Since many later-life outcomes (education, marriage, job market performance) directly depend on the behaviour during adolescence, it is important for good policy design to understand the comparative role of internal motivations, such as feelings of shame possibly implanted by parents, and external motivations, such as peer-group effects, on the sexual decisions of teenagers.
In a recent CEPR Discussion Paper (Fernández-Villaverde et al 2011), we use data from the National Longitudinal Study of Adolescent Health ("Add Health") to address this question. Add Health is a representative sample of about 90,000 adolescents in the US who were in grades seven to twelve at 134 junior- and senior-high schools during the 1994–95 school year. The respondents have been followed in four waves, although for our purposes only the first two (1994 and 1996) are needed. Add Health is well suited for the study of social interactions because it contains information about sexual behaviour, sexual knowledge, shame from premarital sex or pregnancy, religiosity, parental background, school characteristics, etc. Furthermore, there are observations for many students from the same school and respondents are asked to identify their friends from the sample. Thus, peer groups can be constructed from either students who attend the same school or from groups of friends.
So, using Add Health data, which factors do we find affect the chances a teenage girl will engage in premarital sex?
To investigate this, we consider a logistic regression where the independent variable is whether a teenage girl starts having sex between Waves I and II. As explanatory variables we have a set of controls (which includes variables reported in Add Health related to stigma, religion, family background, etc,) and the fraction of the respondent's peers, teenage girls from the same school, who have already had premarital sex in Wave I. The logistic regression and the panel structure of the data set overcome the ‘reflection problem’ prevalent in cross-sectional samples. Group behaviour affects individual behaviour but the group by definition is the sum of the individuals (Manski 1993; Brock and Durlauf 2001a and 2001b). In particular, our analysis focuses on those girls who made the transition from never having premarital sex in Wave I to having had it in Wave II. Therefore, this subset of girls could not have influenced those who had sex in Wave I, which is the peer group that can be taken as exogenous in the regression. Clark and Loheac (2007) use a similar approach to study teenage consumption of alcohol, marijuana, and tobacco, and Patacchini and Zenou (2011) to investigate the transmission of religiosity from parents to children.
Our shame variable from Add Health contains several different variables both on the shame from sex and on the religiosity of the teenager. Since many ofthese variables are correlated, it is not desirable to use them all. Instead, we employ factor analysis to consolidate the variables into a single one, called "shame". The basic idea is that shame is a common factor which affects a respondent's answers. In the factor analysis, eleven variables are used related to the perceived shame a teenage girl would feel from her mother or family regarding premarital sex; the personal shame/concern the teenager would have about sex; her shame/concern about a pregnancy; and the girl's religiosity. These are then statistically aggregated; the resulting single factor explains about 50% of the variation in these 11 variables.
In a simple specification where the only explanatory variables are shame and the fraction of teenagers who have already had sex in the respondent's school in Wave I, both the peer group and shame variables have significant effects on teenage initiation of premarital sex. Teenagers are more likely to start having sex if they have a large group of peers who have already had sex and they are less likely to have sex if they are ashamed of it. This result is surprisingly robust to the inclusion of controls such as parental income in Wave I, whether the respondent has a romantic relationship in Wave II, her grades, whether the teenager believes that she looks older than her peers, maternal education, maternal religiosity, whether the respondent lives with two parents, whether she has an older sibling, whether she received sex or AIDS education at school, whether her parents are satisfied with their relationship with the girl, how much parents talk about sex with her, and whether the teenager works and has an independent source of income.
A simple way to report the quantitative size of our estimates is to calculate the marginal effects, in terms of semi-elasticities, of changes in the shame and peer-group mean. A shift in the shame variable by 1% leads to a change in the odds of having premarital sex of 0.023 percentage points. Similarly, a movement in the school average by 1% adjusts the probability of engaging in premarital sex by 0.068 percentage points. This implies that a change of one standard deviation in the average of peers who had premarital sexwould cause a shift of 2.5 percentage points in the odds of engaging in premarital sex, while a shift of one standard deviation in the shame factor is associated with a movement of 5.3 percentage points in this probability.
A potential problem in identifying peer-group effects is the presence of correlated effects. Peer groups may not be formed randomly. Perhaps a teenager associates with her peers because they have similar unobserved characteristics. If one is interested in having sexual relationships then one may choose to associate with others who share this predisposition – see Greenwood and Guner (2010) for a model of this. In our analysis, peer groups are based on a teenager's school, assuming that the school is exogenous to the individual and parental characteristics of the respondent. Since school choice by parents might be partly endogenous, we consider some further controls.
First, Add Heath asks parents if they chose a particular neighbourhood for school quality and we can use this information as an additional control. Second, it is possible to differentiate between those teenagers who have moved to a neighbourhood recently (within a year) and those who have been living there longer (the idea to differentiate between recent and not-so-recent movers was initially suggested in Gaviria and Raphael 2001). Since correlated effects are likely to be different for these two groups, one can also control for whether the subject is a recent mover. Third, teenagers within a particular school might behave in a similar way since they face the same institutional constraints. To control for this, we add further regressors related to school characteristics.
Our results are remarkably robust. In all these additional exercises the magnitudes of peer-group effects and shame are not affected in a material way, although the peer group effect is often significant only at the 10% level. With respect to the additional controls, none of these variables turned out to be significant and they did not affect the basic findings.
Add Health data strongly suggest that shame is an important driver of sexual behaviour among teenagers even when peer-group effects are considered. Furthermore, ignoring internal motivations such as shame may yield too large estimates of peer-group effects, potentially biasing the design of policy.
Brock, WA and SN Durlauf (2001a), "Discrete Choice with Social Interactions", Review of Economic Studies 68(2): 235-60.
Brock, WA and SN Durlauf (2001b), "Interactions-based Models," in Heckman, James J and Edward E Leamer (eds), Handbook of Econometrics, 5. Amsterdam: Elsevier.
Clark, AE and Y Loheac (2007), "It wasn't me, it was them! Social Influences in Risky Behavior by Adolescents", Journal of Health Economics 26(4): 763-84.
Fernández-Villaverde, J, J Greenwood and N Guner (2011), "From Shame to Game in One Hundred Years: A Macroeconomic Model of the Rise in Premarital Sex and its De-Stigmatization”, CEPR Discussion Paper 8667.
Gaviria, A, and S Raphael (2001), "School-based Peer Effects and Juvenile Behavior", Review of Economics and Statistics 83(4): 257-68.
Greenwood, J and N Guner (2010), "Social Change: The Sexual Revolution", International Economic Review 51(4): 893-923.
Manski, CF (1993), "Identification of Endogenous Social Effects: The Reflection Problem", Review of Economic Studies 60(3): 531-42.
Patacchini, E and Y Zenou (2011). "Social Networks and Parental Behavior in the Intergenerational Transmission of Religion", CEPR Discussion Paper 8443.