Calculating risk difference with covariate

Question

How do I calculate a risk difference for a binary outcome that takes into account a covariate? I want to compare accuracy of a skilled and unskilled tester at detecting 7 different types of tumors. Normally I would use a logistic regression with correct identification (1,0) as the outcome, tester as the predictor of interest, and tumor type as the covariate being adjusted for. However, the PI on this project wants a difference as the output instead of a ratio. How do I calculate that in R?

Answer 1

The thing with risk differences is that they can lead to errors in the typical GLM implementations (see here). One suggestion in the paper I link is to instead use OLS to estimate the risk differences and then use the bootstrap to get valid standard errors to compute confidence intervals.