Difference in Difference after Propensity Score Matching

[mckai]

Hello folks,

I am quite new to SAS, and I am trying to learn the model of Ghaly et al. (2017). In table 7, they have a model where they use Difference in Difference model after using PSM. They compare the cash holdings changes for companies based in Houston (in 2004 and 2006) with those for companies based in neighboring areas (also in 2004 and 2006). Again, they mentioned to have use PSM to generate outputs of pairs then use D-I-D, but then I am lost from here on.

I am currently using caliper matching (0.2 * standard deviation) for match Houston firms (2004) to neighboring-area firms (2004 also, because these firms also need to be matched with each other again in 2006), then I use proc genmod to get the D-I-D. The problem is that, I am not quite sure if I am heading towards the correct direction. I am not sure if proc genmod is taking the pair in my PSM step in consideration.

[SteveDenham]

It is probably likely that GENMOD isn't, but it depends on how you have coded the model statement. If I were doing this, I would use PROC GLIMMIX, and include pair as a random effect, so that the results of any tests about independent variables were conditioned on those effects.

So, to get started, could you share your GENMOD code?

SteveDenham