The productivity growth we experienced over the past 40 years was not enough on its own to generate substantial pay growth for typical Americans. (Scott Olson/Getty Images)

Median pay growth has been abysmal over recent decades: It rose only 12 percent between 1973 and 2015 while productivity rose by 73 percent. This stark fact highlighted in Figure 1 has led some to ask the question: Does productivity growth still benefit typical American workers?

Harold Meyerson, for example, wrote in American Prospect in 2014 that “for the vast majority of American workers, the link between their productivity and their compensation no longer exists.” The Economist wrote in 2013 that “unless you are rich, GDP growth isn't doing much to raise your income anymore.” Jared Bernstein wrote in 2015 that “Faster productivity growth would be great. I’m just not at all sure we can count on it to lift middle-class incomes.”

It’s clear that over the past 40 years, the productivity growth we experienced was not enough on its own to generate substantial pay growth for typical Americans. But this does not necessarily mean that productivity no longer affects pay, and the distinction matters for policy. It could indeed be the case that typical workers’ pay is not much affected by productivity growth and instead determined by other factors. This would imply that policy seeking to raise living standards should focus primarily on equity issues rather than productivity, at least in the short term. On the other hand it could be the case that productivity growth acts to lift typical workers’ pay even as other factors — like declining worker bargaining power, technological change or globalization — are acting to suppress it. Under this view, even in the presence of these other factors contributing to inequality, typical workers would be better off with higher productivity growth than without it.

Our recent research addresses the question of whether incremental increases in productivity translate into pay increases, recognizing that many other things affect pay.  We rely on the natural experiment provided by the fact that productivity growth fluctuates through time. If productivity growth translates into pay, then we should see periods of higher productivity growth coinciding with periods of higher pay growth. If not, the two should be unrelated.

The data supports the first conclusion: Increases in productivity largely translate into increases in pay, holding all else equal. In times of higher productivity growth the typical American worker has seen higher pay growth, as shown in Figure 2 (for both the median worker and the average production/nonsupervisory worker). The relationship holds strongly across a number of empirical tests: There is a large and statistically significant link between productivity growth and growth in median and production/nonsupervisory compensation. Our estimates suggest that a 1 percent increase in productivity growth is associated with two-thirds to 1 percent higher median pay growth and half to two-thirds of a percent higher pay growth for production/nonsupervisory workers. (For more details on these results, see our paper “Productivity and Pay: Is the link broken?” presented last week at the Peterson Institute for International Economics’ conference on Policy Implications of Sustained Low Productivity Growth).

This may seem counterintuitive: If productivity growth largely translates into a rise in typical workers’ pay, then why is it that productivity has grown so much more than pay over the past four decades? Our finding implies that even as productivity growth has been acting to push workers’ pay up, other factors have acted to push workers’ pay down. On net, the overall pay growth for median workers has been close to zero. If productivity growth had been lower over the past 40 years, typical workers are likely to have done even worse.

Understanding what has caused this divergence between productivity and the typical worker’s pay is of central importance. One explanation for this divergence and the concomitant rise in inequality is technological progress. Yet since more rapid technological progress should cause faster productivity growth, if technological progress has indeed caused rising inequality we should see periods of faster productivity growth coincide with faster growth in inequality. We don’t see this; in fact, inequality tended to rise faster during the productivity slowdowns of 1973-1996 and 2003-2015 than during the productivity booms of 1948-1973 and 1996-2003. This should give us pause before accepting explanations of the divergence between pay and productivity based purely on technology.

Our research on this issue was partly inspired by the influential work of Larry Mishel and Josh Bivens at the Economic Policy Institute, who have highlighted the divergent trends in productivity and pay. They have responded to our paper in a blog, discussing points of agreement and disagreement. We and they agree that productivity growth is necessary to raise typical workers’ pay. Similarly, we and they agree on the importance of policies directed at reducing inequality. The productivity growth we saw in the past 40 years was not in itself enough to raise living standards substantially for the median worker. And we tend to agree more than we disagree about the probable causes of the productivity-pay divergence.

Bivens and Mishel argue that we overstate the impact of productivity on pay because of the way we carry out our analysis. First they propose that we include the level of unemployment in our analysis (not just the change in unemployment). We are inclined to agree with their criticism here. There is a strong case for using the level of unemployment as a key indicator of labor market tightness. However, redoing our analysis in this way has only a small impact on our results.

Second, they argue that our results are not robust on the removal of the period of booming pay and productivity growth in the late 1990s and early 2000s. We have confirmed their statistical conclusion. It is true that removing this period meaningfully attenuates but far from eliminates the relation between pay and productivity growth.  However, this finding seems to us unsurprising and does not cause us to doubt our conclusions. The late 1990s/early 2000s represented the only period of persistently high productivity growth in the post 1973 period. So removing it diminishes the power of our natural experiment by limiting the variation in productivity growth.

Martin Sandbu at the Financial Times has also highlighted our work in a recent column. He raises two critiques of our analysis. First, he notes that we find a weaker relationship between productivity and pay before 1973 than since 1973. We agree with him that this is something of a puzzle. As we discuss in our paper, the extremely low variation in rates of productivity and pay growth during the late 1950s and early 1960s goes some way to explaining it, as it would tend to exacerbate the impact of noise and bias our estimates downward. In addition, including the level of unemployment in our regressions substantially raises the pre-1973 estimates and thus reduces the puzzle.

Second, Sandbu notes that our estimated relationships are often below one-for-one. This is correct. For the median worker our estimates are close to one and not statistically significantly different from one, while for production and nonsupervisory workers, our estimates are consistently between half and two-thirds, and often significantly different from both one and zero. As we note in our paper, this apparent difference bears further investigation. At the same time, given measurement error in productivity growth, one would expect the strength of the estimated relationship to be biased downward. The fact that our estimates for the median worker are close to one and for production and nonsupervisory workers are large, positive and significantly different from zero implies to us that productivity growth is doing much more to raise typical pay than an initial look at the productivity-pay divergence in Figure 1 would suggest.

Overall, the slow growth in typical workers’ pay over the past four decades and the large and persistent rise in inequality are extremely concerning on grounds of both welfare and equity. As policymakers and analysts seek to understand and reverse these trends, our analysis demonstrates that productivity growth still matters.

If productivity accelerates for reasons relating to technology or policy, the likely impact will be greater pay growth for the typical worker. This does not mean that policy should ignore questions of redistribution or labor market intervention: The evidence of the past four decades demonstrates that productivity growth alone is not necessarily enough to raise living standards. However, it does mean that policy should not focus on these issues to the exclusion of productivity growth. Strategies that focus both on productivity growth and labor market or redistributive policies are likely to have the greatest impact on the living standards of middle-income Americans.