Toward Clearer Questions in Mendelian Randomization Studies