1

As seen below from Piet De Jong, Generalized Linear Models for Insurance Data, for linear model, we can identify high leverage, if the value of leverage>2p/n (or hii > 2p/n) then the indicates the case is unusual in the space of explanatory variables.

enter image description here

enter image description here

If I have a Logit Regression/Logistic Regression, does anyone know can I use the rule of thumb that $h_{ii}>2p/n$ to identify large leverage? Or does anyone have references for the rule of thumb for $h_{ii}$ (to identify which case is unusual in the space of explanatory variable) in logistic regression?

kjetil b halvorsen
  • 63,378
  • 26
  • 142
  • 467
Jasmine Helen
  • 23
  • 3
  • Relevant (not a duplicate): https://stats.stackexchange.com/questions/81058/how-to-handle-leverage-values – kjetil b halvorsen Dec 21 '21 at 16:42
  • Is your goal to identify the influential observations/outliers for the model? The diagnostic tools are less informative for logistic regression, because it is harder for an individual observation to have a great impact (unlike linear regression). – user344849 Dec 21 '21 at 16:43
  • However, you may still use something like cook distance that measures the standardized change in the linear predictor of your logistic model, when the ith case is deleted. – user344849 Dec 21 '21 at 16:44
  • But I am not sure if the rule of thumb for leverage in the linear regression setting is still applicable here, although you can also get such leverage value from the linear predictor of your logistic model – user344849 Dec 21 '21 at 16:47

0 Answers0