The word “average” can conceal an important choice. A salary table may report the arithmetic mean, while another shows the median. Comparing those figures as though they were the same measure can create a false impression about graduate outcomes, even when both publishers have calculated their statistics correctly.

NIST's statistical handbook defines the mean and median as different measures of location. The mean uses every value in the sum; the median uses the middle of the ordered observations. The following invented list isolates their behaviour. It is not a sample from a university, an estimate of labour-market pay or a recommendation to ignore high earners.

Change one value, retain the others

Suppose five fictional full-time salaries, measured in the same units and period, are 28,000, 30,000, 32,000, 34,000 and 36,000. Their total is 160,000. Dividing by five gives a mean of 32,000; the middle ordered observation is also 32,000.

List Total Mean Median
Original five salaries 160,000 32,000 32,000
Replace 36,000 with 136,000 260,000 52,000 32,000

One observation rises by 100,000, so the mean rises by 100,000 divided by five, or 20,000. The median remains 32,000 because the middle observation does not change. A headline saying that “average salary rose by 20,000” would be incomplete unless it identified the mean and explained the underlying population.

The median's stability is useful when extreme observations dominate a mean, but stability does not make it a full description. In the changed list, the median conceals the increase in total pay. The mean records that increase but no longer resembles the salary of most individuals in this small group. The right choice depends on the question.

Ask about spread as well as the centre

Two populations can have the same median and very different lower and upper tails. A student concerned about financial downside needs more than a central value. Where available, inspect published quartiles, the distribution and the proportion with salaries measured. Never infer a minimum salary from the median.

Equally, do not treat the sample's highest observation as a likely personal outcome. An extreme value may be genuine, miscoded or measured on another basis. Check the publisher's treatment of outliers and verification rules before deciding that it should be removed. Deleting inconvenient observations simply to improve a comparison is not a defensible cleaning method.

Definitions must remain aligned. Annual full-time pay, hourly pay and total compensation are different variables. A change in the proportion working part time can alter an overall earnings summary without any change in pay rates within comparable jobs. Currency conversions and observation dates can create further differences that the choice of mean or median does not solve.

Read the table label before using the figure

For each source, record the named statistic, population, salary period, response coverage and any exclusions. If “average” is undefined, mark it as unresolved and consult the methodology. Do not silently relabel it as median because another institution uses that measure.

Reproduce a small example like this one in the comparison sheet before combining real figures. It tests whether everyone interpreting the sheet understands what can move the statistic. If the decision changes when one plausible extreme observation changes, the evidence may need a distributional view rather than a single-number ranking.

The resulting conclusion should distinguish summary and promise: the invented mean is 52,000, the median is 32,000, and neither is a guaranteed starting salary. Reporting both with their definitions gives readers more information than choosing whichever figure makes a programme appear strongest.