Title: A Bayesian factor analysis model for evaluating an intervention using observational panel data on multiple outcomes
Authors: Pantelis Samartsidis - University of Cambridge (United Kingdom) [presenting]
Shaun Seaman - University of Cambridge (United Kingdom)
Silvia Montagna - University of Turin (Italy)
Andre Charlett - Public Health England (United Kingdom)
Matthew Hickman - University of Bristol (United Kingdom)
Daniela de Angelis - University of Cambridge (United Kingdom)
Abstract: A problem frequently encountered in many areas of scientific research is that of estimating the impact of a non-randomised binary intervention on an outcome of interest using time-series data on units that received the intervention (treated) and units that did not (controls). One popular estimation method in this setting is based on the factor analysis (FA) model. The FA model is fitted to the pre-intervention outcome data on treated units and all the outcome data on control units, and the counterfactual treatment-free post-intervention outcomes of the former are predicted from the fitted model. Intervention effects are estimated as the observed outcomes minus these predicted counterfactual outcomes. We propose two extensions of the FA model for estimating intervention effects: 1) the joint modelling of multiple outcomes to exploit shared variability, and 2) an autoregressive structure on factors to account for temporal correlations in the outcome. Using simulation studies, we show that both extensions can improve the precision of the intervention effect estimates: the first when the number of pre-intervention measurements is small; the second when the number of control units is small. We apply our method to estimate the impact of stricter alcohollicensing policies on alcohol-related harms.