They don't. What they can do is observe cancer rates in populations that have 1 drink per day and compare that to rates of cancer in other populations. There are a bunch of statistical tools you can use to try to identify and correct for other confounding variables.