Wages and Productivity

One of the major assumptions of many economic models is that workers’ pay and productivity are linked: wages are determined by the marginal product of labor. On the other hand, some economists have pointed out a growing divergence between wages and productivity in the US economy, which suggests the link is not so close. This chart is a common example:


Who’s right? Well, this very topic is the subject of one section of a new AEI book, simply titled ‘The US Labor Market: Questions and Challenges for Public Policy’ (free pdf available). The contributor list is a veritable who’s who of economists whose analyses I find invaluable, and it looks to hit on many themes I’m interested in, so I’m excited to dig into this. But for today I’ll stick with wages and productivity.

One the side of divergence is Dean Baker of the Center for Economic and Policy Research. He argues that the top and bottom of the income distribution exhibit a disconnect between wages and productivity. He starts with the example of CEO pay, which as I noted in my post on inequality in firms, has pulled away from the rest of workers in very large firms. If pay were linked to productivity, this would indicate American CEOs have become an order of magnitude more productive than they were decades ago (as well as more productive than similar CEOs in other rich countries). What this suggests is really going on is that due to poor corporate governance structures, CEOs are commanding wages far out of line with what they would earn in a truly competitive marketplace.

Next Baker turns to the minimum wage, and basically argues that a combination of a wage floor and firms’ desire to have a differentiated wage structure means there is another disconnect between wages and productivity. If the minimum wage increases, there is an incentive to raise other workers’ pay as well to maintain distinctions based on seniority, skills, etc. He also argues this means that these workers are being paid less than their productivity already, using a monopsony model as a base. There’s a decent explanation here, but basically a monopsony is the opposite of a monopoly; i.e. one buyer and many sellers, so the buyer (employer) has power over pricing (wages), and in these models wages are well below what they would be in a competitive marketplace.

I don’t find this particular part of the argument so convincing based on my experience in a city that has raised the minimum wage. Over the last few years, the DC minimum wage has gone from $8.75/hr prior to 2014 to $11.50/hr as of July this year, and will reach $15/hr by 2020. In that time, at the (large) retail company I work for, there have been no raises beyond regular cost of living for workers between cashiers and salaried managers. So now workers employees with decades of experience are at the same level as fresh cashiers, and the wage difference between a supervisor (who is typically hired around $12/hr) and a regular cashier has basically disappeared. It still suggests there is disconnect between wages and productivity, just not the way he envisions it.

Moving on, Robert Lawrence of Harvard argues that looking at charts such as the one above is misleading because the deflators used in measuring compensation and output across time are used improperly. The real compensation series uses the CPI deflator, while the real output series uses an implicit price index for GDP. The CPI has some problems; many believe it overstates the real cost of living, and Lawrence argues that in this context an output price index should be used to deflate compensation. Why? Because using a consumption index includes factors that domestic workers do not produce, such as imports and housing. As such, a ‘product wage’ should be used, and doing so accounts for a significant chunk of the observed gap. Using his estimates of productivity, he find that productivity and compensation track each other closely until about 2001, and it’s not until the Great Recession and its aftermath that a real divergence opens up, though not a massive one.

Something that’s often implicit in arguments in this space is that productivity/pay divergence is the main source of inequality. I think the reason for this is that it implies an easy solution: if we can just control the greed of the top 1% and make companies pay workers more, then everything will be ok. If it turns out that people really are broadly being paid their ‘fair’ share, then solving the problem of inequality becomes much harder (as we’ve seen described in The Wealth of Humans).

Where do I come down on this? While I do think there’s something ‘unfair’ going on with things like CEO compensation, I also just don’t see that playing out at the scale required to explain the levels of inequality in the US. And so we’re left to grapple with far more difficult questions.


Firming Up Inequality

Another day, another NBER paper. ‘Firming Up Inequality’, by Jae Song et al., is about a year old, but the latest version is from this summer (pdf). This paper uses a comprehensive matched employer-employee data set to investigate just how changes in inequality have played out between and within firms.

The fact of dramatic increases in wage inequality in the US is well documented, but there is considerable disagreement on its root causes. As we’ve seen argued in books like The Wealth of Humans, skills biased technical change is one hypothesis, which boils down to changes in markets rewarding a particular subset of workers. Another hypothesis is that changes in inequality are mostly due to the powerful members of society extracting economic rents.  A common example of this is the relative rise in American CEO pay vs other developed countries. If this rent seeking hypothesis were true, we would expect to see increasing inequality within firms, as CEOs and other top managers captured greater shares of firms’ profits.

What they find is that the majority of the increase in income inequality is actually explained as a between-firm phenomenon, not a within-firm phenomenon. Furthermore, this finding holds across industry sectors, region, demographics, firm size (with one exception; see below), etc. It’s not the case that there has been any sort of differential growth leading to this.

Finally, they find that there has been one type of firm that has seen increases in within-firm inequality: the ‘mega-firm’, i.e. a firm with more than 10,000 employees. In the US there are around 800 mega-firms that employ about 20% of the workforce, and these have seen both decreases in the lower end and increases in the higher end of the income distribution. In the authors’ estimates, the median worker in such firms saw earnings fall around 7% from 1981 to 2013; in contrast, the top 10% saw increases averaging around 11%. And the managers at the very top in these firms have realized real earnings increases of 137%.

So what’s going on here? Most firms are becoming more equal internally, yet firms as a whole are becoming less equal.  And at the same time, the very largest firms are seeing the opposite pattern.

Song et al. propose two potential explanations for the increased in interfirm inequality: a ‘widening firm premium’ story, or a ‘worker segregation’ story. In the former, firm inequality is driven by the fact that some firms are winners and others losers in the economy; winners may end up distributing their increased gains to their employees, while losers cannot. In the latter story, workers are increasingly being sorted into firms by ability; high-ability and low ability workers are clustering into separate firms, rather than mixing together.

Continue reading “Firming Up Inequality”

Recessions and Skill Biased Technical Change

A very fascinating and topical NBER paper (ungated pdf here) recently came out which investigates the connection between skill biased technical change (or routine-biased technological change, as these authors prefer to call it), which has been a major theme in the posts on The Wealth of Humans, and recessions.

One curious fact about the last several recessions is that their recoveries have been largely jobless. If we eyeball the growth in employment over the last several decades, it’s obvious we never returned to trend even after the 2001 recession.


Furthermore, unemployment continued to rise even after the technical end of the post-1990 recessions.


The reasons for this have not been well understood, but the authors take an adjustment cost theoretical framework as a starting point.

We can imagine that as technology makes certain jobs less relevant while creating others, there are costs to employers in making the appropriate adjustments. Recessions can provide the necessary ‘oomph’ to reallocate labor and capital that wouldn’t happen in normal times. In this understanding, labor market adjustment to RBTC occurs episodically, not gradually.

I’m sure it’s not hard to visualize that person at the office who isn’t really necessary anymore since that new software can do 70% of their work, but is still around since the boss is a human being and can’t find a compelling reason to fire them. In these models, economic downturns provide the excuse to clear out inefficiencies that have built up over time.

While this makes intuitive sense, the question is whether there is data to back it up. Other authors have found that the vast majority of declines in middle skill employment have occurred during recessions, and subsequent recoveries have been jobless in precisely these occupations.

Still, these studies have lacked direct evidence on how firms were restructuring. That’s where this paper comes in: they’ve obtained a data set that contains huge amounts of online job postings in major metropolitan areas for the years 2007 and 2010-15.

They find first that employers in metropolitan statistical areas (MSAs) that were hit harder by the Great Recession had an increase in skill requirements in jobs ads relative to both those same areas before and MSAs less affected by the shock. Now, there are two possible causes of this: one could be that employers simply increased the asking skill levels for all jobs, or happened to be posting mostly high skill levels jobs and less low skill jobs. The authors find the former explains most of this ‘upskilling’: in a weak labor market, employers were able to demand higher skill levels for the same jobs relative to a few years prior or to MSAs with better economic conditions.

This leads to the next question: are these firms actually changing what they do, or are they merely changing the type of people they hire? The former implies Shumpeterian idea of a recession as a cleansing activity (and consistent with the adjustment cost explanation above), whereas the latter is more like taking opportunity of a slack labor market to get the type of people you wouldn’t normally be able to. The latter is temporary; the former, permanent.

Continue reading “Recessions and Skill Biased Technical Change”

The Wealth of Humans, Part VII: Solutions?

<Part I, Part II, Part III, Part IV, Part V, Part VI>

To recap: the Digital Revolution has reduced the bargaining power of the average worker through two channels. The first is by creating an abundance of labor through automation, enhanced productivity, and globalization. The labor supply has outstripped labor demand, forcing workers to accept low wages or no work. The second is by changing the nature of the most productive activities. Serving as a cog in an industrial process was once a high value activity with clear and measurable aspects: how many units did you produce? Are they up to standard? But now the highest value activities occur within firms acting primarily as information processing organisms, and the success of these firms relies on social capital. This social capital is hard to measure, and by its nature cannot be taken or easily replicated. So despite being crucial to the success of their firms, workers don’t have the bargaining power to fully capture their share of gains versus the owners of capital.

This has led to a stagnation of wages and an employment trilemma that suggests that the only sources of mass employment in the future will be low productivity, low wage work. Is there anything we can do?

Avent runs through a few proposals to get the average person more money that I’d like to explore. I think it’s important to keep some questions in mind as we consider each.

  • Is our goal to reduce inequality or simply raise the floor for the bottom end of the income distribution?
  • Is there a trade off between inequality/redistribution and economic growth?
  • What types of redistribution do we want to engage in?

It should be said that the assumption here, and one that has underlain the entire book, is that today’s inequality is chiefly due to skills biased technical change. There are other potential causes, such as growth in economic rent seeking, that should be kept in mind, though I won’t deal with them here.

Continue reading “The Wealth of Humans, Part VII: Solutions?”

The Wealth of Humans, Part VI: Superstars

<Part I, Part II, Part III, Part IV, Part V>

It’s tempting to ask, what’s really different about this time? Isn’t this the same story we’ve heard every time some new advance comes along? It’s the nature of technology to destroy some jobs, and enhance and create others; that’s creative destruction and it’s always been with us.

While I’m generally sympathetic to this sort of ‘Ecclesiastical’ thinking (nothing new under the sun), I do think there are signals that this time really is different in important ways.

I don’t believe Avent touches on this in the book, but what I see as one of the big differences is the prevalence of what’s dubbed The Superstar Effect (pdf). This is most plainly evident in certain industries, like music, film, and televised sports: industries that are based around the transmission of information, or where, to quote from the paper, ‘the costs of production do not rise in proportion with the size of the market’. In the case of information transmission, such as film, television, or recorded music, the marginal cost of distribution is virtually zero. Furthermore, these types of goods have the characteristic of being ‘non-rivalrous’: just because I purchased The Wealth of Humans doesn’t prevent you from buying it also (so you should!).

Another key feature is imperfect substitution in these markets. Buying 10 cheap sponges could be the equivalent of buying 1 good sponge: the good one lasts ten times as long as the cheap sponge, but beyond that there isn’t much difference; they function as perfect substitutes. But watching 10 episodes of Home Improvement doesn’t add up to one episode of Breaking Bad.  Consequently, demand increases more than is proportionate to the degree of quality. To quote again from the paper, ‘If a surgeon is 10 percent more successful in saving lives than his fellows, most people would be willing to pay more than a 10 percent premium for his services.’

When markets have both of these features, non-rivalry and imperfect substitution, the superstar effect comes into play: one or a few players can capture outsize shares of the market. Think of the acting abilities of the average Hollywood star; they aren’t thousands of times better at acting than the people you see on TV, but being just slightly preferable to the average consumer means those stars make huge gains.

Technology, via replication, is what allows non-rivalry. And again, historically this has been most evident in industries based around transmission or communication of information. So it’s quite likely the Digital Revolution will continue to exacerbate this problem by rendering more and more industries subject to replication. Furthermore, technology will continue to reduce the marginal costs of production in many other fields so that they also can also expand more than proportionally. One example could be programmers; in the deep learning market, the good being designed is multi purpose. Once you have your program, you can apply it to all sorts of problems at little additional cost; DeepMind didn’t need to be reworked from the ground up each Go match it played, nor will it have to be totally redesigned for its next endeavor. 

Or consider remote surgery. The best surgeons will be able to service vastly expanded markets, rendering the services of all but the most skilled surgeons largely unnecessary. So I could see superstar effects taking hold in many more industries.

This is one big reason why I think the changes brought by the Digital Revolution aren’t some temporary phase. The very structure of many job markets has the potential to be fundamentally transformed to dramatically favor only the best. And if this comes to pass, what happens to the rest of us?

The Wealth of Humans, Part V: The Employment Trilemma

<Part I, Part II, Part III, Part IV>

What then does this combination of automation, an abundance of labor and the rise of social capital mean for the future? Avent suggests that we now face an employment trilemma between three factors, of which we can choose at most two: high productivity/wages; resistance to automation; and mass employment.

The reasoning is thus: Any job with the potential to have all three characteristics will become self limiting almost immediately. Jobs with high productivity inevitably limit the amount of employment, either because output quickly exceeds demand (lowering prices and thus productivity) or the high wages incentivize producers to automate the tasks (lowering employment).

We can see this in the data. Let’s look again at the BLS data on industry performance from 2004-2014. Again, for this data the measure for productivity is output per worker rather than the usual output per labor hour. I’m working on getting another data set from the BLS together that has the usual productivity measures in it, but this will have to do for now.

First, a look at the headline sectors. The size and color of each bubble are the number of employees and productivity of each sector, respectively. The x-axis is employment growth, y-axis is output growth, and the 45 degree line marks where the two are equal (above the line means output has grown faster than employment, and vice versa).


An industry that avoided the trilemma would be big and dark red, and we do see one of those: manufacturing. But the sector is also shrinking and has been for decades thanks to automation and globalization. The information sector is promising in having high productivity and resistance to automation, but again employment is shrinking. The sectors that are commonly touted as being the future of mass employment, education and health care, have low productivity. Indeed it is clear from this graph that all the large sectors and growing sectors have low productivity, as the trilemma suggests.

What about mining? That has high productivity and rapid employment growth, after all. Could this be a future source of mass employment? Well, consider again that as employment increases, so does output, which will push down oil and gas prices (the main driver of the job increases). This is already happening.

Let’s dive a little deeper into these industries:

Continue reading “The Wealth of Humans, Part V: The Employment Trilemma”

Urbanization and poverty reduction in Africa

Brookings had an interesting article a short while back with an interesting analysis on urbanization in Africa that’s forced me to rethink my priors on urbanization as a tool of poverty reduction.

It seems that there are important differences in the reasons for urbanization in the developed world vs much of Africa. While urbanization has a been a driver of enormous creativity and growth in the Western world, Africa has largely not seen the push/pull dynamic of advances in agricultural productivity (that would allow people to leave rural areas for the city) and industrial productivity (that would lure people to the city in the first place) that was the case here.

If these forces are absent, what is driving African urbanization? The authors point to three strands of research. One is that urbanization is driven by natural resource extraction, the exports of which create wealth. Remi Jedwab at George Washington University has a paper that suggests that in areas that derive their income from the exports of natural resources, the people tend to desire urban amenities with their newfound cash. What are these urban amenities? Well, they tend to be in the non-tradables sector; that is, services which can’t simply be bought from abroad. So this creates a labor pull, but without the commensurate increase in productivity. Instead, basically all economic activity is sourced from the rents from resource extraction. This leads to low productivity, service sector oriented cities, which Jedwab dubs ‘consumption cities’. (The whole paper is worth a read, it serves as a pretty good overview of this topic).

Another strand posits that politics is a driver of urbanization. Apparently, some research suggests that the capitals of autocratic governments are larger than they ‘should’ be, and this makes sense. In many of these countries, the formal sector is severely constrained or practically absent, and a government job is the closest guarantee to success there is. So, people will try to relocate to the source of political power in the hopes of making connections.

The final strand discussed is improvements in health. There seems to be evidence that, contra the history of industrialization in the West, new urban areas have far better health outcomes than rural areas, and the effect is greater the larger the city. Indeed, even growing up in the slums of a large urban area is still healthier than being raised in a rural area, which is pretty crazy. We could imagine that if people are aware of the health related possibilities in urban areas they would relocate even absent any other incentives.

The message here is that the reasons for urbanization matter, and it isn’t a given that the good things associated with urbanization in the developed world will follow just by throwing a bunch of people together in close proximity.