Hello ASA community,
I am working on a study which aims to analyze the effects of a policy using a difference in differences approach. We have 1.5 years of pre policy data and 1.5 years of data after policy implementation. The outcome variable is binary.
I wrote the model as = + X + *Post + *Treated + *Post*Treated where Post=0 denotes pre period, Post=1 denotes post period, Treatment=0 denotes the control group, Treatment=1 denotes the Treated group and X denotes the vector of all the other covariates.
I looked at the interaction term to get a sense of the impact of the policy. But this is a rate of odds ratios which is hard to interpret. Some colleagues have suggested that marginal effects would be a better approach( as described by Pinar Karaca-Mandic, Edward C. Norton, and Bryan Dowd Interaction Terms in Nonlinear Models. Health Research and Educational Trust DOI: 10.1111/j.1475-6773.2011.01314.x. However I have seen several high-impact papers published where they have used the interaction term from logistic regression. One example is Rajaram R, Chung JW, Jones AT, et al. Association of the 2011 ACGME resident duty hour reform with general surgery patient outcomes and with resident examination performance. JAMA. doi:10.1001/jama.2014.15277.
Which is the better approach? Or should I consider other approaches than these two?
Thank you in advance!
-----------------------------------------------------------------
Nicoleta Lupulescu-Mann, MS, Research Associate
OHSU Center for Health Systems Effectiveness
Office: 3030 SW Moody | Mail Code: MDYCHSE
P: 503-418-3604 | E: lupulesc@ohsu.edu | W: www.ohsu.edu/chse