Figuring out model/technique to study change in law on health outcomes in India

[tobias haeberli]

Dear everyone

This question might not be technical enough for this forum. There are some other posts about whether DID or RDD makes more sense, so I thought it might be ok. If it is not, I understand and will delete this post.

As a study project, using Stata, I'm trying to figure out the effect alcohol prohibition in Kerala state had on health outcomes. I did some preliminary descriptive statistics with Stata but don't really know how to proceed. I'm mainly struggling to find the right model or technique to actually analyse the data I have.

Where I stand:
I have survey data (not panel data) on individuals and households from 2005/2006 and 2015/2016. The law was introduced in 2014 and then (slowly) implemented before being repealed in 2017.

Two ideas on how I could study the effects I had so far:

• I could use a DID approach to studying the effect of the prohibition or maybe construct a synthetic control out of the data on the other states.

I'm not quite sure if DID would make sense, especially since it assumes that the other (neighbouring) states have a common trend. I'm also not sure whether it is reasonable to regard the introduction of prohibition as a natural experiment, but many studies of DID look at policy changes. Could a synthetic control be a favourable option or are confounders not such a large concern?

• Kerala state has a relatively large Muslim population that hardly drinks (about 0.1-0.2%, according to the survey data). So my idea is that you could assume that Muslims were not affected (in terms of direct health outcomes) by the prohibition. This way I could maybe use Muslim observations to control for 'within-state' trends and use observations from non-muslims in states other than Kerala to control for 'nation-wide' trends in the drinking population.

I'm not certain what technique would allow me to use both of these controls, or whether doing so even makes sense. I'm also uncertain how bad it is to use a specific religion as a control group since I can then no longer control for religion in e.g. a linear regression (I think).

I'd be very grateful for any advice anyone could give me or literature any literature you could point me to.

Best regards,
Tobias

[tobias haeberli]

Some example code for the first 100 observations:

* Example generated by -dataex-. To install: ssc install dataex
clear
input float(stateid alcohol religion)
25 0 1
25 0 1
25 0 1
25 0 1
25 0 1
25 0 2
25 0 1
25 0 1
25 0 1
25 0 1
25 0 2
25 0 1
25 0 1
25 0 3
25 0 3
25 0 3
25 0 3
25 0 1
25 0 3
25 0 2
25 0 1
25 0 1
25 0 1
25 0 1
25 0 1
25 0 1
25 0 1
25 0 1
25 0 1
25 0 1
25 0 1
25 0 1
25 0 1
25 0 1
25 0 1
25 0 1
25 0 1
25 0 1
25 0 1
25 0 1
25 0 3
25 0 1
25 0 1
25 0 1
25 0 1
25 0 1
25 0 3
25 1 1
25 0 1
25 0 1
25 0 1
25 0 1
25 0 1
25 0 1
25 0 1
25 0 1
25 0 1
25 0 1
25 0 1
25 0 3
25 0 3
25 0 3
25 0 1
25 0 1
25 0 1
25 0 1
25 0 1
25 0 1
25 0 1
25 0 1
25 0 1
25 0 1
25 0 3
25 1 1
25 0 1
25 0 1
25 1 1
25 0 1
25 0 1
25 0 1
25 0 1
25 0 1
25 0 1
25 0 1
25 0 1
25 0 1
25 0 1
25 0 1
25 0 1
25 0 3
25 0 1
25 0 1
25 1 1
25 0 1
25 0 1
25 0 1
25 0 1
25 0 1
25 0 1
25 0 1
end