Social Forces

In an earlier paper, we deployed the European Social Survey (ESS) and the General Social Survey (GSS) to conduct the largest analysis to date examining the question of whether child sex affects parent political orientation. We found null effects in contrast to earlier, smaller studies. In the current issue, Hopcroft (2016) argues that our null findings may have been obtained due to sample restrictions and measurement error arising from the fact that we used the sex of the first child “residing” in home rather than the sex of the first “biological” child.

We believe that in their comment, Hopcroft largely restates the limitations we have already discussed in our original manuscript, adding only details of the GSS and ESS codebook for the readers. More importantly, she has confused identification issues (e.g., measurement error) with inference issues (e.g., sample restrictions) that we discussed in detail in three pages (Lee and Conley 2016, 1112–14). The goal of our systematic sample selection was to reduce the potential attenuation bias arising from measurement error. Further, by ignoring the period/country variations we showed, this critic missed one of the main points of our paper—we asked why we might observe contradictory findings in the UK and the United States in the first place. We concluded that such results are more likely due to publication bias (or possibly period heterogeneity) rather than to treatment effect heterogeneities or country differences. Nevertheless, we are open to the possibility that we made mistakes in our original paper. In the present response, we have decided to play devil’s advocate by taking the opportunity to revisit our case.

There seems to be a straightforward way to measure the sex of the first child; one can ask respondents about the sex/age and biological status of all children they have ever had and infer the sex of the first child from the resulting roster. [End Page 899] This is how the 1994 GSS collected the relevant information (Conley and Rauscher 2013). Despite the possibility of coding errors, interviewer bias, respondent fatigue, and recall bias, one can assume that this approach is by far the gold standard by which one can measure the sex of the first child. By contrast, in our paper, we used the information about the sex of all children who are still living in the household, since a complete fertility roster was not available. We tried to minimize measurement error bias as best we could, by including in the sample only those parents whose oldest child is under seventeen in the ESS and the GSS data, following the strategy of prior studies (e.g., Oswald and Powdthavee 2010). Additionally, in the GSS data, we attempted to capture all children each respondent has ever had only by including those who have the same number of kids they have in their life and in their household (though this strategy was not available in the entire ESS data) – thus lowering the chances that we were picking up stepchildren.

In this response, we focus on the 1994 GSS data to show the actual amount of measurement error by comparing the bio sample (that was only available in the 1994 GSS data) and the cohab sample (which we called the “analytical sample” in the original report), and its likely consequences for the inferences we made. Moreover, we revisit the second round of ESS data (years 2005–2006) that asked respondents if there is any child who does not live together with the respondents, and we check how the additional exclusion of those respondents might affect our original conclusions. Additionally, we show the effects of having biological versus adopted/stepchildren on parental political orientation. Finally, we simulate measurement error processes by extrapolating the predicted equation for measurement bias in the 1994 GSS data to other periods and show whether the model-based correction of measurement error can alter the temporally-mixed patterns we reported in figure 2 of the original report.