[Philip Gigliotti]

Here is the reference for the most current iteration of the mora and reggio paper.

Mora, R., & Reggio, I. (2017). Alternative diff-in-diffs estimators with several pre-treatment periods. Econometric Reviews, (forthcoming).

The earlier working paper (2012) entitled treatment effect identification using alternate parallel assumptions is available here.

https://e-archivo.uc3m.es/bitstream/...5CA?sequence=1

[Ludmila Farooq]

The Regression below is a caption taken from Angrist and Pischke (2015)
In their book they write: " This model presumes that in the absense of treatment effect, death rates in state k deviate from common year effecs by following the linear trend captured by the coefficient Theta "

I want to keep my regression simple. my question now is :
Is it enugh to just make and interaction between time variable and in my case department id and add it in the regression ?

[Philip Gigliotti]

The Regression below is a caption taken from Angrist and Pischke (2015)
In their book they write: " This model presumes that in the absense of treatment effect, death rates in state k deviate from common year effecs by following the linear trend captured by the coefficient Theta "

I want to keep my regression simple. my question now is :
Is it enugh to just make and interaction between time variable and in my case department id and add it in the regression ?
[ATTACH=CONFIG]n1409526[/ATTACH]

In a two way fixed effect model, the summation of department dummies, beta, will be absorbed by the department fixed effect. The summation of year dummies, gamma, is your vector of year fixed effects, so the only vector you would need to include is theta and possibly delta, though I'm not sure what delta represents. If your model lacks department fixed effects, then you should specify the model exactly as seen in the book.

[Ludmila Farooq]

@Clyde Schechter

Does it matter if the pretreatment time is only one year while post treatment time is 11 years ?

[Philip Gigliotti]

@Clyde Schechter

Does it matter if the pretreatment time is only one year while post treatment time is 11 years ?

Yes, it does matter. Without 2 or more pre treatment periods you can't test the parallel trends assumption. A 1 period pre-treatment is still a diff in diff, but the comparison may be meaningless. What if the one pre treatment year happened to have above average and or below average values for your dependent variable. You are only comparing against one year, rather than an established pre treatment trend, so your inferences will be questionable. 1 pre period diff in diffs are probably going to be difficult to publish.

[Abiodun Salako]

Clyde Schechter Philip Gigliotti

I need some clarification about adding state-specific quadratic trends to a difference-in-differences model. If I have the below difference-in-differences model:

reg i.statepolicy covariates i.state i.year

where statepolicy is 1 for each year in which a policy is in effect in a state and 0 otherwise (Note: states began implementation of the policy in different years)

To add state-specific quadratic trends to the model do I do (A) or (B)

(A)

reg i.statepolicy covariates i.state i.year state#c.linearyear state#c.linearyear#c.linearyear

(B)

reg i.statepolicy covariates i.state i.year state#c.linearyear#c.linearyear

where linearyear is coded 1 for the first year in the study period, 2 for the 2nd year and so on.

[Clyde Schechter]

Code (A) is correct. Code B is not because it produces only a quadratic term, with no linear term for linearyear. However, (A) can be shortened and simplified:

reg i.statepolicy covariates i.year i.state##c.linearyear##c.linearyear

Note the use of the ## operator, which is different from the # operator.

Added: You tacked your question on to an old thread. Although both the old thread and your question involve difference-in-differences models, the real question you are asking concerns how to code quadratic effects. It would be better, in the future, to start a New Topic on the forum when asking a question that is different. People come to the Forum to search for particular topics. Somebody looking for advice on quadratics will never find this post. And others coming for general advice on DID modeling will waste their time reading it. So it is important to keep threads on topic. When you have a question that is only tangentially related to an existing thread, it is better to start a new one (and give it an informative title that will facilitate accurate searching.)