When flexibility becomes infrastructure

This series has examined one question: did the post-pandemic expansion of work from home change employment for people with disabilities, and if so, what kind of change was it? The strongest answer supported by the evidence is neither that remote work is merely a preference nor that it solves disability exclusion. WFH appears to have lowered participation barriers and expanded full-time employment for people with physical disabilities in the U.S. population and occupations studied, with labor-supply mechanisms especially plausible. The judgment that follows is conditional: treat flexibility as accessibility infrastructure when it changes a real barrier in a job that can be performed remotely, while preserving other accommodations and resisting claims that the average effect is universal.

The evidence forms one access model

The BLS 2025 release gives the problem its scale. It reports an employment-population ratio of 22.8% for people with a disability age 16 and over, compared with 65.2% for people with no disability. The gap is not simply a comparison between employed and unemployed people: about three-quarters of people with a disability were not in the labor force, and the population with a disability was much older than the population without one. The release's functional CPS definition and its separate labor-force categories show why the statistic is a signal about labor-market position, not a diagnosis of individual capacity.

That descriptive starting point becomes a causal question only after the population and comparison are narrowed. The IZA study examines people ages 18 to 64, focuses on five disability categories whose combined population was relatively stable before and after COVID, and compares January 2018–December 2019 with July 2022–June 2024. It leaves out the pandemic years and excludes cognitive difficulty from the baseline so that a changing composition of reported disability is less likely to account for the result. The study's “physical disability” label is useful for its stable-composition design, but it does not replace the broader CPS classification.

The institutional change is WFH. Before the pandemic, about 7% of full workdays in the United States were worked from home; by 2023, the paper reports 28%. That shift was not designed as a disability program. It occurred because firms reorganized work for the broader workforce. Its accessibility effect could nevertheless be unequal because a standard workplace imposes different hidden costs on different people. Commuting, moving through a building, tolerating noise or lighting, controlling a desk setup, and working fixed hours can be part of the practical price of a job even when they are absent from the job description.

The access model has a gate that matters: occupations and employers determine whether the location of work can change. WFH expanded much more in computer occupations than in teaching, and the paper uses that variation to test whether disability employment rose more where remote opportunities expanded more. A remote-capable task is not enough by itself. The employer must make a usable arrangement available, and the worker must have circumstances in which it actually lowers the relevant barrier. This is why “WFH” is not a complete explanation; the meaningful object is a change in the conditions under which labor can be supplied.

The researchers connect those pieces at the occupation level. They measure the change in WFH among workers without a disability and compare it with the pre/post change in employment among workers with a physical disability. Using non-disabled workers avoids defining the exposure with the same group whose employment outcome is being explained, which reduces a direct reverse-causality concern. Controls for non-disabled employment growth address broad occupation-level labor demand, and composition reweighting addresses measured changes in age, gender, race, and education. A pre-pandemic measure of how feasible an occupation was to perform from home supplies an additional source of predicted WFH variation.

The design does not create a randomized individual experiment. It makes a cross-occupation relationship more credible by connecting different data branches at the occupation level and by testing several plausible alternatives. Its identifying assumptions remain visible: occupation-level WFH changes must not be primarily driven by disability-specific shocks, and pre-pandemic remote feasibility must affect disability employment mainly through WFH after the controls. The result applies most directly to the defined working-age sample and to the average occupation-level relationship, not to a guaranteed outcome for a particular worker or employer.