How to tell an accessibility effect from a pandemic coincidence

This chapter reconstructs how the IZA study tests the relationship between work from home and disability employment. Its focus is the design itself: what counts as WFH exposure, what counts as the employment outcome, which comparisons are made across occupations, and which alternative explanations the researchers try to remove. The value of following those connections is practical as well as statistical. It lets readers treat the later estimates as conditional evidence with known units and assumptions, rather than as proof that any individual worker was helped by any remote-work policy.

The comparison is across occupations, not people paired to policies

The study does not randomly assign disabled workers to work from home and then compare them with a control group. Instead, it uses the fact that occupations experienced different increases in WFH after the pandemic. Some occupations became much more remote-capable in practice; others changed little. The researchers ask whether disability employment increased more in the occupations where WFH opportunities expanded more.

This comparison joins two different kinds of information. The outcome comes from monthly Current Population Survey data on people with a physical disability ages 18 to 64. For each occupation, the researchers compare the number employed before the pandemic—January 2018 through December 2019—with the number employed after the pandemic disruptions—July 2022 through June 2024. The outcome is the percent change in disability employment in that occupation between the two periods.

The exposure is not the change in WFH reported by disabled workers themselves. It is the change in WFH among workers without a disability, measured mainly through the American Community Survey's transportation-to-work question. The researchers compare the share of non-disabled workers in an occupation who answered “worked from home” in 2019 with the corresponding share in 2022. The difference is the occupation's change in WFH opportunity for the purposes of the main analysis.

That separation is the design's first important interface. The disability sample supplies the outcome; the non-disability sample supplies the exposure. If disabled workers have an especially strong demand for WFH, using their own WFH rate could make the exposure partly reflect the outcome the study is trying to explain. A rise in disability employment could lead employers to offer more remote work, or disabled workers could sort into occupations that already offer it. Measuring the change among non-disabled workers makes it less likely that disabled workers' own behavior mechanically defines their occupation's WFH exposure. The authors note that people with disabilities made up about 3% of the employed workforce in the pre-pandemic period, which supports—but does not prove—the assumption that they did not drive the broad adoption of WFH.

The unit of comparison is consequently the occupation. The baseline data are aggregated to two-digit Standard Occupational Classification groups, sparse groups are merged, and the estimates cover 86 occupations. One observation is not “a particular worker received a remote option.” It is an occupation-level change in disability employment paired with an occupation-level change in WFH among non-disabled workers.