imbalanced sex ratios (e.g.,Lichter et al.1992).The CHFLS respondents in our sample are distributed across 37 such communities. To circumscribe the relevant pool of men by age,and to take into account the fact that the sexual behaviors that serve as dependent variables could have occurred many years before the administration of the CHFLS,we assign to each female respondent a seven-year sex ratio with a two-year staggering of the numerator(number of males)and denominator(number of females) when the respondent was age 20.This two-year staggering corresponds to the age difference between spouses in China(Porter 2006).Thus,the sex ratio is defined as the number of men ages 17 to 23 divided by the number of women ages 15 to 21.We use data from the full-count 2000 China census (China Data Center 2004)and the one-percent samples from the 1982 and 1990 censuses(China Population and Information Research Center 2008),along with standard techniques of interpolation and extrapolation,to estimate the value of this community-specific sex ratio for each female CHFLS respondent when she was age 20.2 We include several other explanatory variables in our models.Educational attainment is measured as a 6-point continuous variable ranging from never attending school(=1)to attending university or graduate school (=6).To capture age-related and/or historical trends in sexual behavior,the models include dummy variables for decadal birth cohort(1950s,1960s,and 1970s, with the 1950s serving as the reference category).We include a dummy variable for whether respondents report residing in an urban area(county-level city or larger)when they were age 14. A separate dummy variable differentiates residents of the generally more modernized South and East coast of China from other areas.Table 1 presents definitions for all the variables used in our analyses. Table 1 about here 9
9 imbalanced sex ratios (e.g., Lichter et al. 1992). The CHFLS respondents in our sample are distributed across 37 such communities. To circumscribe the relevant pool of men by age, and to take into account the fact that the sexual behaviors that serve as dependent variables could have occurred many years before the administration of the CHFLS, we assign to each female respondent a seven-year sex ratio with a two-year staggering of the numerator (number of males) and denominator (number of females) when the respondent was age 20. This two-year staggering corresponds to the age difference between spouses in China (Porter 2006). Thus, the sex ratio is defined as the number of men ages 17 to 23 divided by the number of women ages 15 to 21. We use data from the full-count 2000 China census (China Data Center 2004) and the one-percent samples from the 1982 and 1990 censuses (China Population and Information Research Center 2008), along with standard techniques of interpolation and extrapolation, to estimate the value of this community-specific sex ratio for each female CHFLS respondent when she was age 20.2 We include several other explanatory variables in our models. Educational attainment is measured as a 6-point continuous variable ranging from never attending school (= 1) to attending university or graduate school (= 6). To capture age-related and/or historical trends in sexual behavior, the models include dummy variables for decadal birth cohort (1950s, 1960s, and 1970s, with the 1950s serving as the reference category). We include a dummy variable for whether respondents report residing in an urban area (county-level city or larger) when they were age 14. A separate dummy variable differentiates residents of the generally more modernized South and East coast of China from other areas. Table 1 presents definitions for all the variables used in our analyses. Table 1 about here
Analytical strategy:We use logistic regression to examine the impact of the community-and cohort-specific sex ratio on women's sexual outcomes.Although the CHFLS respondents are nested,or clustered,within communities (as well as within single-year birth cohorts),the relatively small number of communities and the absence of hypotheses invoking cross-level interactions limits the potential utility of a complete hierarchical or multilevel modeling approach.However,we adopt the logic of the multilevel approach,and achieve one of its principal objectives,by adjusting the standard errors of the logistic regression coefficients for the clustering of observations within communities.Using STATA's cluster procedure(StataCorp 2005),we compute robust standard errors that derive from the Huber-White estimate of variance (Wooldridge 2002). RESULTS Table 1 presents(weighted)descriptive statistics for all variables used in the analysis. About 89%of the female CHFLS respondents aged 20 to 44 report having engaged in sexual intercourse in the past year.Almost 7%of the respondents report having been forced to have sex against their will at some point in their life.Nearly 17%of the respondents report having engaged in sexual intercourse outside of marriage.Of the respondents who agreed to provide a urine sample,fewer than 5%tested positive for a gonorrheal,chlamydial,or trichomoniasis infection. Descriptive statistics for the primary explanatory variable indicate that,on average,there were about 108 men aged 17 to 23 per 100 women aged 15 to 21 in the respondents' communities when these respondents were twenty years old.Thus,on average these women tended to face a surplus of men in their local community during early adulthood.Average educational attainment falls between elementary school and junior high school.Fourteen percent 10
10 Analytical strategy: We use logistic regression to examine the impact of the community- and cohort-specific sex ratio on women's sexual outcomes. Although the CHFLS respondents are nested, or clustered, within communities (as well as within single-year birth cohorts), the relatively small number of communities and the absence of hypotheses invoking cross-level interactions limits the potential utility of a complete hierarchical or multilevel modeling approach. However, we adopt the logic of the multilevel approach, and achieve one of its principal objectives, by adjusting the standard errors of the logistic regression coefficients for the clustering of observations within communities. Using STATA’s cluster procedure (StataCorp 2005), we compute robust standard errors that derive from the Huber-White estimate of variance (Wooldridge 2002). RESULTS Table 1 presents (weighted) descriptive statistics for all variables used in the analysis. About 89% of the female CHFLS respondents aged 20 to 44 report having engaged in sexual intercourse in the past year. Almost 7% of the respondents report having been forced to have sex against their will at some point in their life. Nearly 17% of the respondents report having engaged in sexual intercourse outside of marriage. Of the respondents who agreed to provide a urine sample, fewer than 5% tested positive for a gonorrheal, chlamydial, or trichomoniasis infection. Descriptive statistics for the primary explanatory variable indicate that, on average, there were about 108 men aged 17 to 23 per 100 women aged 15 to 21 in the respondents’ communities when these respondents were twenty years old. Thus, on average these women tended to face a surplus of men in their local community during early adulthood. Average educational attainment falls between elementary school and junior high school. Fourteen percent