the result is a model with an intercept for each binary category and separate coefficients for each category-independent var combination, but only one R squared value. when i compare the result to just doing separate lm models for each category of data, the coefficients are very similar but slightly different