LPM or OLS in survey

[Oskar Lindgren]

I'm doing a study on policy acceptance using a large-scale survey across countries. As my dependent variable I use policy acceptance (measured on a 7-point scale) and I want to assess the effect of various determinants on policy acceptance (e.g. sociodemographic factors, political trust, ideological orientation etc) in both single and multiple models. Is it preferable to use LPM or OLS for this? Although I see many studies using OLS, I was told in a PhD course that LPMs are better for interpretation.

[George Ford]

LPM (linear probability model) is OLS.

If you have a categorical DV, then you need to use ordered probit or logit.

[Bruce Weaver]

Adding to George's helpful response, LPM = OLS with a dichotomous (0/1) outcome variable. And IIRC, the vce(robust) option is often recommended. So it would not be at all suitable for your 7-point (ordinal) DV. For info about the other models George suggested:

help ologit
help oprobit