Hi,
Have conducted a few interrupted time series analyses and am having an existential crisis about my most recent, particularly in regards to coding and outcomes.
First question relates to coding of dummy variables:
My default coding of dummy variables is to code the INTERVENTION as 0.....0 (pre) 1....1 (post) acting as the identifier of the pre- and post-intervention periods AND 0.....0 (pre), 1,2.......23, 24 (post) as the TIME SINCE INTERVENTION. However, I have since seen data coding the TIME SINCE INTERVENTION as starting at 0 rather than 1 (i.e. 0...0 (pre) 0,1,.......22, 23 (post))
I assume this won't make a difference to my outcome model, but just wondered if there was conventional wisdom as to either being the better/prefferd approach? I've seen a mix of both across the majority of literature
Second question relates to outcomes. I've used a variety of GLM models and estimation methods to derive exponentiated outcomes.
I wondered whether you would anticipate the derived value for the Level Change to equate to the difference between the post-policy modelled value at time 1 post-intervention and its corresponding counterfactual value.
This is not the case with my currenty outcomes and I wondered whether it is a poetntial issue with the exponentioation estimations or just my incorrect interpretation?
Any advice hugely appreciated