The height study has received a lot of press in the last few days. But Frances Woolley warns against taking its conclusions wholesale:
There is no data at all for people born before 1916. The first 20 years of “sprouting up” are generated by assuming that the 1896 to 1916 period was characterized by the same kind of increase in height as later periods. The data for the people born just after 1916 comes from surveys carried out in 1998 or later – i.e. from measurements of the heights of people up to 80 years old. To estimate the average height, at age 18, of people born in 1918 by observing that cohort in 1998 when they are 80 years old involves some heroic assumptions – assumptions about shrinkage with age, survival rates, etc. It would make a lot more sense to choose a shorter time span for the analysis, and give results that involved a bit less guesswork.
Is this bad science? I would say yes. It’s bad science because it oversells the results. The article overstates both the amount of height data the research team has (it’s not a century, in many cases it’s more like 50 to 75 years, especially for women), and also how recent the data is (in most cases the data is not for the 1996 birth cohort, but rather for earlier birth cohorts). It’s bad science because it presents headline grabbing results – and makes them readily available to journalists – without attempting to convey, in ways that are easy for reporters to understand, the amount of uncertainty associated with those results. Are Latvian women tall? Yes. Are they the tallest in the world? We can’t know that for sure unless we know the margin of error associated with the estimates of Latvian, Dutch, and other groups’ heights. It’s hard to put a standard error around the results of complicated projections – but that’s an argument against making complicated projections, and disseminating them to reporters, not against reporting standard errors.
H/T R. Briggs.