Key to causal inference from observational data is adequate adjustment for confounders (i.e., factors which cause both the exposure and outcome). However, knowing whether a variable is a confounder (requiring statistical adjustment) or a mediator (i.e., a variable caused by the exposure which in turn causes the outcome; not requiring statistical adjustment) is often difficult to establish with certainty and often relies upon potentially-debatable assumptions. By making use of the repeated data collected by ALSPAC - both in terms of religiosity and data on other covariates which may be plausible confounders and/or mediators - it is possible to make reasonable inferences as to whether religiosity causes the covariate, the covariate causes religiosity, or indeed whether there is bidirectional causation (i.e., religiosity causes the covariate and the covariate also causes religiosity). If data permit adjustment for baseline confounders, prior exposure and prior outcomes, then it may be possible to infer causality using longitudinal observational data (VanderWeele 2021; VanderWeele et al., 2016). We intend to apply these methods in ALSPAC to explore potential bidirectional causation between religiosity and a range of covariates to better understand these patterns and to help inform future work using these data.

References:
VanderWeele, T. J. (2021). Can sophisticated study designs with regression analyses of observational data provide causal inferences?. JAMA psychiatry, 78(3), 244-246.
VanderWeele, T. J., Jackson, J. W., & Li, S. (2016). Causal inference and longitudinal data: a case study of religion and mental health. Social psychiatry and psychiatric epidemiology, 51(11), 1457-1466.