Saturday, 12 February 2011

Driving your friends to drink

Sally Casswell's latest piece shows that if your friends are more depressed and in worse health, you're more likely to be a heavy drinker. Your friends and family can drive you to drink. Or at least that's the conclusion I drew from her correlations showing that individuals with worse health and lower reported personal well-being were more likely to report knowing heavy drinkers.

She of course frames it as heavy drinkers reducing the quality of life of those around them. But there's no way of sorting out the direction of causation's arrow.

I'd also worry a bit about selection on unobservables - a bunch of demographic covariates also affect the various well-being scores. Such results aren't reported in the paper, but I'd like to know the extent to which the correlation between knowing a heavy drinker and the well-being measures is attenuated by adding in the various covariates. If a big chunk of the correlation is explained by observable differences between cohorts, we'd have to wonder whether other unobservable things that correlate both with hanging out with heavy drinkers and having a generally poor life might explain the rest. For example, suppose that you have very poor personal characteristics such that the best mate you can attract is a heavy drinker. Is your poor reported life satisfaction due to the mate's heavy drinking? Or are your underlying characteristics, only some of which are observable, driving results?

Fortunately, TVNZ sourced its report from Reuters, who seem to know that they need to ask about causality. And so they note:
Casswell acknowledged, however, that the findings do not prove that being around a heavy drinker was the root of study participants' problems, noting that the study is just "a snapshot at one point in time".

"So...some other explanation is possible," she told Reuters, adding that people with poorer well-being may be more likely to attract heavy drinkers into their lives.
Always always ask how the researcher established causality.



  1. Judging only from the abstract, it appears they had no way of controlling for geographic proximity. This would appear to be a biggie - it is well known that problems cluster in neighbourhoods; this shouldn't surprise anyone, but for people who need formal evidence, sociologists have been documenting this at least since the 1930s. If you disregard proximity, you could also, I'm pretty certain, "show" that people's single parenthood causes their friends to be unemployed etc. etc. Even while controlling for individual SES.

    I'm not at all against properly labeled explorative research, but there comes a point at which research, which costs resources, becomes almost useless.

  2. Completely understood dude, I sent my friends a text says, jeez I depressed, come round I bought in this booze for you, and nobody came.