I picked up the folders for the two courses required of every student at the school. Statistics and epidemiology. Epi—what?
In the first lecture, we ‘reviewed’ all the major study types. For example, in the case-control study you find a group of people with a disease, and then look for people who are much the same but without the disease. You compare the two groups to see if they have different risks. It’s a relatively cheap method, but it doesn’t tell you much about the order in which things happen. I can’t remember all the examples used in the lecture, but let’s say you want to look at causes of depression in women. You start with 600 depressed women, find another 600 who match them in age, ethnicity and educational status, and then ask them all about their lives. Let’s say you find out that women who are depressed are six times more likely not to have had sex in the last year as women who are cheerful. That means if you’re not having sex you get depressed, right? But hang on, couldn’t it be that women who are moping around looking miserable don’t get laid much?
Perhaps you’d be better off with a cohort study. You start off with several thousand women who are perfectly happy. Then you follow them over time, recording their behaviours, and see which of them get depressed. If you find that women who have sex are less likely to become depressed than women who aren’t getting any, it suggests it is the lack of sex that causes the depression, not the depression which stops you getting laid. You can throw out the ‘misery guts’ theory and recommend more good sex as an intervention to promote mental health.
Chronology is evidence of causality, but it’s weak evidence. In this case, there are (at least) two problems. First, there could be some other factor (disruption of social network? increase in pro-inflamatory microbiota?) which causes both, but the sex is caused faster. Alternatively, it could be that depression causes low sex drive, but that kicks in immediately whereas it takes months to get a depression diagnosis.
There are good ways to determine causality from observational data, but timing isn’t one of them.
Hm? What Pisani is pointing out here is that of the 3 major causal patterns that a cross-sectional correlation can reflect, A->B, B->A and A<-C->B, a longitudinal correlation in which observations of A are followed by observations of B, will let you rule out 1 of the 3 patterns (B->A), reverse causation, which leaves either the hypothesized direct causation or confounding. This is much better evidence than just the cross-sectional approach, although I think confounding is much more likely in general so the boost is not as big as the trichotomy makes it sound.
Reverse causation is not ruled out because diagnosis can be delayed.
It seems entirely plausible to me that it takes several months of worsening depression symptoms (during which time sex drive is effected) before a patient sees a psychiatrist.
I suppose it’s ruled out if we separate “depression” and “diagnosed with depression” into separate nodes, but that doesn’t rule out anything interesting.
-- Elizabeth Pisani, The Wisdom of Whores, p. 16
Chronology is evidence of causality, but it’s weak evidence. In this case, there are (at least) two problems. First, there could be some other factor (disruption of social network? increase in pro-inflamatory microbiota?) which causes both, but the sex is caused faster. Alternatively, it could be that depression causes low sex drive, but that kicks in immediately whereas it takes months to get a depression diagnosis.
There are good ways to determine causality from observational data, but timing isn’t one of them.
Hm? What Pisani is pointing out here is that of the 3 major causal patterns that a cross-sectional correlation can reflect, A->B, B->A and A<-C->B, a longitudinal correlation in which observations of A are followed by observations of B, will let you rule out 1 of the 3 patterns (B->A), reverse causation, which leaves either the hypothesized direct causation or confounding. This is much better evidence than just the cross-sectional approach, although I think confounding is much more likely in general so the boost is not as big as the trichotomy makes it sound.
Reverse causation is not ruled out because diagnosis can be delayed.
It seems entirely plausible to me that it takes several months of worsening depression symptoms (during which time sex drive is effected) before a patient sees a psychiatrist.
I suppose it’s ruled out if we separate “depression” and “diagnosed with depression” into separate nodes, but that doesn’t rule out anything interesting.