Ploubidis et al. Respond.

Abstract

Listen

Translate article

Thank you for the opportunity to respond to Strohschein’s letter regarding our published article in AJPH. The main issue raised by Strohschein is that our study lacks a test of interaction for one of our findings, namely that men derive more health benefits from marriage than women. Strohschein cites a newspaper article, which appears to reference our article. Unfortunately, the newspaper article misrepresents our findings. We simply did not find and certainly did not report that marriage does not benefit women. Our findings clearly demonstrate that marriage is beneficial for both men and women. The newspaper article advises young women they should not get married; however, there is nothing whatsoever in our article that would “invite” or suggest this. Furthermore, we have never shown that married men are less likely to have metabolic syndrome compared with unmarried men. We are disappointed that Strohschein reiterates this false representation of our findings. Strohschein demonstrates a commonly held misconception, namely that interaction and effect modification are identical concepts. In fact, they are distinct, although in some settings they may coincide.1 Effect modification, which is the concept relevant to our study, is defined as the effect of an independent variable (partnership status) varying across the strata of another variable (gender), while interaction is defined as the joint effect of two variables.1 The two concepts do not coincide in our analysis, since not all paths from gender to health are blocked.1,2 Since interaction terms could not be directly interpreted as a test of modification in our analysis, and given the heterogeneity of latent partnership status trajectories, we reported analysis stratified by gender. Our interpretation and reporting of model parameters did not rely heavily on significance tests, but on actual effect size, as is increasingly recognized as best practice within statistics.3 We note however, that alternative analysis with an interaction term for gender and partnership status returned “significant” results. Additional analysis with multigroup models representing effect modification also showed a better fit to our data compared to restrictive models that assumed lack of modification. All approaches did point to the same conclusion: the presence of effect modification by gender in our data, such that men benefitted more from marriage than women. Strohschein cites other studies where the health effects of marriage are equally distributed among men and women. Unlike ours, these studies are concerned with mental health outcomes and—crucially—none of them is set in the United Kingdom. We are surprised that Strohschein perceives these findings as conflicting with ours. As stated in our article, our findings on the association between life-course partnership status and biomarkers in midlife are applicable to the 1958 birth cohort and those born close to that year in the United Kingdom. They are not generalizable in other—particularly younger—cohorts or in non-UK settings.