Is "Pseudo R2"considered to be important in logistic regression? If so,what does it mean?

[Cassie Liu]

Dear statalists，
Sorry to disturb you all.
As what says in the topic——Is "Pseudo R2"considered to be important in logistic regression? If so,what does it mean?
We all know that R-square means how the equation suit for the research.The value range starts from 0 to 1,The closer that this value is to 1, the more well-suited the equation is.But in logistic regression,I wondered if it is the same with other normal regression?
My Pseudo R2 is 0.0491,does it mean good to logistic regression?
Many thanks in advance.

[Joao Santos Silva]

Dear Cassie Liu

As Arthur Goldberger noted in his 1991 book, the only important thing about the R2 is that the R2 is not important. I know he said that in a particular context, but the general idea is that the value of the R2 generally provides little useful information, and that applies at least as strongly to pseudo-R2.

Best wishes,

Joao

[Davide Consalvi]

In the pseudo-R2 you’re comparing the loglikelihood of your model with that of the trivial alternative without explanatory variables. It is important and your R2 is quite small.
to better evaluate the goodness of your model you Should do an “estat classification”.

[Carlo Lazzaro]

Cassie:
I'm probably late to the party.
As an aside to others' helpful replies, I would focus on your model specification (ie, did you include all the predictors an interactions that are necessary to give a fair and true view of the data generating process underlying your sample?).

[Richard Williams]

There is nothing that says that much of what happens in the world can't be randomly determined. If a correctly specified model yields an R^2 of .05, then so be it.

Of course, your model may not be correctly specified. If you think R^2 is implausibly low, then think about what you might change in your model. Are important variables omitted? Should squared or interaction terms be included?

Of course, you should think about these questions regardless of what R^2 is. An R^2 of .5 may look good but it might be even higher if the model was specified correctly.

[Mike Lacy]

Leaving aside the arguments about whether R2 values are useful, I'd note that the particular pseudo-R2 that Stata reports for logistic regression is arguably one of the least useful such measures. See:
DeMaris, A., 2002. Explained variance in logistic regression: A Monte Carlo study of proposed measures. Sociological Methods & Research, 31(1), pp.27-74.

[Cassie Liu]

Thank you all for providing so many useful advices about Pseudo R2,it really helps.