of the orphan status indicator. The first used the indicator presented in Table 3. The second used indicators of whether maternal or paternal death had occurred in the past six months. Results were the same regardless of the measure used.

I also examined to what extent the descriptive results in Table 1 were affected by the removal of left-censored individuals. With the 34 individuals included, the results for the independent variables did not differ from those shown in Table 1, although the significance level increased for the difference between young women and men in mean age at sexual debut and in the proportion debuting at age 14 or younger. This latter finding might be anticipated, given that young men are disproportionately represented in those left-censored because they have greater representation in the left tails of the distribution for sexual debut.