csdid with binary outcome

[Ahmed Saade]

Hi!
I would like to know if csdid is appropriate in the case where the dependent variable is a dummy variable. What about when the dependent variable is ordinal? Finally, how would you graphically show treatment in the case of csdid with binary outcome?

Thanks!
A.

[FernandoRios]

hi Ahmed
So, CSDID for binary dep variable is just as appropriate as Linear Probability model would be.
However, Prof Wooldrige has shown in some simulations, that it may not always produce the correct results because of the parallel lines assumption (due to the nonlinearity of the logit/probit model)
For Ordinal variable, i would say its not appropriate, unless you make the assumption the relationship across values is constant (basically, if you can explain why it could be appropriate to use a linear regression with that type of variable. In that case, I would say its not the best approach.
HTH
Fernando

[Ahmed Saade]

Many thanks Fernando! Very helpful.