Especially with more cutting edge and Not Machine Learning based statistics, R has packages where Python does not. I've found SciPy/NumPy and statsmodels to be an adequate "What if I need a GLM in my Python code?" solution, but less of a great solution to heavy stats work.