There is one randomised controlled trial of a coding bootcamp in an emerging market. It was run by the World Bank across Buenos Aires and Bogotá, 803 participants, 402 of them offered a scholarship, and it reported two effects.
The programme raised the probability of working in a technology job by 9.2 percentage points — a 38% improvement over the control group — and raised the probability of passing an employer-style coding exam by 20 points. Those are the numbers a training provider would put in a report.
It also lowered the probability of being employed at all by 9.8 percentage points.
The authors call that second result surprising, and explain it: participants left the labour force to keep studying. They were not unemployed in the discouraged sense. They were out of work because the programme had persuaded them to retrain further.
Both effects are real, both are causal, and a programme that reported only its placement rate would have found the first and never looked for the second.
Why the second number is invisible almost everywhere
Nearly every training outcome published in this region is a completion figure or a placement figure, measured on the people who finished. That design can only find people who got jobs. It cannot find people who left the labour market, and it cannot distinguish the participants a programme helped from the participants who would have got the job anyway.
The operators who are careful about this say so. Correlation One ran a cybersecurity programme for women across Jordan and Egypt where employment among participants rose from 35% at baseline to 66% at endline — a result most providers would lead with. Their evaluators put the limit in writing: the improvements "are consistent with — but not proof of — program impact", because the design was a cross-sectional measurement after the fact with no control group.
That is an operator publishing the ceiling on its own favourable number, which is rare enough to be worth naming.
The same organisation's larger Jordan programme, Tech for Jobs, has enrolled 4,360 and graduated 3,630, and reports 2,164 placements into new or better work, 76.1% of them participants who had no work at baseline. Those are monitoring figures. The programme was also designed from the outset as a randomised trial with J-PAL MENA — three arms, 6,199 participants, follow-up surveys at 12 and 24 months — and that evaluation has not published. So the best-instrumented training programme in Jordan has a causal answer coming and does not have it yet.
Everyone else is reporting the numerator.
The pay result nobody quotes
The World Bank trial measured one more thing that complicates the regional pitch.
Technology jobs in the sample paid about the same as non-technology jobs. The earnings difference was not statistically significant. What did improve was the shape of the work: technology jobs were 24 percentage points more likely to offer remote work and flexible hours, and participants reported 12% higher job satisfaction.
That is a meaningful outcome, especially for women's labour force participation, and it is not the outcome being sold. Training programmes across the region are pitched on earnings. The only trial that measured earnings found the gain was in flexibility.
Two caveats belong next to this. The follow-up data refers to February 2020, two to five months after completion, so it measures the short run only. And baseline employment in the sample was 85% — these were mostly women already working, switching into technology, not entering from unemployment. Neither caveat rescues the earnings claim.
What Jordan's own report says about the other end
The capability question sits underneath all of this, and Jordan has measured it.
The Digital Skills Supply and Demand Gap Analysis, produced by the DigiSkills Association with the Ministry of Digital Economy and Entrepreneurship, finds that entry-level candidates meet employers' hiring criteria in fewer than 20% of cases, and that 35% to 40% of ICT graduates are absorbed into relevant employment within their first year.
The same report says employers rank communication, teamwork and adaptability above coding, and 77% report graduates falling short in those areas specifically. The binding deficit named by Jordanian employers is not technical.
Set that against the national figures. Unemployment among Jordanians holding a bachelor's degree or higher was 25.8% in 2024, above the overall rate of 21.4%. In Jordan, more education is associated with a higher chance of being unemployed, which is the clearest possible signal that the problem is not quantity of graduates.
What this means for how we read a programme
We are asked to look at training-adjacent companies regularly, and we run programmes ourselves. The lesson from the one trial that exists is not that bootcamps do not work. It is that placement rate is the wrong instrument, because it is structurally incapable of detecting the cost the trial found.
Three questions do more work than a placement rate.
What happened to the people who did not finish. A completion-based number excludes them by construction, and they are the group most likely to have been harmed by the time commitment.
What the comparison group did. Without one, a 35%-to-66% employment jump and a strong local labour market are indistinguishable.
Whether anyone measured labour force participation, not just employment. That single variable is what separated the two headline findings of the World Bank trial, and almost nobody in this region collects it.
Jordan produces graduates that fewer than one employer in five will hire at entry level, and funds programmes measured by how many of their graduates got hired. Those two facts are answers to different questions. Until the J-PAL evaluation lands, the honest position is that we do not know what the region's training spend is buying, and the one place it was measured properly it bought something other than what was advertised.