LPM with dichotomous dependent and independent variables

[Carlos Ingeberg]

Hello Statalist, i have a question of methodological character.
I have a dataset that only consists dichotomous variables (6 independent and 1 dependent). As my dependent variable is dichotomous the statistical model with the best fit would be logistic regression (this is also what my stata handbook recommend). However, i have also read that it is possible to use a linear probability model with a dichotomous independent variable - with the advantage of getting coefficients that can be easily interpreted. Using logit i would interpret the marginal effects and the log-odds (although log-odds isnt very intuitive).

What model would you use and why?

Any advice is much appreciated!

[Joseph Coveney]

What model would you use and why?

In your case, all categorical predictors, wouldn't

regress outcome i.explanatory?, vce(robust)

and

logit outcome i.explanatory?
margins , dydx(explanatory?)

give essentially the same results? Well, except in cases of complete or quasicomplete separation (and you'd probably want to know about that).

To answer your question, I suppose that it would depend upon the context (whether risk difference is more to the point in addressing the research question). As a default, though, my preference tends toward using logistic regression and interpreting the odds ratios.

[Carlos Ingeberg]

Thanks for the reply! i think ill include both models and compare the marginal effect with the OLS coefficient

[Nick Cox]

See Battey. Cox (Sir David) and Jackson on the linear probability model: https://royalsocietypublishing.org/d...98/rsos.190067