Labour economics · Contrarian
The average vocational programme in developing countries raises employment by about two percentage points. In a South African trial, certifying skills that job seekers already had, with no training at all, raised it by five. One of the most rigorous training experiments points to the certificate as the likely reason its gains lasted.
The default response to frontline unemployment is a training programme. The experimental evidence points to one reason such programmes underperform that their budgets rarely cover: the employer cannot see what was learnt. When researchers gave South African job seekers their skills results in a form employers could check, employment rose by 17 per cent. The same results given privately did nothing for employment. In Uganda, vocational graduates kept gains that firm-trained workers lost, which the authors say "likely stems from the fact that the skills of VT workers are certified and therefore can be demonstrated to potential employers".
Start with the case for training, because it is a good one.
When a labour market leaves large numbers of people out of work, the simplest diagnosis is that they lack what employers want, and the simplest remedy is to teach it. Training addresses that deficit directly. It is visible, politically legible and easy to count in enrolments.
The best meta-analytic evidence also supports it over the longer run. Card, Kluve and Weber, summarising estimates from more than 200 evaluations of active labour market programmes, found average impacts close to zero in the short run that become more positive two to three years after completion, with larger gains for programmes that emphasise human capital accumulation.1
On that account, disappointing early results are a matter of patience. Skills take time to turn into jobs, and the programmes that build the most skill pay off the most, eventually. I accept most of that. What it leaves out is the step between a skill and a job.
McKenzie's review of recent evaluations in developing countries is the most careful summary available. Across nine evaluations of vocational training, "the simple unweighted average across the studies is a 2.3 percentage point increase in employment", and many of the evaluations find no significant impact on employment or earnings at all.2
The standard explanations are quality, relevance and demand: the training was poor, it taught the wrong things, or there were no jobs to be had. Each is sometimes true. McKenzie adds a more sweeping one, which cuts against this paper as well as against training: "urban labor markets appear to work reasonably well in many cases, with fewer market failures than is often thought".2 If that were the whole story, a certificate would add as little as a course.
There is a further explanation, and it predicts something the others do not. If the binding constraint is that employers cannot see what a trainee has learnt, then an intervention that adds no skill at all, and only makes existing skill visible, should move employment. If it does, a programme that funds only the course has been paying for the half of the problem that was not binding. The experiments in section four test exactly that, and in the settings they studied they found one failure that does persist: employers cannot see skills that job seekers already have.
The mechanism is the one paper nine set out, applied to the moment a course ends. Training raises what a worker can do. Employment depends on what an employer believes the worker can do. Between the two sits a step many programme budgets leave out: evidence that a stranger can check.
Where that step is missing, three things follow. The newly trained worker enters a pooled market in which employers pay for the average capability of everyone who claims the skill. Any employer who does observe the new skill, usually the first to hire the worker, gains an informational advantage over every other employer, and with it the pricing power Acemoglu and Pischke described.3 And the worker, who bore the cost of general training in time and forgone earnings, is paid as though the skill were specific to one firm.
None of this says training is useless. It says the value of training is the value of the skill multiplied by the chance an employer can see it, and a programme that ignores the second term can deliver real skill and a small employment effect at the same time. That combination is consistent with what the averages show.
If the missing step is binding, adding it on its own should work. Three experiments did exactly that.
The first is the South African trial introduced in paper nine. Carranza, Garlick, Orkin and Rankin assessed 6,891 work seekers and randomised who received what. Those given results they could credibly share with employers saw employment rise by 5 percentage points on a control-group rate of 30.9 per cent, a gain of 17 per cent, with weekly earnings up 34 per cent and hourly wages up 20 per cent three to four months later, averaged across the group with those out of work counted as zero. Those given the same results privately changed their beliefs and search in similar ways, but "private certification has no effect on employment and raises earnings by less than public certification", by about 16 per cent against 34. Giving the results only to firms shifted which candidates the firms interviewed.4
What did visibility buy per 1,000 job seekers?
The trial figures come from one randomised study at three to four months; the training average spans nine evaluations in other countries and at other horizons, so the comparison shows scale, not a controlled contrast. Neither source reports costs in comparable terms, so no cost per job is stated.
The second experiment concerns references. Abel, Burger and Piraino ran an audit study of job applications in South Africa and found that including a reference letter from a former employer "increases callbacks by 60 percent", with women driving the effect. A second experiment, which encouraged job seekers to obtain and use a letter, found that letters raised interviews and employment for women. The letters worked, the authors argue, because they carry information about workers' skills that employers use to select more able applicants.5
The third tested a certificate for skills that are hard to observe. In Uganda, Bassi and Nansamba gave young job seekers certificates of their soft skills to show employers, in a sample of 787 workers and 422 firms. Over the following two years, employed workers with a certificate earned 11 per cent more, about US$7 a month, but the certificates "did not increase the probability of employment"; they reallocated workers across jobs instead.6
None of the three interventions included any training, though the Ugandan job seekers had already completed vocational courses, so there the certificate was added to training rather than substituted for it. The pattern across them is consistent in direction and uneven in size, and the uneven part matters. The Ugandan certificate moved wages and not employment, the South African certificate moved both, and the reference letter's effect was driven by women. A signal is not a universal employment lever. It is a price lever that sometimes moves employment too.
| Study | Setting | What was added | What moved | What did not |
|---|---|---|---|---|
| Carranza et al. | South Africa; 6,891 work seekers | Skills results that employers could check | Employment +17%, weekly earnings +34%, hourly wages +20% at three to four months | The same results given privately: no employment effect |
| Abel et al. | South Africa; audit and encouragement experiments | A reference letter from a former employer | Callbacks +60%; interviews and employment for women | Effect driven by women |
| Bassi and Nansamba | Uganda; 787 workers, 422 firms | A certificate of soft skills | Wages +11% among the employed over two years | Probability of employment |
The value of training is the value of the skill multiplied by the chance that an employer can see it.
The best evidence against this paper's thesis is also the best evidence for training, so it deserves a full hearing.
Alfonsi, Bandiera, Bassi, Burgess, Rasul, Sulaiman and Vitali followed 1,700 young workers and 1,500 firms in Uganda for four years, in a setting where youth unemployment exceeded 60 per cent. Workers were offered six months of either classroom vocational training or training provided by firms. Relative to the control group, both groups gained large and similar amounts of sector-specific skill, both improved their employment rates, and both improved on an index of labour market outcomes.7 Training worked.
The dynamics are where the paper becomes decisive for this argument. Gains from firm-provided training "materialize quickly but fade over time", while gains from vocational training "emerge slowly but are long-lasting", so that the vocational group's employment and earnings eventually rise above the firm-trained group's. The authors' job ladder model locates the reason in job offers: vocational trainees receive offers at a higher rate when unemployed, which "likely stems from the fact that the skills of VT workers are certified and therefore can be demonstrated to potential employers".7
So the most rigorous training result I know of does not break the thesis; it sharpens it. Two groups gained similar skills. The group whose skills could be demonstrated to a stranger kept its gains, and the group whose skills were known to the firm that trained them saw theirs fade. Training without a signal did not fail outright. It failed to last, and for frontline workers, whose careers run through a sequence of employers, a gain that cannot travel is a gain that does not last.
One caution belongs here. The firm-training arm was delivered through wage subsidies to firms, so some of its fade may reflect the subsidy ending rather than the absence of a certificate. The authors' own model points to job offers after unemployment, which is where a certificate would matter, but the design cannot separate the two cleanly.
The strongest case against putting the signal first is displacement. A certificate that helps one job seeker win a post may simply move that post away from another, uncertified job seeker. The Ugandan certificate result, higher pay through reallocation across jobs with no change in the probability of employment, is what displacement would look like, and the South African trial measured individuals rather than the labour market as a whole. The researchers' own audit of real vacancies found that the advantage of a certified application shrank as more of the competing applications carried certificates too.4
That concession should be made plainly. None of the three signalling experiments shows that signals raise total employment. What they show is that signals change who is hired and what they are paid, in favour of people whose skills were previously invisible. Better matching is a real gain even when the number of jobs is fixed, and the South African researchers found that shareable results also improved the alignment between workers' skills, beliefs and search strategies.4
The second objection is horizon. The South African effects were measured at three to four months, and training's advantage in the meta-analysis appears two to three years out. A signal might win early and lose late. The Ugandan experiment runs long enough to test that for training, but it compares two kinds of training and cannot say whether a signal on its own lasts. An Ethiopian experiment can. Abebe and colleagues gave young job seekers in Addis Ababa a job application workshop designed to help them signal their skills, and compared it with a transport subsidy. "Four years later, the workshop significantly increases earnings, job satisfaction, and employment duration, but the effects of the transport subsidy have dissipated." Their conclusion states this paper's premise in their own words: "young people possess valuable skills that are unobservable to employers".8
The third is that a certificate cannot create a skill that is not there. That is true, and it is why this paper argues about sequence and bundling rather than substitution. The failure mode is not training. It is training that ends without a signal.
For Saudi Arabia and the wider GCC there is a larger gap than any of these. I found no randomised or quasi-experimental evaluation of a Saudi or Gulf skills programme. The only published analysis of outcomes for graduates of the Technical and Vocational Training Corporation that I located is a World Bank tracer study, which describes what graduates went on to do and has no comparison group.9 Every claim about what training achieves in the region, this paper's included, is imported from elsewhere.
If the argument holds, three things change for anyone funding frontline skills.
Every training budget needs a signal line. That means an assessment at the end of the course, scored against a published standard and recorded in a form that an employer who has never met the worker can check. Without it, a programme pays for skill and leaves the price of that skill to chance.
Evaluate by visibility, not only by completion. The South African design, which randomised whether results could be shared, is simple to copy. A Saudi programme that randomised the form of its certificate would answer within months the question this paper can only raise for the region.
Fund the signal for skills people already have. The fastest effects in this paper came from certifying existing capability rather than building new capability. For a workforce that already holds years of unrecorded experience, the most productive employment programme available may turn out to be an assessment.
Disclosure
The case above stands on its own evidence; nothing in it depends on what follows. I am the founder of UpSkillMe, which builds assessment and a worker-owned capability record for household work in Saudi Arabia and the trades in South Africa, so this paper argues for the half of the problem my company works on. The reader should weigh section seven with that in mind. The studies cited are public, were designed and run by researchers with no connection to us, and stand independently of our interest in them.
Next in this series: what a skills passport must contain, and why the difference between a passport and a registry is who holds it.
Written in British English; source titles and quotations keep their original spelling. Effect sizes come from different countries, samples and horizons and are compared for scale only. Point estimates for the two Ugandan training arms that circulate in secondary commentary disagree with one another and were not used; the dynamics are described from the published abstract. South African earnings and wage effects are averaged over all participants, with those out of work counted as zero. No randomised or quasi-experimental evaluation of a skills programme in Saudi Arabia or the GCC was located.