A note on variance reduction