In other words, during the nearly half-century from 1970 to 2018, there was not a single example of Summers' stated condition — unemployment under 4% and inflation over 4% — ever existing.
There are some earlier examples, at least in a limited way.
For about a year in 1951, unemployment was under 4% while inflation was above 4%. But inflation then dropped to 1.1% or less for the six months immediately before a recession in 1953, so this is not unequivocal evidence for Summers’ larger point about high inflation being a precursor to a recession.
We found only one clear example of Summers’ rule holding — a two-year period in 1968 and 1969 when unemployment was under 4% and inflation was above 4%. That came in the immediate run-up to a recession.
Still, this means there’s just one example of Summers’ rule holding over nearly 75 years of economic data. That’s not exactly ironclad empirical evidence.
This is known in data analysis as the "small-n" problem, short for "small number of examples." Social scientists and economists generally warn against drawing sweeping conclusions based on small numbers of examples.
So what’s going on with Summers’ rule?
We didn’t hear back from Summers, but we did hear back from a research collaborator of his, Alex Domash of the Harvard Kennedy School. When Domash walked us through the supporting data, it became clearer to us that the research upon which Summers was basing his formula had more nuances than his talking point to journalists indicated.
Domash pointed us to a blog post that is the source of Summers’ talking point. In it, he and Summers looked at data similar to what PolitiFact used, although they used quarterly figures rather than monthly figures.
A chart in the blog post breaks down the data in some detail. It covers every quarter from 1955 to 2019, or 256 separate quarters.
Using the thresholds cited by Summers in his media interviews — inflation above 4% and unemployment below 4% — they found that the likelihood of a recession was 100%. However, these conditions only existed in seven quarters out of 256, or less than 3% of the time.
Domash told PolitiFact that the "small-n" issue "is a valid critique," and one that he has since sought to improve upon by enlarging the base of research. He said he recently used the same analysis to study 30 member nations in the Organization for Economic Cooperation and Development, a group of richer nations, and found that low unemployment and high inflation made the probability of an impending recession "around 90%."
Still, Summers’ offering of a straightforward formula in the media obscures the reality that there’s barely any historical precedent, making it less than a robust method for extrapolating into the future.