Graduate Employment and Salary Data in Rankings: What Is Measured and What Is Imputed

How graduate employment and salary indicators are built: ranking weights, survey windows, response rates and missing-data substitutions.

Employment and salary figures in rankings can look more direct than their production process. A published percentage may combine employer perceptions, answers supplied by graduates, or events recorded by public bodies. It may also rest on a different survey population, observation period and statistical rule from a figure published elsewhere. The displayed number is therefore an endpoint of several methodological choices, not simply a direct reading of graduate success.

Who Measures Employment and Salary, and with What Weight?

QS makes employment-related evidence part of its overall ranking structure. Its Employability and Outcomes lens accounts for 20% of the World University Rankings, divided into Employer Reputation at 15% and Employment Outcomes at 5%.

These components measure different objects. Employer Reputation measures the reputation of institutions and their programmes among employers. QS collects those views through its Employer Survey. The 15% is a ranking weight attached to that employer-perception indicator; it is not the percentage of a university’s graduates in employment.

Employment Outcomes is a separate indicator within the lens. QS defines it as measuring the extent to which institutions can ensure high employability for graduates and produce graduates who go on to make a meaningful impact on society. Its 5% weight must therefore not be conflated with the 15% assigned to employer reputation.

The Times Higher Education World University Rankings 2026 follows a different structure. None of its 18 indicators measures employment or salary. Its published weighting includes Teaching at 29.5%, Research environment at 29% and Research quality at 30%, together with International outlook and Industry. The survey used in the main-ranking framework is the Academic Reputation Survey, conducted from November 2024 to January 2025 and drawing more than 108,000 responses; it is not an employer survey.

Consequently, QS contains an employment-related component in its weighted score, while THE’s 2026 main ranking has no corresponding employment or salary component. THE’s omission does not establish that a university’s graduates had poor employment outcomes. It means that the main weighted score does not measure that dimension.

Employer Surveys, Graduate Tracking and Administrative Records

Three distinct data-production mechanisms appear in graduate employment measurement.

An employer reputation survey records perceptions rather than the circumstances of identified graduates. In QS, employers provide views of institutions and programmes. The resulting indicator is a measurement of reputation: it is evidence of how employers assess an institution, not a direct count of that institution’s employed graduates.

A graduate tracking survey asks graduates to report their own circumstances after completing a course. Depending on the survey, respondents may classify themselves as working, unemployed or in further study, and may also supply salary information. Graduate Outcomes in the United Kingdom, QILT in Australia and the Higher Education Authority in Ireland use surveys of this kind. Their figures are self-reported observations, linked to particular graduate populations and collected at specified points after completion.

Administrative records provide a third route. They use information held by public bodies rather than answers to a graduate questionnaire. Ireland’s Central Statistics Office uses a longitudinal approach that brings together salary data from Revenue, benefits data from the Department of Social Protection and graduate data from the HEA. This records events visible in those administrative systems and permits a different time horizon from the HEA’s nine-month survey.

These mechanisms are not interchangeable. Employer surveys measure perceptions, graduate surveys measure reported individual status, and administrative systems measure recorded events. None is simply a direct census of every graduate’s labour-market position unless its methodology establishes that coverage.

Why the Tracking Window Matters

The elapsed time between graduation and measurement is fundamental because each system observes a different point in graduates’ trajectories.

Graduate Outcomes in the United Kingdom asks graduates to respond 15 months after they finish their course. It describes this as picking up with graduates over a year after completion. Australia’s QILT Graduate Outcomes Survey is completed approximately four to six months after students finish their studies. Ireland’s HEA has surveyed graduates nine months after graduation annually since 1982.

These are not three labels for the same moment. A QILT result at approximately four to six months and a UK result at 15 months describe graduates at different stages after completion. Employment, further-study and job-search statuses are not fixed properties of a graduating class: an individual’s recorded activity at the earlier survey point is not necessarily the activity reported at the later one. No common observation point is created merely by expressing all three as post-graduation outcomes.

A national annual survey also does not necessarily observe one graduating cohort at one uniform age or elapsed time. QILT conducts its survey annually across rounds in November, February and May, while Graduate Outcomes centres its published question on the point 15 months after course completion. The timing built into the survey design is therefore part of the indicator’s meaning.

Response Rates and Non-Response

The scale and coverage of a survey must be established before its percentages become ranking inputs. Ireland provides an explicit response-rate anchor. The HEA reports that just under 71,000 graduates in the Class of 2024 were surveyed, producing a 50.9% response rate. The results were published in November 2025.

That rate measures survey participation, but it does not by itself establish that respondents have the same employment circumstances as nonrespondents. Nor does it permit a nonresponse to be classified as unemployment, employment or further study. A missing survey response is an absent observation, not an observed labour-market status. Any conversion of nonresponses into estimated outcomes would require a separate, stated method.

The United Kingdom source provides a different form of scale measure: 353,755 responses for the Class of 2023/24. This is a response count, not a response rate, because the cited page does not provide the corresponding number of graduates approached. The two published quantities should not be compared as though both expressed the same form of coverage. One is a rate against a survey population; the other is a raw volume of responses.

For institution-level outcomes, a survey-based percentage rests on the responses associated with that institution. Response volume, response rate and the composition of the respondent pool all enter the production process even when the ranking table displays only the resulting percentage.

What “Salary” Means Statistically

A salary figure is not meaningful without its population, work threshold and statistical summary.

QILT defines full-time employment as working 35 hours or more per week across all jobs combined. The hours are therefore added across jobs rather than requiring one job by itself to meet the threshold. QILT separately defines overall employment as one or more hour of work per week, so the full-time category used for its salary measure has a narrower definition.

Its median salary is the median salary level of graduates who are in full-time employment. It is therefore a median rather than an arithmetic mean, and its population excludes graduates outside the stated full-time-employment category. The median cannot be treated as a salary average for all surveyed graduates, including those who are working part-time, unemployed or engaged in further study. QILT’s results also draw on responses from domestic undergraduate and postgraduate coursework graduates; research-master graduates are outside that stated pool.

The reported “salary” should not be assigned additional attributes that are not present in the definition. The cited QILT rule establishes the work threshold and included population, but the word “salary” alone does not establish a pay period or whether remuneration is presented before or after deductions.

Ireland’s HEA survey uses different status categories at its nine-month observation point: 80.2% of graduates were working or due to start work, while a further 10.7% were engaged in further study. “Due to start work” is not identical to currently being in employment, and the Irish survey is not a salary measure like QILT’s full-time-employed median.

Graduate Outcomes reports that 57% of UK respondents were in full-time employment, 5% in full-time further study and 7% unemployed. Its full-time categories cannot automatically be assigned QILT’s 35-hour threshold because the cited UK material does not state an equivalent hours definition. The Irish, Australian and UK percentages are consequently measurements with different populations, categories and elapsed-time rules.

Imputation, Substitution and Uncertainty

An imputed value is not itself an observed outcome. The cited graduate-employment sources, however, use several different techniques that should not all be called imputation.

Ireland’s Central Statistics Office substitutes an administrative route for some questions about longer-term outcomes. It collates Revenue salary data, Department of Social Protection benefits data and HEA graduate data to form a longitudinal view extending up to 10 years after graduation. This is a different data source and time horizon from the HEA survey completed nine months after graduation. The HEA page presents the administrative approach as an alternative form of longitudinal measurement; it does not state that individual nonresponses to the Class 2024 survey are filled with administrative salary values.

ARWU provides a direct, explicit example of a non-observed ranking value being substituted. In its per capita performance calculation, weighted scores are divided by full-time-equivalent academic staff. If the academic-staff number for institutions in a country cannot be obtained, ARWU uses the average number of academic staff for the world’s top 1,000 universities. This is not a graduate-employment indicator, but it shows how a missing denominator can be replaced by a proxy in published ranking arithmetic.

QILT uses pooling and confidence intervals rather than a stated salary-imputation formula. Survey results from the three most recent collection years are pooled to improve reliability, and 90% confidence intervals are displayed to indicate accuracy. Pooling combines observed results across years; it does not assign an invented salary to each nonrespondent. The resulting estimate also represents a pooled period rather than only one collection year, while the confidence interval makes the precision of that survey estimate visible.

These mechanisms should remain analytically separate. Administrative linkage changes the source and duration of observation. Pooling changes the period over which survey evidence is combined. A confidence interval represents uncertainty around an estimate. ARWU’s missing-data rule substitutes a value for an unavailable denominator. None of the cited graduate surveys publishes a general rule that imputes an individual salary for every graduate who fails to respond.

Why One University Can Produce Different Numbers

A ranking label such as “employment” or “salary” does not guarantee a common measurement. QS and THE already differ at the ranking level because the former includes an Employability and Outcomes lens and the latter’s 2026 main ranking does not. Across graduate-outcome systems, the source can be a graduate survey or administrative record; the observation point can be approximately four to six months, nine months or 15 months; and the counted categories can include full-time work, any work, further study, unemployment or a future start date.

Survey design adds further differences. Results may come from different respondent pools or collection periods. QILT pools three years and reports confidence intervals, while instrument revisions can interrupt a series. The HEA reviewed its survey in 2024 and states that a revised instrument will be used in 2026.

QILT also supplies a direct example of a definition change producing a movement in the statistic. It reports that undergraduate full-time employment moved from 74.0% to 75.4% and attributes the movement to the change in how the labour force was defined in 2025. No improvement in the graduates’ real-world circumstances is required to produce that statistical movement: changing the classification rule changes the reported rate.

The central distinction is therefore between observation and construction. Employer surveys observe perceptions; tracking surveys observe graduate-reported statuses; administrative systems observe recorded events. Pooling, confidence intervals, alternative administrative routes and proxy denominators then serve different purposes. A final ranking percentage may be precise in presentation while resting on nonresponse, a different time window, a revised definition or an explicitly substituted value.