Mar 3, 2023Liked by Brian Albrecht

Good point on the double S of “Selective Start dates” distorting tbe findings.

In U.K., I happen to have tracked income inequality over the past 2-3 centuries.

Absolutely not deflating into fictitious £s. Retaining the actual money pay of the day,

And using Pareto analysis to get the upper shape. And good U.K. wage statistics since 1968.

Conclusion for C20th is that the U.K. experienced highly atypical income compression 1948-78.

Fully restored pay differentials took equally 30 years to 2000-10.

Thatcher elected by desparate working class in 1979, no coincidence nadir maximum compression.

For “beware the timeframe to fit the narrative” point, most U.K. income inequality studied start 1970s

Ignorance of prior history makes the authors think their starting point is normal. In fact atypical.

The 1970s in U.K. so atypical that 2mn of best Emigrated. (See GDR 1950s, Venezuela, Zimbabwe).

Birth rate declined sharply as no future seen. Study all printed words showed gloom worse than WW2

Decadal productivity growth collapsed by one third from 1950s and 1960s.

Key point is that all these indicators reversed during 1980s & 1990s. With restoration of pay inequality

The 30 year V-shape down to 1979, fully reversed with reversal to usual pay differentials.

Lesson for “Selective Start dates” is check widely around, both economic & social data, if is typical.

A fair starting date will be in typical period, or made typical by using several year trend averages.

Expand full comment

With the long run data, you can at least have a debate about whether the trend or previous highs/lows are sustainable or if policy caused the change. If you start at the peak or trough, it’s “everything is going to hell “

Expand full comment

Great post! I like the football weight analogy especially. A base year always needs to be chosen, but the key is that choice is defended. There are a bunch of ways to do this. The first, as the authors do in the QJE article, is to state why a year is picked as a base year. Another is a sensitivity analysis, where the starting year is moved forward/backward ~5 years to show the results are similar. A third is to pick a year that provides a intuitive cutoff, for example 1945 or 1989. While politicians can get away with just picking the base year that makes them look best, peer review will generally require a robust defense.

Expand full comment

This is one benefit of all the robustness checks that referees ask for. But the Barkai paper doesn't defend the 1984 timeline so it doesn't always happen. Maybe he did to referees but it's not in the final draft unless I missed it.

Expand full comment

A base year doesn't necessarily need to be chosen. Couldn't you use a base year range e.g., 1980-1990? Then your finding could measure off the peak and trough within that period (between 2%-5.5%). Sort of a quasi sensitivity analysis - output is a range versus a definite. Like super simple corporate finance math - not complicated academic stuff. Well dang, wrote all this and saw Patrick already beat me to it

Expand full comment