Pseudo R2

[Carla Hope]

HI,

I'd be grateful if you could help me clarify this please. I have performed 2 mutlivariate logsitic regressions with a slight change in the variables used as I had mutliple variables for socioeconomic variables with adequate data. Both models make equal sense clinically. Can I use the pseudo R2 statistic to help chose which model is a better fit or is this just of use in linear regression? Otherwise is this just a judgement call based on my knowledge of my dataset?

Thank you,
Carla

[Carlo Lazzaro]

Carla:
I would consider -estat ic- outcome (the lower, the better) to choose between your two regression models.
As an aside, a pedantic teacher/professor/supervisor would probably highlight that:
- you should use adjusted R-square to choose between two linear regression models (by the way, Pseudo-R is not available for OLS);
- multivariate means more than one regressand, whereas you probably meant multiple (ie, with more than one predictor).

[Carla Hope]

Thanks Carlo, I did mean multiple

[David Radwin]

If the models are nested, such as one has covariates a b c and the other has covariates a b c d, you could use a Wald test, Lagrange multiplier test, or likelihood ratio test. For more on these, consult a textbook or https://stats.idre.ucla.edu/other/mu...-andor-similar.