Online Program

Return to main conference page

Thursday, January 11
Thu, Jan 11, 11:00 AM - 12:45 PM
Crystal Ballroom E
Difference in Differences (DiD) and Propensity Scores

Identifying causal effects from longitudinal data: a comparison of Interactive Fixed Effects and Generalised Synthetic Controls (304230)

Richard Grieve, Department of Health Services Research & Policy 
NoĆ©mi Kreif, Centre for Health Economics 
*Stephen O'Neill, National University of Ireland Galway 
Matthew Sutton, Manchester Centre for Health Economics 

Keywords: Synthetic control method, difference-in-differences, policy evaluation

Difference-in-Differences (DiD) estimation and the Synthetic Control (SC) method have been widely used to evaluate the effects of changes to health policy. However, DiD estimation relies on a parallel trends assumption that is often questionable. While the SC method can provide approximately unbiased estimates under non-parallel trends given sufficient pre-intervention data, SC estimates can be relatively inefficient. We consider two promising modelling approaches which have been largely overlooked in the health policy evaluation literature. First, including Interactive Fixed Effects (IFE) in place of the additive fixed effects commonly employed in DiD models, allows for non-parallel trends and can be more efficient than using the SC method. However, where policy effects are heterogeneous, IFE models provide biased estimates. The second approach, the Generalised Synthetic Control (GSC) method, combines insights from the literature on synthetic controls with an IFE model for the control units, permitting non-parallel trends while avoiding bias when policy effects are heterogeneous. We contrast the DiD, SC, IFE and GSC methods through a Monte Carlo simulation study and a case study.