Pay, Productivity, and the Labor Share
On Twitter I’m often asked to respond to EPI’s viral chart showing a striking wedge between pay and productivity. I’ve done so many times, including in this long form essay in National Affairs, but I’ve never published a succinct, shareable response with charts — so here it is.
I’ve pasted the viral chart below. It contrasts economy wide net productivity (meaning output per hour net of depreciation) with the real average hourly compensation of production and non-supervisory workers. There is so much that’s problematic about this comparison, especially given the authors’ heavy implication that the two lines should move together 1:1. And the implications are pretty important. The notion of whether workers are paid in accordance with what they are able to produce cuts to the core of our economic system.
Here’s what the data actually tells us. Since 1947, worker pay and productivity in the nonfinancial corporate sector have moved together basically 1:1, with some temporary deviations. Net productivity has increased 341% from 1947Q1 to 2023Q1 while real hourly compensation measured with the same deflator (more on that later) is up 331%. In level terms, workers were producing an average of roughly $10.83 per hour and earning $8.55 per hour in 1947, and $47.71 and $37.02 now, respectively (all in 2012 dollars). In the rest of this post, I walk through the issues with EPI’s chart and explain the appropriate adjustments.
Issue 1: Comparing productivity for one set of workers to pay for a different set of workers. If we’re going to use economy wide productivity, we should be looking at economy wide compensation. If we are going to use non-supervisory and production compensation (the supplements to wages portion of which needs to be imputed because the payroll data only contains earnings, mind you), we should look at production non-supervisory worker productivity. The latter productivity series doesn’t exist and the former comparison doesn’t shed light on the question of whether workers in the private sector, on average, are paid in a manner commensurate with their productivity (because of bias from the housing and government sectors).
One could use the nonfarm business sector and many have such as Robert Lawrence and James Sherk. But we’d need to make adjustments for proprietors’ income, which is a mix of both labor and capital, but is treated as non-labor income in national accounts. In other words, without assigning a portion of it to labor, it would bias the gap between labor income and productivity over time. So it’s by far the best and cleanest solution to look at the nonfinancial corporate sector as it doesn’t suffer from any of these shortcomings. Now, you might be asking, what does the split between labor and capital have to do with productivity and pay? It has everything to do with it.
By definition, productivity is output (Y) per hour worked (N).
Productivity = Y/N
Compensation (C) per hour is, analogously, C/N
Because N is in the denominator for both calculations, a gap between pay and productivity growth mechanically implies a change in the labor share of income. Like so:
The labor share is C/Y.
If %ch Y/N > %ch C/N, then %ch Y > %ch C
If productivity is growing faster than compensation per hour, that implies declining labor share of income because output is growing faster than compensation. If pay is rising faster than productivity, then the labor share is increasing. This is why it’s so important to set up the analysis from a starting point where we’ve corrected for biases in the labor share.
Matt Rognlie, who is well known for his contributions to the literature in this space, sums it up very well:
“In short, the labor share should be as simple concept, but without an in-depth understanding of the national accounts, we can quickly be led astray. Amid the complexity, is there a standard measure of the labor share that we can use as our first pass at the question? Yes. The single best measure is the net labor share of domestic corporate factor income. This measure divides labor compensation by the sum of labor compensation and net operating surplus for the domestic corporate sector. This excludes proprietors’ income, excludes depreciation, and is unaffected by the split between capital income accruing to debt vs. equity. It also excludes income from foreign investments, which has no labor income counterpart and is conceptually not part of the domestic production function.”
Rognlie uses the total corporate sector rather than the nonfinancial sector like I do but there are some practical reasons why I use the nonfinancial sector. Most importantly, I’m not aware of an hours worked series for the corporate sector, but the BLS regularly publishes a series of hours worked for the nonfinancial sector. We need that to compare pay and productivity as opposed to just looking at the trend in the labor share. Second, the financial sector has its own measurement quirks, such as 1) in financial crises the financial sector can have negative profits causing odd spikes in the labor share and 2) it includes the profits (and losses) of Federal Reserve banks.
So bottom line: use the nonfinancial corporate sector (NFC) and define net output as compensation plus the net operating surplus (profits and capital income). This way, we’ve corrected for important biases and have an apples-to-apples series of pay and productivity. Here’s what the labor share looks like when measured properly. You’ll notice that there’s no secular downward trend over time.
Issue 2. Different deflators for productivity and compensation. EPI will use an output deflator for the productivity series and the CPI-U-RS deflator for wages. Given that we are interested in whether workers are paid for what they produce, not whether they are paid for what they consume, this adjustment doesn’t make sense conceptually. We should be using the same deflators (Mike Strain makes this point well here). This assumption also matters a great deal empirically. Over time, there’s a big difference between inflation as measured by output prices and consumer prices (the latter has grown more rapidly than the former). Another approach would be to ignore inflation measurement altogether and simply compare nominal pay and nominal productivity. Either way we’d get the same result showing no gap. But for our purposes, we are interested in real changes in productivity and compensation so I use an output deflator. Specifically, I use the same deflator that the BEA uses to deflate nonfinancial corporate net value added which you can derive in NIPA table 1.14. Note, you can also find additional net value added deflators for other sectors in table 1.9.4.
Issue 3: Comparing an average to a median. EPI often talks about pay for the “typical” worker, meaning the median worker, which they then compare to productivity for the average worker (we are talking about productivity for the average worker by definition when we divide total output by total hours worked). Why should we be surprised that the average is growing faster than the median? In this sense, EPI is baking in a wedge into their analysis.
A separate and far more interesting question is whether median productivity is lagging average productivity. In my mind, this would make a lot more sense for explaining why median and average pay have diverged over time. Sadly, we can’t measure median productivity, yet. But I wrote about the concept in my National Affairs essay. The late Ed Lazear also wrote a great paper exploring the topic.
NB: Now, if you’ve been reading closely, you’ll see that I claimed EPI is talking about “median” pay when they are technically talking about average hourly compensation for production and non-supervisory workers. Well, the answer is both. EPI quite literally uses an average pay figure to describe pay for the “typical” worker, because compositional issues (such as excluding managers) biases the pay series to more closely resemble changes in median pay over time rather than average pay. You can take a look by comparing the average hourly earnings series from the payroll data to their median wage series from EPI’s CPS extracts here.
