Direct and Indirect Effects of Grandparent SES on Grandchildren’s Cognitive Development: A Prospective Mediation Analysis

Tuesday, 17 July 2018: 17:30
Oral Presentation
Markus KLEIN, University of Strathclyde, United Kingdom
Michael KÜHHIRT, University of Cologne, Germany
Using the British Cohort Study 1970, we investigate whether grandparents’ SES affects their grandchildren’s cognitive development solely through parental SES or also through direct pathways such as financial aid or caregiving. Assessing the contributions of resources and behaviours of different generations on children’s life course outcomes potentially faces severe methodological challenges. More specifically, estimating the direct effect of grandparent SES and its indirect effect through parents’ SES may be plagued by two issues of endogenous selection bias (Elwert & Winship, 2014). Firstly, adjusting for a mediator (parental SES) via the traditional approaches to mediation analysis may induce spurious association between exposure (grandparents’ SES) and outcome through unmeasured common causes of mediator and outcome. Secondly, conventional approaches ignore family formation and childless individuals, thereby potentially introducing selection bias into estimates of how characteristics in one generation affect an outcome in the next.

Our contribution to the literature is threefold. Firstly, we integrate the literature on social stratification and child development and consider multigenerational SES effects on children’s cognitive development. Secondly, we use a novel approach to mediation analysis (VanderWeele 2016) addressing the shortcomings of the traditional approach and allowing for effect decomposition in the presence of an exposure (grandparent SES)-mediator (parent SES) interaction using a potential outcomes framework. Thirdly, we follow the prospective approach on intergenerational social mobility (Breen & Lawrence, 2016) and correct for grandparent and parent SES effects on having a child in the parent generation by estimating marginal structural models with inverse probability of treatment weighting.

The BCS70 age 34 wave (in 2004) randomly selected half of the cohort members who lived with their children for an additional battery of questions about their children. Hence, we have information on children’s cognitive ability and SES information for parents (at various ages) and grandparents (measured when parents were aged 10).