topic Re: difference in rate differences in Statistical Procedures

difference in rate differences

GreenTree1 — Fri, 07 Feb 2020 22:04:02 GMT

Hello everyone,

I am trying to do a DID analysis and difference in rate difference, where the rate is death rate (no of deaths/total number of infections) where death= 1/0

before and after an introduction of a vaccine (timeperiod= 0 is before and 1 is after), where vaccine= 0/1 for two different counties A and B. For county B, the vaccine was never introduced.

I have tried looking this up but most of the examples are based on differences in means. Also since my outcome of interest is death, the population is not the same across two the time periods.

Below is how my dataset looks like

county death infection vaccine timeperiod

A . 1 . 1 . 1 1
B . 1 . 1 0 1
A . 0 . 1 1 1
A . 0 . 1 1 1
B . 0 . 1 0 1
A . 1 . 1 0 0
B 1 1 0 0

Please let me know if more I can clarify this further.

Appreciate your help

Re: difference in rate differences

GreenTree1 — Sun, 09 Feb 2020 15:50:07 GMT

Post update: proc genmod worked

proc genmod;
class a b / ref=first;
model y = a b a*b;
estimate "Diff in Diff" a*b 1 -1 -1 1;
lsmeans a*b;
lsmestimate a*b "Diff in Diff" 1 -1 -1 1;
run;

I had the interaction term mixed up.

Re: difference in rate differences

StatDave — Mon, 17 Feb 2020 18:58:07 GMT

Your response mean is a probability (limited to [0,1] range), not a rate that can exceed 1. To properly model a probability, you should either use PROC LOGISTIC or specify the DIST=BINOMIAL option in the MODEL statement in PROC GENMOD. If you want to estimate the difference in difference on the proportion scale, then you should use one of the methods discussed in this note.

Re: difference in rate differences

GreenTree1 — Wed, 19 Feb 2020 16:55:50 GMT

Thanks for pointing that Statdave. It makes sense to use dist=binomial in proc genmod to model probability.