Long-Term Changes in Attitudes Toward Premarital Sex in the United States: Reexamining the Role of Cohort Replacement

Abstract

Attitudes toward premarital sex in the United States have changed considerably since the 1970s. However, it is unclear whether these changes are due to cohort replacement, broader changes within cohorts, or a combination of the two processes. We examined within- and between-cohort changes in attitudes toward premarital sex in the United States from 1975 to 2008 using hierarchical age-period-cohort models based on data from the General Social Survey. We used a religious plausibility structure framework to examine several possible mechanisms for within- and between-cohort variation in premarital sex attitudes over time, including changes in religious participation, marital patterns, and family structure. The results provided mixed support for our hypotheses. Attitudes toward premarital sex became more permissive over time in the United States in part because of the process of cohort replacement, but the level of permissiveness peaked with the baby boomer cohorts. This cohort effect is due in part to differences in rates of religious service attendance and educational attainment. However, the overall increase in permissive attitudes toward premarital sex is also due to period effects that are not captured by the measures included in our analyses.

View correction statement:

Erratum

Acknowledgments

The authors presented a previous version of this article at the 2011 meeting of the Midwest Sociological Society in St. Louis, MO. The authors thank Mike Stern, the editor, and the reviewers for comments on previous drafts of this article.

Notes

¹In other words, we can perfectly predict age based on period and cohort. Therefore, we cannot include the continuous versions of all three variables in the same model. For example, if a respondent was interviewed in 2000 (year) and born in 1978 (cohort), then the respondent must have been 22 years old (age) at the time of the survey.

²It would be ideal to have an index of attitudes toward premarital sex, but this is the only measure that is available in the GSS from 1975 to 2008, and it is often used in studies of long-term changes in premarital sex attitudes in the United States (e.g., Harding & Jencks, 2003; Petersen & Donnenwerth, 1997; Singh, 1980; Treas, 2002).

Note. Data come from the GSS. Numbers presented are unweighted frequencies and column percentages (in parentheses).

Note. Odds ratios are x-standardized for age and unstandardized for other variables.

*p < .05. **p < .01. ***p < .001.

³We tested the parallel regression assumption for every independent variable in all three models presented in Table 2. Using a .01 alpha level (which is recommended for tests of the parallel regression assumption), the results from Brant tests indicate that the parallel regression assumption was reasonable for all three models. For the key variables, the assumption was violated in only 3 out of 117 cases, and the effects were in the same direction across cutpoint equations.

⁴The Wald tests were significant for period and cohort in all three models. The results for model 2 were: period (Wald X ²  = 51.43, df = 19, p < 0.001) and cohort (Wald X ²  = 215.12, df = 18, p < 0.001). The results for model 3 were: period (Wald X ²  = 62.56, df = 19, p < 0.001) and cohort (Wald X ²  = 114.05, df = 18, p < 0.001).

⁵Changes in coefficients across models are potentially due to differences in unobserved heterogeneity between models (Mood, 2010). Therefore, we also examined changes in y-standardized logit coefficients (based on the standard deviation of the latent Y) using the listcoef command in Stata (Long & Freese, 2006). According to these results, none of the key or control variables in models 2 and 3 helps explain the effects of age. However, the other key variables in model 2 (attendance, marriage, teens in home, and divorce) explained 3% to 24% of the coefficients for the period variables for 1977 to 1982 and 3% to 17% for 1994 to 2002. In addition, the key and control variables in model 3 explained 61% to 75% of the coefficients for the cohort variables for 1930 to 1960.

Note. N = 22,287. Data come from the General Social Survey.

Note. N = 22,287. The odds ratios for education, log family income, political ideology, newspaper, and TV hours are x-standardized. The remaining odds ratios are unstandardized.

*p < .05. **p < .01. ***p < .001.