I am running a stepwise regression using the F test as the criterion. Is there a way to explicitly set the add and drop thresholds (alpha levels) in R? The documentation does not make it clear.
The fact that this is not easy in R is a feature rather than a bug or deficiency representing advances in the science.
For example after the first step of a stepwise regression the t or f statistics computed conditional on that first step are no longer distributed according to the t or f distributions and therefore any specified alpha levels will be wrong.
Further, stepwise regression is known to give coefficient estimates that are biased away from 0 and tests on those coefficients that are generally meaningless. Sometimes stepwise regression becomes essentially an inneficient method for selecting a random subset of predictors (rather than a meaningful one). These days penalized methods such as lasso or ridge regression among others are preferred to stepwise methods.
Even better is to not do model selection if not needed.