An exploration of childhood antecedents of female adultmalaise in two British birth cohorts: Combining Bayesianmodel averaging and recursive partitioning

We use information from two prospective British birth cohort studies to explorethe antecedents of adult malaise, an indicator of incipient depression. Thesestudies include a wealth of information on childhood circumstances, behaviour,test scores and family background, measured several times during childhood.We are concerned both with incorporating model uncertainty and using aperson-centred approach. We explore associations in both cohorts using twoseparate approaches: Bayesian model averaging (BMA) and recursive trees. Thefirst approach permits us to assess model uncertainty, necessary because manychildhood antecedents are highly correlated. BMA also aims to produce morerobust results for extrapolation to other data sets through averaging over therange of plausible models. The second approach is concerned with partitioningthe sample, through a series of binary splits, into groups of people who are asalike as possible. One advantage is that the approach is person-centred in that itretains real groups of respondents. We compare and contrast the insightsobtained from the two approaches and use the results from each to inform theother and thus refine our understanding further. Moreover, we explore theclaimed added robustness for extrapolation by using a split-sample for the 1970cohort. The consistency of results across methods and cohorts is discussedthroughout.[...]