Retention rate heavily embeds student selectivity in it, so any valid comparison of it to find treatment effects needs to adjust for student characteristics, rather than use raw numbers. Of course, any such remaining variation may come from other factors besides teaching quality.