Thread by @ernietedeschi

Here's a quick, slightly esoteric thread on how we measure wage growth at the bottom. /1

We're seeing more and more stories like this good one from @EricMorath and @jeffsparshott, talking about heating wage growth at the bottom of the income distribution. /2
wsj.com/articles/rank-…

@EricMorath @jeffsparshott There's growing evidence from different sources that this is the case. One key dataset often cited is the Atlanta Fed Wage Growth Tracker, which shows faster wage growth in the bottom quantile than in the other three. /3 frbatlanta.org/chcs/wage-grow…

@EricMorath @jeffsparshott The Atlanta Fed Wage Growth Tracker is different from most other wage growth measures you probably see bandied about, in that it's a measure of typical wage *growth* among the *same* workers employed 12 months apart. /4

@EricMorath @jeffsparshott The Atlanta Fed measure uses the Current Population Survey (CPS) to link the same workers across a year, and then calculates the median wage growth those workers report between those two periods. I've (more or less) replicated their measure below using the CPS. /5

@EricMorath @jeffsparshott Now, when you see median wage growth by "wage quantile", you might *assume* that means the quantile of wages at the *start* of each 12 month period. At least, that's the way I first interpreted it years ago.

But you'd be wrong! /6

@EricMorath @jeffsparshott In fact, if we look at wage quantiles by where workers *start*, the lowest-wage workers **consistently** out-perform higher-income workers. /7

@EricMorath @jeffsparshott And this actually makes some sense if you think about it: it's just the law of large numbers. Wage growth here is measured as a percent, so any fixed nominal raise is going to be a much larger percentage raise for low-wage workers relative to everyone else, by definition. /8

@EricMorath @jeffsparshott What about quantiles by where workers *end up*? It's almost completely the opposite ordering! High-wage workers consistently do better.

But this is an even less useful measure, because ordering workers by *outcome* necessarily penalizes workers who end up low-wage. /9

@EricMorath @jeffsparshott So what does the Atlanta Fed do? They take the quantiles of the *average* wage across *both* periods. So it's a hybrid of the two approaches. /10

@EricMorath @jeffsparshott I'll conclude by saying I don't think the Atlanta Fed approach is "wrong" or "bad". It gives us useful insight about wage growth along a longer-than-single-point-in-time distribution of income. But... /11

@EricMorath @jeffsparshott ...by the same token, many people naturally think of wage growth quantiles in terms of *starting* wage.

So if you ever have to bet which workers now will end up with the highest *percent* wage growth after a year, *always bet on the lowest wage workers*. /FIN

@EricMorath @jeffsparshott POSTSCRIPT: Let me add that other wage measures are showing firmer growth at the bottom. This chart is from @nick_bunker and uses the payroll survey to show growth in average wages by industry, ranked by each industry's wage level...

@EricMorath @jeffsparshott @nick_bunker ...while this is another chart from me using the CPS, but rather than look at longitudinally-linked workers, it simply shows how average wages in each tercile have grown over time.