2

I am using the RMS package in R to conduct a logistic regression that contains a three-way interaction. As part of my modelling approach, I have conducted chunk tests of the interaction (using Wald chi-square) followed by odds ratios for significant interactions. I have been asked by a reviewer to provide betas and standard errors for all factors in the model. I know how to obtain beta estimates for different levels of an interaction, but I can't find a way to obtain a single beta to represent the entire three-way interaction term. Is there any way of doing this with an lrm object?

EDIT:

Model is predicting police outcome (court/no court) Independent variables are ethnicity (binary), age (continuous), sex (binary), previous violent charge (3 categories) and number of charges (3 categories).

Model

model <- fit.mult.impute(`Outcome`~ rcs(Age,5) + `Ethnicity` + 
        `Violence` + Sex + `N_Charges` + rcs(Age,5)*`Ethnicity` + 
         rcs(Age,5)*Sex + `Ethnicity`*`N_Charges`+ 
        `Ethnicity`*`Violence` + **rcs(Age,5)`*Ethnicity`*Sex**, 
         x=TRUE, y=TRUE, fitter = lrm, xtrans = imp)

I am interested in obtaining the beta for the bolded (wihin ** **) interaction term.

kjetil b halvorsen
  • 63,378
  • 26
  • 142
  • 467
user183974
  • 361
  • 1
  • 15

1 Answers1

1

If you by the beta means one coefficient value for the interaction term, you are out of luck. That is a multiple-df term, so has multiple betas. But with a chunk test you can get one p-value for the term.

For chunk tests see What are chunk tests?

kjetil b halvorsen
  • 63,378
  • 26
  • 142
  • 467