# Why could pseudo $R$-squared (pseudo $R^2$) increase when I remove variables?

I have a multinomial logistic regression with $11$ independent variables. When I remove variables, the pseudo $R$-squared increases.

Isn't this not supposed to happen? Why could this be happening?

• Can you specify precisely what you mean by pseudo-$R^2$? I've seen a few different quantities called that. Jul 21 '16 at 15:44
• See stats.stackexchange.com/questions/3559/… for different defs Apr 5 '19 at 20:28