If an outcome rate is 60% in two successive observations, it is tempting to say nothing changed. That statement is true of the aggregate percentage. It need not be true of the people behind it. Some students may have entered the outcome while others left, with the two movements cancelling in the total.

The ONS describes linked longitudinal data as a way to investigate the same people across observations. That methodological distinction motivates this fictional example. No actual ONS student outcomes are used, and the illustration is not an evaluation of a university or evidence of the effect of its teaching.

Track the same people in both periods

Suppose a fictional study follows exactly 100 students at two dates, with complete and consistently defined records. Sixty meet an outcome at the first date. Forty do not. By the second date, twenty of the original sixty no longer meet it, while twenty of the original forty now do.

First-date status Second date: meets outcome Second date: does not Row total
Meets outcome 40 20 60
Does not meet outcome 20 20 40
Column total 60 40 100

The outcome rate is 60% on both dates. Forty students changed category: twenty in each direction. Sixty did not change category. The net change of zero students is therefore compatible with substantial movement. It would also be compatible with a completely different table in which nobody changed.

Conditional transitions add detail. Among the sixty initially meeting the outcome, twenty left it: one third. Among the forty initially not meeting it, twenty entered it: one half. These denominators differ, so the two conditional percentages should not be added or compared as if they were shares of the same starting group.

Separate net change from gross movement

Net change is entries minus exits. In this example it is twenty minus twenty, or zero. Gross movement between the two categories is entries plus exits, or forty. Both calculations are useful, but they answer different questions.

The observation dates still limit interpretation. A student counted in the same category twice might have changed status between the dates and returned. The table captures endpoint transitions, not every intervening spell. More frequent observations or event histories would be needed to describe durations and repeated changes.

An outcome also needs a precise definition. “Employed” could include different hours, contract types or simultaneous study. A binary transition does not reveal whether conditions improved within employment. Do not let a convenient two-category table stand in for evidence about job quality or personal wellbeing.

Check whether the observations are actually linked

Two annual university summaries may describe different graduating classes. Even if each contains 100 students and reports 60%, they do not provide individual transitions. You cannot populate the off-diagonal cells above from repeated cross-sectional totals alone.

If a panel loses respondents, identify who remains in the matched sample and whether replacements have been added. A complete-case table may describe a selected subset. Record response loss and the producer's handling of missing records; do not treat someone absent at the second date as automatically moving into the negative category.

The data year alignment guide helps check reference periods, while this method checks whether the unit itself is followed over time. For a research note, state the aggregate rates, the matched sample size, the transition counts when available, and what is not observed between dates. “The percentage is unchanged” can then remain a precise descriptive statement without becoming an unsupported claim that every student's situation stayed the same.