Workforce Wonkery. Policy. Data. Practice. Decoded.
National Practice Deep Dive · WorkAdvance

The placement report is not the end of the story

A placement report is a milestone. How much does it tell us about what happens to earnings years later?

Read the original MDRC evaluation

Four within-provider randomized comparisons; no head-to-head provider ranking or proof of a specific mechanism.

Prepared October 10, 2026 · Independent reader review waived for this collection; no independent reader test was conducted.

The Decode (30 seconds)

We often judge training when someone gets a job. But the story may not end there. A study followed 2,564 people at four training providers for ten years. At Per Scholas, a clear earnings gain in Year 7 was no longer clear in Year 10. At St. Nicks Alliance, the offered group’s average annual earnings were $8,580 higher than its own comparison group in Year 10. Its Year 7 gain had not been clear. The study cannot say why these results differed, or prove the impact changed over time. A better question for a workforce board is: what will we learn about wages after the first job?

The story (3-minute overview)

A placement can be a genuine achievement and still be an incomplete measure of success.

Imagine two sector partnerships reporting back to their boards. One has strong early wage gains; the other has modest results. Which one should receive the next investment? The usual answer might be obvious, until you ask a question that few contracts can answer well: what happens several years later?

That is what makes the WorkAdvance evaluation so useful. The study began with 2,564 adults who enrolled between 2011 and 2013 and were randomly assigned either to be offered WorkAdvance or to a comparison group that could use other community services. Four nonprofits carried out versions of the same model in different places and industries. The package combined career preparation, occupational training, employer connections and support for advancement after a job started.

MDRC returned to the results over time. Its March 2022 seven-year report found a significant average earnings gain at Per Scholas, the Bronx-based IT provider. Its September 2025 ten-year follow-up found that Per Scholas still had a positive observed difference, but it was too uncertain to call a statistically detectable impact at that point. Meanwhile, St. Nicks Alliance, a Brooklyn provider focused on environmental remediation, moved from no detectable average earnings effect in Year 7 to a sizable one in Year 10. Seven-year MDRC report · Ten-year MDRC report.

The contrast is worth seeing together:

ProviderYear 7: difference in average annual earningsYear 10: difference in average annual earnings
Per Scholas, Bronx, IT+$5,123, p=0.027+$1,971, p=0.489
St. Nicks Alliance, Brooklyn, remediation+$2,608, p=0.264+$8,580, p=0.005
Madison Strategies Group, Tulsa, transportation/manufacturing+$231, p=0.880−$690, p=0.707
Towards Employment, northeast Ohio, healthcare/manufacturing+$998, p=0.454+$1,493, p=0.392

Source: MDRC, 2025 full report, Tables 2–5, pages 10, 15, 20–21. All comparisons are with each provider’s own control group. Bold marks a result meeting the common 5% statistical threshold (p below 0.05). A p-value tells us about uncertainty, not whether a program will work everywhere. A result above 0.05 is not proof of no benefit.

This does not mean St. Nicks “beat” Per Scholas. The evaluation compared people offered each provider’s program with that provider’s own control group. It did not randomly assign the same people between providers or isolate which one of WorkAdvance’s services caused a later difference. The researchers reported yearly estimates, not a statistical test proving that the size of one provider’s impact changed between Year 7 and Year 10. Different sectors, participants and employer conditions also matter.

Even so, it raises an uncomfortable but productive question for the rest of us: how often are we making long-term judgments with short-term evidence?

Go deeper (10-minute reading)

What actually changed for participants?

WorkAdvance was more than an occupational course. Its five connected elements were:

  1. Screening for readiness and fit.
  2. Career-readiness support.
  3. Occupational training that could lead to a credential.
  4. Job development and placement.
  5. Retention and advancement help after employment.

The fifth element is easy to admire on paper and harder to maintain when the current contract, caseload and funding cycle are all calling for attention.

This was real operating work. The nonprofits had to build employer relationships, adapt to local industries and maintain multiple forms of support. MDRC’s 2016 implementation review reported operational expenditures of $5,200 to $6,700 per participant at the four sites. That is a historical program-cost range, not a 2026 California bid and not the price of maintaining earnings records for ten years. Some parts of the model also took more than a year to fully establish. 2016 MDRC findings.

The target sectors were not interchangeable: IT at Per Scholas; environmental remediation at St. Nicks; transportation and manufacturing in Tulsa; and healthcare and manufacturing in northeast Ohio. These differences matter when the economy changes. A worker’s skill may gain or lose value, a control-group member may find training elsewhere, or a local industry may weather a downturn differently.

Four things the ten-year study cannot settle

The pandemic affected the study years. Depending on when someone entered, Year 10 fell in 2021, 2022 or 2023. The COVID-19 shock changed working conditions across industries. The researchers considered possible explanations, but the earnings records do not tell us which sector each participant was working in a decade later.

One measure changed between reports. The original Year 7 report examined the share earning $40,000 or more; the Year 10 report selected $45,000 or more as its high-earnings outcome. These are not the same threshold. In Year 10 at St. Nicks, 34.6% of the WorkAdvance group crossed $45,000, versus 27.6% of controls, a seven-point difference. The reported p-value was 0.091, so that result does not reach the conventional 5% significance level, even though MDRC marks it at its 10% threshold. The stronger Year 10 finding is the average earnings difference (p=0.005).

The cumulative earnings story is promising, but more uncertain. MDRC estimated that each site’s participants earned more in total over the full decade than controls. It explicitly did not run statistical significance tests for these cumulative estimates, because constructing them required combining different data sources and making assumptions. The findings should not be presented as four proven decade-long causal gains. The annual randomized impacts and their p-values give a firmer basis for this story. 2025 full report, Box 1 and methods.

The earnings records leave parts of career mobility unseen. The National Directory of New Hires records cover unemployment-insurance employment but not all self-employment or contract work, and they do not show whether a worker stayed in the original training sector. No one should read the earnings result as a direct count of promotions or an explanation of every career path.

Those limits do not make the research unhelpful. They tell us what a reasonable decision can and cannot rest on.

The California translation: track a pathway, not just a first job

California workforce boards do not need to run a ten-year randomized trial before improving how they learn from their own training investments. They can begin with an honest review of one existing sector pathway.

Borrow now: Alongside placement and near-term wages, ask whether participants can move to better work. Identify the job ladder the training promised, the next skill or credential on that ladder, and who could reconnect a worker when opportunities change.

Adapt first: Choose one clearly defined cohort and decide which follow-up measures are actually available and authorized. It may be realistic to review verified advancement at 24 or 36 months. It may not be possible to link decade-long wage records. Be clear about missing people, missing data, and the cost of keeping contact. Do not put identifiable participant information into an unapproved workbook or shared system.

Explore further: Put provider capacity and labor-market change alongside outcomes. Did employers still need those skills? Were second-step credentials accessible? Was there a funded person responsible for advancement support? WorkAdvance cannot tell us that any one of these steps caused the later earnings pattern, but it gives us good reasons to investigate them.

Before the next sector-training renewal, the director and program manager could ask the provider for a one-page advancement learning plan, not a promised ten-year return. One page is enough to identify:

  • Who: The original training cohort, including people who did not complete or get placed.
  • What: One clearly defined advancement outcome, and what data will be missing.
  • When: Feasible follow-up dates, such as 24 or 36 months, rather than a promise of ten-year tracking.
  • How: An approved data source, an owner for the follow-up work, and the staff time or funds needed.
  • Then what: A decision date and one finding that would justify keeping, changing or stopping the pathway.

This is a meeting prompt, not a tested or authorized data-collection form. Fiscal, privacy and partner responsibilities still need local approval.

One question for the next contract discussion

If the first placement went well, what evidence would tell us three years later that the pathway is still helping people advance?

The answer may be a new course, a stronger employer commitment, help reaching the next credential, a reengagement call, or simply better data about what is happening. Do not pick the remedy before learning where progress slows.

WorkAdvance reminds us that an encouraging early result deserves attention, but not a permanent victory lap. An underwhelming early result deserves scrutiny, but not an automatic verdict. Time does not replace evidence. It changes which evidence we need.

Sources and interpretation