Suppose that one-third of patients have some gene that makes them respond to Prozac with an effect size of 1.0 (very large and impressive), and nobody else responds. In a randomized controlled trial of Prozac, the average effect size will show up as 0.33 (one-third of patients get effect size of 1, two-thirds get effect size of 0).
Economists instinctively fall back on the “representative agent” model, in which you average the population results of whatever study you do. So an economist would say that the effect size is 0.33. But the point is that there is not one parameter that represents the whole population. One needs to take into account differences.
Where this bothers me the most is in the realm of expectations. Someone will take a survey of, say, consumer expectations for home price increases. The results will diverge across consumers. But the economist will report a single number for consumer expectations.
This shows that statistical medicine benefits governments and authorities, but has limited use for individual patients.
What is needed is coming from Silicon Valley – Genomic individualised medicine.
Although many established in authority will discourage it as much as possible, it will come eventually because there are obvious economic advantages to tax funded and insurance funded medicine in not giving people medicines or other treatments that will not help them.
Those whose lives were built around statistical methods in medicine, however, will find that their investment of time and money in their education will drop just like the stock quotations of companies who manufacture outdated technologies.
“Where this bothers me the most is in the realm of expectations. Someone will take a survey of, say, consumer expectations for home price increases. The results will diverge across consumers. But the economist will report a single number for consumer expectations.”
It can get very bad indeed. If I think home prices increase by 50% next year and you think they decrease by 50% the Fed will report that home prices are expected to be stable while the actual expectations are that prices will not be stable and the disagreement is in the direction.
Personalized medicine is now and the future. Researchers and regulators are trailing badly. If you are going to genotype for other outcomes, why not health?
With what serene conclusiveness a member of some Useful-Knowledge Society stops your mouth with a figure of arithmetic! To him it seems he has there extracted the elixir of the matter, on which now nothing more can be said. It is needful that you look into his said extracted elixir; and ascertain, alas, too probably, not without a sigh, that it is wash and vapidity, good only for the gutters.
Genomic medicine has not panned out compared to case control. Personalized medicine is “try this”, see if it works when titrated to effect, “start low go slow”, and then try an alternative if no effect is seen. Algorithmic medicine, ie “cookbook”, is the opposite. We used to do personalized medicine but now it is assembly line medicine. What we have learned from genomics is that everyone is a “one of” when DNA and its interaction with the environment is taken into account.
I think that’s a bit of a straw man. A well-done study by an economist would look not only at the average effect but also other moments like the average effect size conditional on nonzero effect (which in this case is 1), from which they would easily surmise that the treatment is effective for 1/3 of patients.
If treatment effects are not binary, economists would likely explore the distribution of effects using tools like quantile regressions.
The point is that no study is limited to a single statistic, and indeed to get published one would have to explore various dimensions of the result.