Dynamics of Occupation Choice, Skill Investment and Output Pricing with Multidimensional Skill

Project Outcomes Statement

The project aimed to understand how the labor market responds to technological change. This research is embedded in an even larger research program aimed at understanding economic disparities and how they are affected by policy.

We began by examining how individual workers should be expected to respond to technological change that increases the value of some skills while decreasing the value of others. For example, word processors reduced the demand for typists but probably increased the demand for people who could write well. Older workers who have invested heavily in typing skills and skills complementary to typing will generally try to remain in similar jobs despite the reduced demand. Somewhat younger workers will invest in the complementary skills and seek jobs where those skills are relatively important. The youngest workers will take advantage of the new opportunities. Modeling this process formally produces some surprising results. 1) Technological change that is broadly beneficial to workers can have significant adverse effects on relatively young workers because workers invest heavily in skills early in their careers. 2) Liquidity constraints that prevent workers from making desirable investments may be hard to detect. 3) Employers may have an incentive to invest in their workers’ skills even if they are fully transferable to other jobs.

Our more recent work has focused on understanding the nature of technological change. Using broad categories of skills (manual, social, abstract, routine/adaptability), we examine the change in skill use within and between jobs over roughly sixty years. We find that changes in skill use from how workers do a given job often dwarf changes due to shifts in the job distribution. We also examine how the cost of acquiring skills relates to workers’ education. We find that the major advantage highly educated workers hold is adaptability, the opposite of what some researchers call routineness.

The research discussed above examines technological change experienced by workers. However, technological change is often anticipated but uncertain. We were motivated by the likely advent of self-driving trucks. We show that when technological change is on the horizon, but its arrival date is uncertain, wages will rise, employment will fall, entering workers will be older, and exiting workers younger in the occupation that will be adversely affected. We show that these predictions are consistent with what happened to workers who drove horse-drawn trucks at the dawn of the motor truck. There is some evidence of similar developments in truck driving.

Our focus on disparities has led us in a couple of unanticipated directions. We examine how monitoring technology can affect racial disparities. Under some conditions, even if the productivity distribution is independent of race, equilibrium may entail heavier monitoring of black than white workers. This equilibrium is self-sustaining. The model leads to several predictions, of which the most unique is that recently hired black workers will be more likely than white workers to be fired but that this “firing gap” will fall with seniority, a result we confirm empirically.

Complementary to our results on liquidity constraints, we examined the market for illegal moneylending. Using a unique data set of over 1,100 clients of loan sharks, we provide basic information about this market and its customers. We show that borrowers and lenders are constrained by their desire to continue transacting with each other. This “relational capital” reduces the loan shark’s need to rely on extra-legal enforcement because borrowers are anxious to repay even without threats. Likewise, the lender does not want the borrower to switch to other lenders for future loans.

Finally, our broader program on disparities led us to a meta-analysis of policy research and hypothesis testing in economics. We treat the distribution of test results as coming from a mixture of (nearly) true and false null hypotheses. We put sufficient structure on the model to estimate the fraction of rejected null hypotheses that are falsely rejected. In passing, we also cast light on the presence of p-hacking. This term refers to adjusting the sample or the statistical approach to ensure the result is statistically significant. We find no evidence of p-hacking among papers published in leading economics journals. However, we find that nearly half of rejected null hypotheses are falsely rejected. This proportion is even higher when the rejection is narrow. To be clear, we do not suggest that economists engage in unethical behavior. Instead, we argue that economists investigate many policy effects. If we study enough policies and outcomes, we will find many results that are unlikely when examined in isolation. The results suggest strong caution when acting on a single published study, even if it is well-executed and published in a good journal.