How Mismatch Gets Measured—and Misread
Once credentials and capabilities are treated as separate dimensions, the next question is methodological: how do we know what a job requires and what a worker can do? This chapter compares the main ways educational mismatch, skills mismatch, and labour-market demand are measured. The practical goal is not to crown one method as the truth, but to identify what each observes, where it is likely to mislead, and what additional evidence is needed before a number supports a decision.
A mismatch statistic is never just a count. It is the result of choosing a reference point, a respondent, a classification, a time period, and a definition of “required.” Change any of those choices and the apparent size or location of the problem may change with it.
Three ways to infer the education a job requires
The most direct-looking approach is the subjective method. A survey asks workers what level of education was needed to get the job or to do it satisfactorily, then compares that answer with the worker’s highest qualification. This method is relatively easy to include in a questionnaire and can distinguish a hiring requirement from a performance requirement. It also depends on the respondent’s interpretation, memory, and willingness to describe their position accurately. Question wording is not a minor technical detail: “needed to get” can capture a screening rule, while “needed to do” is closer to the educational content of the work.
A second approach uses realised matches. Instead of asking what a job requires, the researcher looks at the mean or modal education level of workers in an occupation and treats that pattern as the occupation’s typical requirement. The method can be applied to existing labour-force or administrative datasets containing education and occupation, which makes it convenient for comparisons. Its convenience is also its central limitation. The typical qualification of workers is not the same thing as the knowledge required by every job they hold.
An occupational average can be pulled toward older workers who entered their jobs under earlier requirements. It may also be calculated for a broad group that combines positions with different tasks. If newer entrants face higher requirements because work has changed, the average can make those entrants look overeducated even when their qualifications are appropriate for current practice. The method describes a realised pattern of credentials; it does not directly inspect job content.
The third approach is job evaluation. Professional analysts assess the educational requirements of occupations for occupational dictionaries or classification systems. This can bring more explicit attention to the work itself, but it is expensive to carry out at national scale and requires regular updating as occupations change. Expert judgement does not eliminate subjectivity; it makes the judgement more structured and potentially more consistent. Broad classifications such as ISCO also involve assumptions about whether jobs with the same title require the same education across countries, and they tend to give limited attention to informal training.
The consequences of these choices are visible in the estimates. One comparison cited in the ILO review reported an Austrian overeducation rate of 9.6% using a subjective approach and 1.1% using an empirical approach. These figures come from different approaches and should not be assembled into a single Austrian time series. Their value is diagnostic: disagreement can arise from the estimator rather than from careless arithmetic. The review also reports cases in which the ranking of countries changes depending on the method used.
That is why a precise percentage can still be a weak guide to action. Before comparing two rates, ask whether they use the same definition of the job requirement, the same occupation boundaries, the same population, the same time period, and the same treatment of entry versus performance requirements. If not, a difference may reflect the measurement design rather than a real difference in labour-market fit.
The choice is not simply between “subjective” and “objective.” A worker’s report can reveal a requirement that no administrative dataset records. An occupational average can provide broad coverage that a costly job evaluation cannot. A professional classification can make job content explicit while becoming stale if not maintained. Each method trades coverage, directness, cost, and timeliness. There is no single winner independent of the decision.
A measurement stack for skills and demand
Skills mismatch adds another layer of difficulty because skills are not fully recorded by the highest qualification. Overskilling and underskilling are usually measured through worker questions about whether their skills are higher or lower than what their current job requires. Those questions can capture valuable information about capability and use, but they remain judgements. A question about unused ability is not identical to one about inability to complete current duties, and a question about needing more training may include aspirations for future work rather than a present performance deficit.
The European Skills and Jobs Survey shows why an overall skills measure should not be treated as a universal scale. It asked about eleven specific areas as well as overall skills. The overall mismatch measure had correlations ranging from 0.19 to 0.26 with mismatches in particular technical or transversal skills. In other words, a person’s broad sense of being mismatched is related to specific skill experiences but does not identify them completely. Literacy, digital technique, planning, communication, and problem-solving can point in different directions.
The source of skill also matters. Formal education, work experience, workplace training, and personal practice can all contribute to capability, but most datasets cannot separate their contributions cleanly. A curriculum document can describe intended learning outcomes, yet that is not direct evidence that every graduate has mastered every stated skill. An employer’s report can reveal a practical difficulty, yet it may reflect the design of a job or the support provided as much as the worker’s preparation. The measurement should name whether it observes intended learning, reported capability, demonstrated performance, or skill use.
Demand data create a related set of trade-offs. Administrative employment records and labour-force surveys cover actual employment across occupations and sectors relatively completely. They are useful for observing how employment changes, but they do not show unfilled positions and usually do not describe skills at a fine level. They therefore support an occupation-level picture more readily than a direct measure of shortages or of the capabilities used inside a job.
Online job vacancies provide a more granular and frequent view of what employers say they want. They can reveal skill terms, emerging occupations, and changing demand sooner than slower surveys. But a vacancy is still an advertised demand signal. Some jobs are more likely to be posted online than others, and vacancy data do not show how many qualified candidates are available, whether the employer’s requirements are realistic, or whether a position remains unfilled because of pay or conditions. Counting mentions of a skill is not the same as measuring a shortage of that skill.
Employer surveys can ask directly about hard-to-fill vacancies, current-worker gaps, and training needs, which makes them valuable for questions that employment records cannot answer. They also introduce an attribution problem. An employer may call a vacancy a skills shortage when the real obstacle is salary, working conditions, location, or recruitment practice. Response rates can be low, and some firms or sectors may be less likely to answer. The report therefore treats employer evidence as important but not self-validating.
Worker and employer observations can disagree even inside the same firm. The ILO review describes an Irish study in which employees reported skill gaps more often than employers did. Mutual recognition ranged from 64% for communication deficiencies to 33% for literacy and numeracy deficiencies. The disagreement does not show that one side is simply wrong. Workers may think about future responsibilities while employers focus on immediate performance; managers may not see hidden difficulty; employees may interpret a desire for development as a current gap. The observation point changes the result.
Graduate surveys and linked administrative records help connect education to later outcomes. A graduate survey can report labour-market experience by qualification, which is useful for universities considering programme changes. Online response rates may be weak because graduates have left the institution. Matching education records with labour-market administrative data can improve coverage, but outcome data still do not directly reveal every skill a person uses or every reason a job was obtained.
Higher education has a distinctive supply-side blind spot. VET learning outcomes are often connected to qualification frameworks in a common format, while higher-education programmes may describe their skills at institution level and in inconsistent language. New technologies can help identify skill content from programme descriptors, but inferred curriculum content is not the same as demonstrated competence. A system that sees detailed vacancy demand but only rough education supply can create an illusion of precision on the demand side while leaving the supply comparison unresolved.
The stack in the diagram is not a recipe for collecting every possible data source. It is a reminder that different questions require different evidence. If the question is whether hiring filters demand more education than the work uses, worker reports and job-content evaluation are relevant. If the question is whether a new technical task is difficult for current staff, specific skill evidence and employer or worker reports matter. If the question is whether a country should expand a programme, graduate outcomes, employment trends, vacancies, supply projections, and expert interpretation may all be necessary.
From a signal to a defensible claim
A defensible claim begins by naming the observation point. Is the statistic about a worker’s perception, an employer’s report, an advertised vacancy, an occupational average, a qualification outcome, or an aggregate employment pattern? The answer determines what the statistic can support. A vacancy can support a claim about advertised demand. It cannot, by itself, support a claim about a genuine shortage. A qualification can support a claim about formal education. It cannot, by itself, support a claim about demonstrated skill.
Next, name the comparison and its denominator. “Overqualified” may compare a degree with the education judged necessary to do a job, while another indicator compares it with a typical qualification in a broad occupation. “Lower wages” may compare a mismatched worker with a similarly educated matched worker, or with less-educated colleagues in the same occupation. Without the comparator, the direction and size of an effect are easy to misread.
Then make time explicit. Current employment records, a vacancy posted this month, a graduate outcome several years after completion, and a ten-year projection answer different questions. A short-term signal may be useful for immediate training while being too unstable to justify a new degree programme. A long-term projection may inform capacity planning while being too uncertain to prescribe a precise skill list.
Finally, state what evidence would change the conclusion. If an apparent shortage disappears after accounting for wages and working conditions, it was not primarily an education-supply problem. If a high overeducation rate falls when the analysis uses job content rather than occupational averages, the policy interpretation should change. If a graduate programme has weak employment outcomes but its graduates demonstrate strong transferable skills in a changing market, a single outcome measure may be insufficient. Good measurement does not eliminate uncertainty; it makes uncertainty actionable.
The most useful indicator is therefore not necessarily the most detailed one. It is the one whose observation point, comparison, timing, and limitations fit the decision. When indicators disagree, the correct response is usually not to select the most dramatic number. It is to ask which part of the education-to-work relationship each number can see—and what remains outside its field of view. Only then can the consequences of mismatch be interpreted without merging unlike problems.