Piecing Together the Wage Puzzle
Tom Heintjes: Welcome to another Economy Matters podcast episode. I'm Tom Heintjes, managing editor of the Atlanta Fed's Economy Matters magazine, and today we're joined again by John Robertson, a senior policy adviser at the Atlanta Fed. John, you've been with us on the podcast before, so I appreciate your coming back and sitting down with us again today.
John Robertson: You're welcome, Tom.
Photo: David Fine
Heintjes: John is one of the architects behind our Wage Growth Tracker—which, of course, tracks wages, as you would expect from something named the Wage Growth Tracker. And with unemployment low and 2018 largely in the rearview mirror, I thought it would be a good time to talk to John about what we're seeing in wage trends. But before we get into recent observations, could you just briefly tell us what sort of information goes into the tracker—you know, what sets it apart from other measures of wage growth and so forth?
Robertson: Sure, Tom. Most measures of wage growth look at it from the perspective of the change in average wages—so, what is the average wage bill, and is it higher than the average wage bill last year? What's unique about the Wage Growth Tracker, and kind of makes it a cool thing, is it looks at the distribution of actual wage changes as opposed to the change in the average wage.
Heintjes: Right, interesting. Well, actually I want to touch on that again later, so keep that thought in mind. But I wanted to ask you—did wages in 2018 behave largely as you would have expected? At the end of 2017, if you foresaw an environment of very low unemployment and low inflation, is wage growth about where you would have pegged it?
Robertson: Well, if you looked at measures of average wages, I think the answer is pretty clear that there's a puzzle, where the growth in the average wage bill for firms is not growing as quickly in relationship to the unemployment rate decline as it has in the past. And so I think we saw an environment of low growth in average wages—which is not totally inconsistent with low growth in average labor productivity, which we also see in the aggregate statistics. If you look at the Wage Growth Tracker, that also has not moved over the last couple of years upwards in the way that it had in the past in relationship to the unemployment rate. So that, too, was a bit of a puzzle.
Heintjes: So what do you think is behind the puzzle? Or is that sort of what you're working on right now, and sort of untangling?
Robertson: That's an ongoing exercise, trying to figure that out.
Heintjes: Well, that would be a good reason to have you back on at some point. John, the Wage Growth Tracker looks at anonymous actual people and their wage changes, as you mentioned earlier. Something I've wondered is, when someone currently captured in the tracker leaves the workforce for whatever reason—retirement, disability, what have you—how is that person replaced in the dataset? In other words, how do you maintain whatever demographic mix you need since these people are not identified?
Robertson: Well, actually, we only see an individual for that one wage growth observation. That is, we see them last year and we see them this year. And then in the survey—the Current Population Survey, which is the source for this data—those people are replaced with other survey participants. So what we're really relying on is the randomness in the survey, so that we're always getting new people who kind of look like the current population.
Heintjes: So again—not to get too in the weeds here—but when you say they're replaced, who is replacing them? Is that you or someone at the Census Bureau?
Robertson: We're using the Current Population Survey, which is the survey where the unemployment rate comes from, so it's a survey of households conducted by the Census Bureau on behalf of the Bureau of Labor Statistics. They have a procedure for making their sample representative of the population. And then, as people rotate in and rotate out, they're always being replaced with people to keep the mix in the sample representative.
Heintjes: So they keep it sort of demographically balanced over time.
Robertson: Yes, so as the demographics change, you're going to see the mix of people in the survey change.
Heintjes: Got it. Well, I had wondered about that, so I appreciate your clarifying it. John, you and your team in Research recently tweaked some of the architecture of the Wage Growth Tracker—sort of the underlying structure of it. What kind of modifications did you make, and what improvements did they bring to it?
Robertson: Well, that's a good question. It allows me to mention that the real brains behind the operation—
Heintjes: Oh, you're so modest, John!
Robertson: —is Ellie Terry, my colleague. She had the idea that because we can only measure the wage changes of people who are actually employed both this year and last year, we're kind of missing some people who, for one reason or other, are in and out of employment. And that especially is true of younger workers, who we might miss even though they get back into employment. They just don't happen to be employed exactly when we measure their employment this year and last year. And so she developed a way to reweight the sample to make it more representative of the employee population today—so basically, give more weight to these younger workers who the Wage Growth Tracker is not picking up because of their absence.
Heintjes: Interesting. Well, I guess the Wage Growth Tracker is sort of an ongoing refinement process?
Robertson: That's right. Another thing that we did recently was—even though, as I said, the sample's always being refreshed to look like the population looks today—and then there's tomorrow, it'll get changed again to look like the population tomorrow. The population characteristics are changing over time, and so there can be pressures on the wage growth that are coming from a shift in the composition of the workforce. The most obvious one today is the aging of the population. Research is showing that older workers tend to experience lower wage growth than younger workers, so as the population becomes older you're going to see a downward trend in wage growth that's just coming from the shift in the composition of the population. So we developed another weighting mechanism for the observations so that the population would look the same as it did in 1997—just so that it would allow a better apples-to-apples comparison of wage growth over time.
Heintjes: Right, that makes sense. Well, John, as we speak here it's late 2018, November 2018, and we're well out of the recession—and as I mentioned earlier, in a period of very low unemployment. In your view, are trends in wages behaving as you would expect—not just by current standards, but by historical standards?
Robertson: Well, as I mentioned earlier, there seemed to be developing a bit of a disconnect between those traditional relationships between the unemployment rate and wage growth. In the last few months, we've actually seen an uptick in wage growth as measured by changes in average wages. And indeed, the most recent Wage Growth Tracker observation for October of this year is up 3.7 percent from a year ago, which is higher than where it's been earlier in 2018. So maybe it's also showing signs of a pick-up more consistent with every other signal that we get, either in the data or anecdotally, about the tightness of the labor market.
Heintjes: Right, because conventional wisdom holds that when unemployment is very low, wages rise to attract workers because there's great competition for labor. Is this what the Wage Growth Tracker is now telling us is happening?
Robertson: It may be. I think it's a little too early to draw that conclusion definitively, but as I mentioned we're starting to see the Wage Growth Tracker moving higher—more consistent with the movements in the unemployment rate. If you looked historically and you said, "Give me a wage growth measure that had the tightest connection to the unemployment rate," you would definitely pick the Wage Growth Tracker. It had a very tight relationship with the unemployment rate. So as I said, that relationship seemed to soften some but maybe it's becoming reestablished again.
Heintjes: Well, I know in wages—as in sports—it's very hard to compare eras. It may be impossible. But just for the sake of discussion, if we were to go back and find another period of consistently strong jobs numbers, as we see now, and very low unemployment, what would the trend in wage growth have been? Similar to what we see now, you think, or different in some way because of structural reasons in the economy?
Robertson: Yes, so comparing wage growth over time is tricky. One of the factors, as I mentioned, was about shifts in the composition of the workforce. Another is the underlying rate of inflation has changed over time. Another is workers tend to get paid for what they bring to the table, and so productivity—the productivity of the workforce—also matters for what's going on in wages. If you compare back to the late 1990s—which was a period of somewhat higher inflation, but also higher average labor productivity—you saw the Wage Growth Tracker up around 5 percent growth. The most recent number of 3.7 [percent] is lower than that, but then so is inflation and so are measures of labor productivity. So the connection isn't necessarily broken, but comparing over time you need to take into account all the other factors that have changed over time as well.
Heintjes: Well, John, let's break this down a little and talk about different segments of the labor force and the different wage changes they're seeing, they're experiencing. You've written in the Atlanta Fed's macroblog about how educational attainment affects wage growth.
Robertson: Wage growth varies a lot, and in fact, if you step back and look at the variation that people who otherwise look very similar, the wage growth that these individuals have varies a lot. And so there's a lot that's going on that's not easily explained. If you focus in on the kind of typical wage growth, then yes—you do see some patterns. Younger workers early in their career tend to experience faster wage growth than older workers...
Heintjes: Because they're starting from a lower level?
Robertson: Exactly. And that's something that I want to stress. A thing that sometimes confuses people when thinking about wage growth versus wage levels. A one-dollar increase in the hourly rate of pay for somebody who's earning $10 an hour is a 10 percent wage increase.
Heintjes: Yes, a very healthy percentage boost.
Robertson: But for somebody who's earning $50 an hour, one dollar is not so high a percentage of a boost. So that's why you tend to see high wage growth for people when they're in kind of early parts of their career. And then there are other reasons why the returns to an extra year of experience when you're young, early in your career, are pretty big, but less so as people approach retirement age.
Heintjes: Well, what other demographic factors could be at work? I mean, age, obviously, gender. How is wage growth across these dimensions? What did you see in 2018?
Robertson: Well, one of the things I've been looking at recently is the relationship between wage level and wage growth. If you look at somebody who, on average this year and last year, was a low-wage worker, do they tend to have higher or lower wage growth than somebody who was a high-wage worker?
Heintjes: And what did you see?
Robertson: And you actually see an interesting pattern where low-wage workers right now are experiencing high median wage growth—higher than even high-wage workers.
Robertson: And that seems consistent with the anecdotal reports that we've received that are documented in places like the Federal Reserve's Beige Book, where employers complain about the shortage of relatively low-skill workers, or lower-wage workers. And we're actually seeing the labor market respond through higher wage growth for those workers. And if you look historically, you see that play out in the past as well. When the unemployment rate is very low, you tend to see higher wage growth for that segment of the population.
Heintjes: Right, the competition we were talking about before.
Robertson: The group where you don't see strong wage growth is the group in the middle—the middle of the wage distribution—and that seems consistent with the literature and economic research that's talked about job polarization and the polarization of the wage distribution. And inequality, where low-wage workers are in demand, are seeing good wage growth. High-wage workers, or workers in more skilled occupations, are seeing healthy wage growth, but the people in the middle are not.
Heintjes: So when you talk about people in the middle, it puts me in the mind of what you call it the "prime-age worker"—those roughly 25-54 years old. What sort of wage increases are they seeing?
Robertson: They are the bulk of the workforce, so what you see for them is pretty much what you see for the overall Wage Growth Tracker. The median worker is going to be a prime age worker, but, as I just mentioned, you see a lot of diversity even within prime age depending on the types of jobs that those workers have or the types of skills that those workers have.
Heintjes: Well, speaking of aging—how has the aging of the workforce in general in recent years affected wages and the way wages are measured?
Robertson: They have an effect on wage growth because of the shift in composition of the workforce. When we constructed this demographically fixed version of the Wage Growth Tracker, where we kept all the demographics fixed like they were in 1997, you see slightly higher wage growth starting around 2010 when you fix the demographics, than if you just let them play out, because you see lower wage growth amongst older workers, and as you get more older workers, that pulls it down. So once you control for that, you see somewhat stronger wage growth.
Heintjes: We talked briefly before about prime-age workers—those who are roughly 25–54 years old. For this group, does it come down to educational attainment within that group that affects their wage increases, or are there other factors at work beyond educational attainment?
Robertson: I think there are a lot of factors in play. Again, this goes back to the distinction between wage growth and wage levels. Economic researchers have looked a lot at the sources of variation in wage levels across workers, and they can account for a fair amount of that variation through demographic and labor market job characteristics. The variation in wage growth is much greater and much harder to account for.
Heintjes: As we've noted now, John, the Wage Growth Tracker allows us to look at wage trends among people who switch jobs versus those who remain with the same employer. As you put it in one of your recent macroblog posts, "Does loyalty pay off?" What have you seen in terms of wage trends among these two groups of workers: those who stay with the same employer long term versus those who switch employers? So I guess, does loyalty pay off? [laughter]
Robertson: In the Wage Growth Tracker data, we're able to identify people who were with the same employer, probably, as they were a year ago. I say "probably" because, as a result of the structure of the survey, we can't track people every month over the year.
Heintjes: The anonymous nature of it.
Robertson: The anonymous nature of it. But when we say, "Well, to a good approximation, these are people who are with the same employer," and then compare them to everyone else, we see lower wage growth during strong labor markets than with people who we don't identify as staying with their employer—many of whom we think switched jobs. We can't definitively say they did switch jobs. We can only say they're not in the group that we can definitely say did stay. But we do see, when the labor market is tight, the job switchers have stronger wage growth—median wage growth—than the job stayers. During bad times, you see the reverse—everybody's wage growth slows, but it slows much more for the people who switch jobs because in that environment switching jobs is probably not voluntary. They're just having to change jobs because the current job has gone away.
Heintjes: Sure. Well, John, one of the things economists such as yourself look at is "slack in the labor market," which we can summarize as unused potential in the labor market. Does the Wage Growth Tracker allow us to infer anything about the amount of slack currently in the labor market?Robertson: If you look at the relationship between the Wage Growth Tracker and the unemployment rate, you can see that they move together in general over time. There has been a little bit of a disconnect recently, but perhaps with the rise in the Wage Growth Tracker in the last few months, that connection's being reestablished again. So you could think of the Wage Growth Tracker as a proxy for slack in the labor market, given that it's signaling very similar things to what the unemployment rate is signaling.
Heintjes: Interesting. That would make it doubly useful, because slack is such a closely watched metric. You know, John, I was watching Fed chair Jay Powell discuss wages recently. He was being interviewed on stage, and he actually cited the Atlanta Fed's Wage Growth Tracker as one of the key metrics the Board uses in looking at and assessing wage growth. For you, as one of the architects of the Wage Growth Tracker, that had to be pretty cool to hear.
Robertson: Yes, I'll be looking for it in my own personal wage growth tracker. [laughter] But it's good when the work that a team does gets some recognition. I'm pleased that people find the Wage Growth Tracker useful. I think it's important to never focus on one particular metric as "the" metric, and so that's why all people involved in the Federal Reserve policymaking process are always looking at lots of different data. It's easy when all the data is all pointing in the same direction. It becomes much more challenging—and more interesting—when they don't, because then you have to figure out, well, why would they not all be telling me exactly the same thing?
Heintjes: As you noted, it's an ongoing puzzle to construct. But I think Chair Powell said it was one of the four key metrics they consider when assessing wage trends, so it's not "the" tool they use but it's definitely one of the core tools they consider—which I think is a real tribute to the work you've done.
Robertson: Thank you.
Heintjes: John, as always, I really enjoyed talking with you, and thanks so much for sharing your insights with us today.
Robertson: You're welcome, Tom. It's a pleasure.
Heintjes: And if you're listening to this, you'll surely want to visit the Atlanta Fed's Wage Growth Tracker. We'll have a link to it on our website, and it's the tool that serious wage watchers keep an eye on. And you should also bookmark our macroblog, where John and our other economists write about wages and many other economics topics.
And that brings us to the end of another episode of the Economy Matters podcast. I appreciate your spending time with us, and hope you'll join us again next month when we'll talk to Cynthia Goodwin of the Atlanta Fed's Supervision, Regulation, & Credit Division. We'll be talking about the year in banking, and it should be an interesting conversation, so be sure to come back for that. And thanks for joining us today.