Two courses can report the same median graduate earnings while having very different distributions below that midpoint. A median is useful, but it cannot describe every result in the lower half. If a student is concerned about meeting a future financial commitment, the spread of outcomes can matter alongside the typical figure.

The ONS earnings interpretation guide explains the usefulness of the median and discusses variation in earnings distributions. That methodological context supports looking beyond a single average. It does not supply the course data below: both sets are invented, and the exercise makes no claim about actual graduate salaries or a student's personal earning probability.

Hold the median constant

Suppose each fictional dataset contains nine annual earnings observations, already sorted and measured in thousands of one currency. All nine are assumed to meet the same imaginary reporting definition and reference period. Dataset A is 28, 29, 30, 31, 32, 33, 34, 35, 36. Dataset B is 12, 18, 24, 30, 32, 40, 50, 65, 90.

Measure Dataset A Dataset B
Number of observations 9 9
Median, the fifth observation 32 32
Observations strictly below 26 0 3
Proportion strictly below 26 0 out of 9 3 out of 9
Minimum to maximum 28 to 36 12 to 90

The medians match exactly. Yet three B observations, one third of that small fictional set, fall below 26, while none of A's do. A median-only comparison would conceal this difference. B also contains larger upper outcomes, so the exercise is not a complete ranking of one distribution over the other.

The threshold of 26 is invented for the demonstration. A real planning threshold needs consistent units and purpose. Gross annual earnings cannot be compared directly with a monthly after-tax repayment commitment. Nor should a threshold in one country's currency be applied to earnings in another without an explicit conversion and reference date.

Make the threshold rule reproducible

State whether the rule is strictly below, at or below, or within a band. With a threshold of 30, for example, A has two observations strictly below and B has three. Counting at or below instead gives three and four. A one-word definition changes which records qualify.

Do not invent the missing distribution from a published median. If only the midpoint is supplied, the number below 26 is unknown. A median of 32 alone cannot establish the lower tail's shape. Ask whether the producer publishes relevant percentiles, bands or other distribution information, together with definitions and disclosure limitations.

Even reported percentiles may not identify the exact proportion below an arbitrary threshold. Keep bounds or uncertainty visible rather than interpolating a precise figure without justification. Small samples, rounding and suppression can further limit what a table supports.

Check who is absent from the earnings distribution

Establish whether the publication covers employed respondents, all graduates with recorded earnings, full-time employees or another group. A lower-tail calculation among included earners does not automatically include graduates with no recorded earnings. “Not observed” must not silently become a zero salary, and a zero must not silently disappear.

The graduate salary comparison guide helps align populations and timing before comparing distributions. Avoid treating these descriptive results as the causal effect of attending a course: entrants, occupations, locations and working hours can differ.

For a practical note, record the median, the available lower-distribution evidence, the population covered and the threshold's definition. Add “not available” where the evidence stops. The result is a clearer assessment of what a salary headline can support, rather than an unsupported prediction that a student will earn exactly the reported midpoint or avoid every lower outcome.