Odds Ratios instead of Coefficients using teffects?

[Rachel Ryding]

I'm running a series of models using teffects psmatch, teffects ra, and teffects ipw and comparing against a logistic regression model, both my treatment and outcome variables are binary. I think I've specified my teffects ra model as a logit model, but unsure about the matching or inverse-probability weighting code for this (or if there is one).

I want to make sure that I'm not making an error in syntax or interpretation of output across my respective models before I get into troubleshooting my data; right now there is a sizable difference between the coefficient from the logistic regression model (1.438264) and my three models using teffects psmatch (.1641289), teffects ra (.120325), and teffects ipw (.1400092).

So my questions are: (1) is there a command to be used with teffects that specifies a binary rather than continuous outcome variable? All of the examples I could find use a continuous outcome variable in their model; and (2) are the coefficients from the teffects outputs comparable to the coefficients from a logistic regression model? Or is there a way to get the teffects output to give me an odds ratio instead of the coefficient?

Thank you

[Jorge Eduardo Perez Perez]

(1) There shouldn't be any difference when using teffects with a binary dependent variable instead of a continuous one. The usual caveats about the linear probability model would apply though, e.g. there could be predictions outside of the 0-1 range and there is heteroskedasticity to be accounted for.

(2) In principle the coefficients from -teffects- should be comparable to the coefficients from a linear probability model, but not to the coefficients from a logit. You should compare the marginal effects across the different models.

Hope this helps,

[Rachel Ryding]

Thank you! This is a helpful explanation.