The standard list of barriers facing marginalized workers — credential gaps, transportation, childcare, bias, criminal records — is correct but unranked. That ranking matters, because the workforce-development sector has roughly a fixed amount of money to spend, and it tends to allocate that money to the barriers it understands best (training) rather than the ones with the largest effect sizes (job networks, geography, and what economists call "frictions"). The result is a system that produces credentialed workers who still cannot find their way into the jobs the credentials were supposed to unlock.
The honest argument is that workforce development has under-invested in connection and over-invested in instruction. Recent research from Raj Chetty and the Opportunity Insights team at Harvard, published in Nature in 2022, found that "economic connectedness" — the share of high-income friends a low-income person has — is the single strongest neighborhood-level predictor of upward mobility identified in any large-scale study to date, outperforming school quality, family structure, and even local labor-market strength. If a marginalized worker's most important missing asset is a network, the standard workforce-development response (a six-week training course) is targeting the wrong constraint.
Geography is more determinative than the sector admits
The data on where good jobs sit, and where unemployed workers sit, is brutally consistent. The Brookings Institution's Metro Monitor and Federal Reserve regional studies have documented for years that "labor market thickness" — the density of nearby employers in a worker's skill cluster — explains a large share of the wage gap between similar workers in different regions. A welder in a thick metals-manufacturing labor market earns substantially more than the identically credentialed welder one state over. Workforce dollars that train someone in a region with no employers for that skill are not just inefficient; they are actively misleading.
Marginalized communities are disproportionately located in thin labor markets. The 2024 Economic Policy Institute analysis of Black labor force participation found that residential segregation, not skill gaps, accounted for a majority of the unexplained Black-white employment differential among prime-age workers in the metros studied. The implication is uncomfortable: place-based interventions (transit, broadband, relocation assistance) and employer-side interventions (where the jobs sit) will move outcomes more than another tranche of certificate programs.
The credential trap
This connects to a problem the field rarely names. Many workforce programs are measured on completion and short-term placement — six-week credential earned, ninety-day job logged — because those are the metrics that funders can verify. Long-run wage growth, the metric that would actually tell you whether a program worked, is harder to track and easier to lose in the political cycle. The most rigorous longitudinal evaluations — for example, the Department of Labor's Workforce Investment Act impact studies and MDRC's WorkAdvance evaluations — generally find that programs heavy on sector-specific employer partnerships and "navigation" produce real long-run wage gains, while generic short-credential programs produce close to zero. That is not a secret. It is, however, easier to fund the second kind.
Bias and the hiring funnel
Bias remains a binding constraint, but the empirical literature has narrowed where it bites. The 2017 meta-analysis by Lincoln Quillian and colleagues in PNAS pooled every résumé-audit study from 1989 to 2015 and found that callback discrimination against Black applicants had not declined measurably over 25 years — about a 36% lower callback rate for identical résumés. Comparable audit work from the Urban Institute on formerly incarcerated applicants finds even larger penalties, with one prominent estimate cutting callbacks by half.
The interventions that have actually moved that number — as opposed to the ones that perform inclusion theater — are surprisingly few. Frank Dobbin and Alexandra Kalev's work, summarized in their Harvard Business Review piece "Why Diversity Programs Fail" and the follow-on book Getting to Diversity, shows that unconscious-bias training in isolation does not improve representation; structured mentoring, accountability metrics tied to managers' performance reviews, and explicit succession planning do. The implication for workforce development is that pushing candidates into employers without structural commitments at the employer end produces poor placements and high turnover.
What better looks like
An honest reform agenda for inclusive workforce development would do five things simultaneously. First, fund and report long-run wage outcomes — 24-month, 36-month, 60-month earnings — for any program receiving public money, the way the Workforce Innovation and Opportunity Act's outcome reporting is finally beginning to do. Second, weight investment toward sector partnerships in thick local labor markets (manufacturing in the Midwest, healthcare almost everywhere, semiconductors in the CHIPS-corridor metros) rather than spreading thinly across every region equally. Third, invest in "navigators" — paid case managers with employer relationships — because the audit evidence and Chetty's connectedness research both point to network access as the missing input. Fourth, condition employer subsidies on retention and promotion data, not just hire counts. Fifth, treat childcare, transit, and housing as workforce policy, because they are; the OECD's Employment Outlook 2023 documented that U.S. childcare costs are now the single largest barrier to women's labor force participation among OECD high-income countries.
For the NWLB-specific framing, our Racial Equity at Work → pillar reviews how the post-SFFA legal environment is changing what employers can and cannot do here. The short version: race-conscious recruitment is now legally constrained, but place-conscious, sector-conscious, and class-conscious programs are not — and most of the highest-leverage interventions were always one of those three.
Stop measuring effort. Start measuring lifts.
The hardest sentence to write in a field that is mostly populated by people of goodwill is that effort is not the binding constraint either. There is no shortage of well-intentioned programs. There is a shortage of programs willing to publish their five-year wage data, partner with employers who will commit to structural changes, and concentrate resources where the labor market is thick enough to absorb their graduates. That is a strategy problem, not a values problem, and the field would do better by treating it as such.
Workforce development has spent two decades training workers for jobs that aren't where the workers are. The next decade should spend at least as much on closing the geographic and network gap as it does on closing the credential gap.
Updated May 21, 2026. This piece was substantively rewritten as part of NWLB's 2026 editorial refresh.
