Difference in differences method with panel data. (4 years of time period)

[Yangjun Park]

Hi,

I'm studying on health insurance coverage policy effect using Difference in differences method with panel data.
The aim of policy was to increase insured's coverage reducing out of pocket spending directly. This was introduced and implemented from June 2013 to Dec 2016.
I've read lots of similar article and they normally set a year as a time interval. e.g. intervention(policy implemented 2013, before time 2012 and after time 2014)
However, my study would be different because the policy period was 4years(2013, 2014, 2015, 2016).
Would that be fine to set 4 years of time period? and evaluate the out of pocket spending before time point 2012 and after time point 2017.

Thank you.

[Jeff Wooldridge]

So you have annual data, not monthly data? It will be tricky if you don't have a pre-treatment period (such as 2012), but you can do something. It's hard to incorporate units always treated -- that is, starting in 2013 -- and so it's easiest to drop them. Then you'll have three treated cohorts, 2014, 2015, 2016. You need to define a variable, say first_treat, that is either zero (never treated) or the year of the first treatment. Then use xthdidregress or jwdid (user-written) or csdid (also user-written).

[Yangjun Park]

Thank you for replying my post.

I have annual data. As I mentioned, the policy started in June 2013 and ended in December 2016. The analysis I intend to conduct involves using the Difference in Differences method to compare the out-of-pocket expenses in 2012 (before the policy) and 2017 (after the policy) between the treatment group and the control group.
In other words, the data from 2013, 2014, 2015, and 2016 will not be used. I am wondering if it is appropriate to use a pre-post research model with a four-year gap.

[Yangjun Park]

So you have annual data, not monthly data? It will be tricky if you don't have a pre-treatment period (such as 2012), but you can do something. It's hard to incorporate units always treated -- that is, starting in 2013 -- and so it's easiest to drop them. Then you'll have three treated cohorts, 2014, 2015, 2016. You need to define a variable, say first_treat, that is either zero (never treated) or the year of the first treatment. Then use xthdidregress or jwdid (user-written) or csdid (also user-written).

Thank you for replying my post.

I have annual data. As I mentioned, the policy started in June 2013 and ended in December 2016. The analysis I intend to conduct involves using the Difference in Differences method to compare the out-of-pocket expenses in 2012 (before the policy) and 2017 (after the policy) between the treatment group and the control group.
In other words, the data from 2013, 2014, 2015, and 2016 will not be used. I am wondering if it is appropriate to use a pre-post research model with a four-year gap.