Papers, Please #1: Various Papers on Employment, Wages and Productivity
For a while, I’ve been keeping a bookmark folder called ‘Papers, Please’ of all the papers I’d like to check out in the future. For those I do get to look at, I’ve compiled my observations, with the intent of making this another kind of roundup. I noticed a bunch of them were focused on questions of employment, wages and productivity, so it made sense to pull those out into a post, and stay on the lookout for similar groupings in the future as the section expands.
If there is a central theme here, it is the outsized role of norms, expectations and comparisons in determining outcomes, as well as underlying differences, and how such considerations seem to be neglected in many of the papers.
What interventions are implied, if any, although that is very much not the question I’m looking to ask here? Highly unfashionable ones.
Aggregate U.S. labor market dynamics are well approximated by a dual labor market supplemented with a third, predominantly, home-production segment. We uncover this structure by estimating a Hidden Markov Model, a machine-learning method. The different market segments are identified through (in-)equality constraints on labor market transition probabilities. This method yields time series of stocks and flows for the three segments for 1980-2021.
Workers in the primary sector, who make up around 55 percent of the population, are almost always employed and rarely experience unemployment.
The secondary sector, which constitutes 14 percent of the population, absorbs most of the short-run fluctuations, both at seasonal and business cycle frequencies. Workers in this segment experience six times higher turnover rates than those in the primary tier and are ten times more likely to be unemployed than their primary counterparts.
The tertiary segment consists of workers who infrequently participate in the labor market but nevertheless experience unemployment when they try to enter the labor force. Our individual-level analysis shows that observable demographic characteristics only explain a small part of the cross-individual variation in segment membership. The combination of the aggregate and individual-level evidence we provide points to dualism in the U.S. labor market being an equilibrium division of labor, under labor market imperfections, that minimizes adjustment costs in response to predictable seasonal as well as unpredictable business cycle fluctuations.
This seems right to me. If you can check off a standard set of boxes that make you capable of ‘normal work’ and you are willing to accept what the market has on offer, you will be highly employable. When you lose one job, you will be able to find another. It would take quite a lot of AI automation or economic decline to change this.
Whereas if you can’t check enough of the boxes, and you have some incompatibility with the jobs on offer and can’t find a niche that fixes this that you will accept, you will struggle. And it makes sense that there’s mostly a sharp line between these two groups, with a third group that doesn’t always want to work and is mostly in category two when they try.
Note that a large majority of workers can check off the necessary boxes. This rules out anything too onerous, or the possession of an especially valuable skill, as a barrier to joining the first group.
Is Pay Transparency Good? In some ways yes, in some ways no. Here’s the abstract.
Countries around the world are enacting pay transparency policies to combat pay discrimination. 71% of OECD countries have done so since 2000.
Most are enacting transparency horizontally, revealing pay between co-workers of similar seniority within a firm. While these policies have narrowed co-worker wage gaps, they have also lead to counterproductive peer comparisons and caused employers to bargain more aggressively, lowering average wages.
Other pay transparency policies, without directly targeting discrimination, have benefited workers by addressing broader information frictions in the labor market. Vertical pay transparency policies reveal to workers pay differences across different levels of seniority. Empirical evidence suggests these policies can lead to more accurate and more optimistic beliefs about earnings potential, increasing employee motivation and productivity.
Cross-firm pay transparency policies reveal wage differences across employers. These policies have encouraged workers to seek jobs at higher paying firms, negotiate higher pay, and sharpened wage competition between employers. We discuss the evidence on pay transparency’s effects, and open questions.
At this point I strongly believe that horizontal within-firm pay transparency is a large negative. If pay is known then it will be forced to reflect the status hierarchy. People will spend far more time and effort on comparisons. That means what I pay you will help determine what I pay everyone else, so I am forced to negotiate hard with everyone, and the effective marginal cost of giving out a raise has a large multiplier attached.
Pay transparency between firms plausibly leads to good competition, the danger being that it can effectively also lead to the dynamics of within-firm transparency. So you need to be careful to not give away too much specific information. Transparency on future expectations seems good, again if direct personal comparisons can be avoided.
The new result reported by the Economist is that pay transparency successfully narrowed the gender pay gap by making male wages go down.
Another misconception about pay-transparency laws is that they strengthen the bargaining power of workers. A recent paper by Zoe Cullen of Harvard Business School and Bobby Pakzad-Hurson of Brown University analysed the effects of 13 state laws passed between 2004 and 2016 that were designed to protect the right of workers to ask about the salaries of their co-workers. The authors found that the laws were associated with a 2% drop in wages, an outcome which the authors attribute to reduced bargaining power. “Although the idea of pay transparency is to give workers the ability to renegotiate away pay discrepancies, it actually shifts the bargaining power from the workers to the employer,” says Mr Pakzad-Hurson. “So wages are more equal,” explains Ms Cullen, “but they’re also lower.”
The model in people’s heads was ‘if workers can compare salaries, they will know to demand raises and negotiate higher salaries.’
The actual model is ‘if workers can compare salaries, management cannot negotiate.’
As an extension of the above two findings: I’ve previously stated my model for why labor markets have failed to adjust to the current labor shortage, and why letting people compare salaries often backfires. The model says that within a business, once they have taken the job, people care deeply about relative pay.
Everyone’s salary, when visible to others, is a claim about their relative status.
Thus, one cannot pay (for example) superstar programmers what they are worth.
Raising the minimum wage raises wages for many other workers, perhaps all.
The cost of raising your wage for new hires is likely also raising wages for most or all workers.
Thus, it often makes sense not to raise wages, even if the new hire is needed.
Under this model, when you pass pay transparency laws, you change relative pay from being based on market value to being based on social status and what can be justified.
An NBER working paper via MR uses a dynamic difference-in-differences approach and finds minimum wage increases fail to reduce probability of long term poverty, and also that less than 10% of those for whom a $15 minimum wage would bind live in poor families. The first result seems entirely unsurprising and I do not expect anyone to be convinced who wasn’t already convinced. The second result seems more potentially compelling, especially since I wouldn’t have expected it. Also its implications are rather bleak for the minimum wage.
Robust negative minimum wage effects on employment found, overturning Dube, Lester and Reich (2010), when commuting zones are used to define border regions. The core logical insight here seems sound.
Having had an incident of depression when surveyed at ages 27-35 predicts 10% lower hourly wages (conditional on occupation) and 120-180 fewer work hours annually, lowering total wages 24%. 25%-55% of the gap being attributed to ‘disruption of human capital,’ the rest largely to depression being a recurrent condition.
Who has an alternative hypothesis that explains this data? Anyone? Ooh ooh, pick me, pick me. Perhaps being depressed has something to do with your life being depressing, due to things like lack of human capital or job opportunities, life and career setbacks or alienation from one’s work. Income increases life satisfaction, as I assume does the prospect of future income.
It is amazing to see the ‘depression is purely a chemical imbalance unrelated to one’s physical circumstances’ attitude in this brazen a form. Mistaking correlation for causation here seems like a difficult mistake for a reasonable and reflecting person to make.
Electronic monitoring of workers does not improve their productivity, does stress them out. The stress part is in no way news. I could have seen productivity going either way here, but the null hypothesis does make the most sense.
Our finding of constant returns to marginal search effort, combined with limited spillovers onto other jobseekers, raises the possibility of suboptimally low search effort due to psychological costs of initiating applications.
I can no longer take this kind of language seriously. This is all overwhelming evidence of what we knew already, that job seekers do not do enough searching, even if one does not ‘believe one’s own press’ here. It would completely blow my mind if people were engaging in adequate job search.
What was their intervention?
Our main experimental treatment shifts how jobseekers communicate with the platform, lowering the psychological cost of initiating job applications. All users receive a monthly text message listing new vacancies that match their education, experience, and occupational preferences. These matches are determined by information jobseekers report at sign-up, before treatment assignment. Control group users must initiate job applications by calling the platform or asking the platform to call them. Treatment group users also receive a follow-up phone call after the text message that invites them to begin the application process, reducing the psychological cost of actively initiating a job application.
Yes. That does lower the psychological cost. It also raises salience, puts active social pressure on someone, makes it seem like the default and so on. I do not think it is reasonable to mostly call this ‘lowering psychological costs.’ Calling someone is one of our best known technologies for selling things to people, selling them on a job application is no different.
What are the results?
Treatment increases the share of jobseeker × vacancy matches getting applications from 0.2 to 1.5%. Using treatment as an instrument for applications shows that marginal treatment-induced applications have a 5.9% probability of yielding interviews. This is neither substantively nor statistically significantly different from the 6.3% probability for inframarginal applications from the control group, implying that returns to job search are approximately constant over this large increase in search effort.
Returns are also approximately constant for ‘value-weighted’ interviews, weighted by their desirability in terms of salary, hours, commute times, and non-salary benefits. This finding is not explained by differences in the ‘quality’ of jobseekers who submit marginal vs inframarginal applications: we reject this type of selection using observed quality proxies and we replicate our main finding using an additional within-jobseeker through-time randomization.
No. You can’t know that equal numbers of interviews implies equal value in applying, even if you hold compensation, hours and commute time constant. There are plenty of other reasons why someone might prefer one job to another. We also can’t equate interviews with marginal value of getting hired even if all jobs were random or identical, because we can’t assume zero correlation between applications. Job seekers reasonably assume that one rejection means likely additional rejections for similar jobs.
Economists (and other academics) have a long history of thinking that because the things they measured were the same, that two groups of things are the same.
They propose a theory of psychological cost of applications as the explanatory variable. This is no doubt one important factor, yet they do not earn their claims.
Even dumber, we have this:
Importantly, we do not find evidence that this additional search has negative spillovers on other jobseekers. We treat 50% of jobseekers on the platform, which increases total search by enough that quantitatively large spillovers are possible. Instead, we find that individual jobseekers’ interview probabilities are unaffected by competing against more treatment-induced applications from other users.
Seriously, what? It’s fine to say ‘each individual job seeker is wrong not to apply more.’ It’s also fine to say ‘increased applications improve job-applicant fit, reduce search times for employers and create value.’
You can’t equate interviews with jobs.
It’s not fine to say that applicants seeking out lots more interviews don’t hurt the job prospects of others because the others still get similar interview rates. Indeed, it is mathematically impossible for people who submit more applications to benefit in any real way here, without hurting those who don’t submit more applications. The only way would be to assume the job market improves so much, and employment improves so much, that even those now at a relative disadvantage are better off due to improved general matching driving growth. Which, in the short term, seems pretty ludicrous.
In a matching competition, you putting in more effort to match better and faster hurts others. Deal with it.
Re: #5: the second finding of minimum wage workers not being from impoverished families 90% of the time doesn’t surprise me at all (although I have done a lot of research on the matter, so I am cheating). The vast majority of minimum wage workers are high school/college students and stay at home spouses who want a part time job while their kids are in school or the like. Usually full time workers making minimum wage stop making minimum wage quickly if they are sticking with the job, or quickly move to a new job once they have proven themselves.
For example, I used to know a guy who head hunted fast food restaurant drive through employees. The good ones made 3-4$ per hour more than minimum wage and were actively sought, and the best made 5-6$ more, and this was back when the minimum wage was about 7$ or so (2000’s). Most fast food employees were kids on the first jobs and were bad at their job, but the good ones quickly got a premium to stay.
Condemnation of item #7 seems inadequately justified. It is plausible that their statistical controls were inadequate even in light of the online Appendix G, and personally I would have liked a lot more discussion because I have nothing like the background to judge. But it is unfair to imply they did not consider the hypothesis; they inadequately discussed it, but they do seem to consider that they have largely ruled it out by statistical means.