This doesn't seem to clearly identify anywhere what it is that is being measured; everything is based on a "latent factor for consumtion of alcohol" that is said to be standardized, but never given a proper definition. The closest I could find was:<p>"Indicators of alcohol consumption in the Add Health data include the frequency of binge drinking (drinking five or more units of alcohol in one sitting) and the frequency of getting drunk."<p>That doesn't say if those are the only factors used, or if they mixed in several different data sets until they found a combination that made for a compelling graph. I'll give them credit for mentioning that they've controlled for income though, as that was my first assumption.<p>I would be interested to know if the skew is really in total alcohol consumption, or if one side tends to binge occasionally while the other gets lightly buzzed frequently, or if there is a different split entirely that this is glossing over.