An employment percentage and a survey response percentage have different denominators. Reading one without the other can make a result sound more complete than the evidence allows. Start by distinguishing the people eligible for the survey, the people who replied, and the people included in a particular outcome calculation.

The Office for Students' data FAQs define a survey response rate using respondents divided by people eligible to be surveyed. Its Graduate Outcomes guide explains that the survey asks about activities after completing higher education. Neither point makes every published outcome percentage a percentage of all graduates. Check the actual measure's definitions and treatment of partial responses.

Separate three counts before interpreting a rate

Write down the eligible population, the number of usable responses for the question, and the number reporting the specified outcome. A person may respond to some questions while leaving another unanswered, so a survey-wide participation count may not be the correct denominator for a salary or employment item.

Also retain the outcome definition. Employment, further study, a combined positive-outcome category and a particular occupational classification are different measures. This article uses a deliberately simple binary employment outcome to illustrate missing information; it does not calculate or redefine an official regulatory progression indicator.

Work through an invented population

Imagine a fictional college with exactly 100 eligible graduates. Sixty provide a complete answer to the employment question; 48 of those 60 report the defined employment outcome. Assume each person appears once, all 100 are in scope, all recorded answers are accurate and the remaining 40 outcomes are unknown.

Quantity in the fictional exercise Calculation Result
Response rate for this question 60 / 100 60%
Employment rate among respondents 48 / 60 80%
Known employed as a share of all eligible graduates 48 / 100 48%
Unknown outcomes as a share of all eligible graduates 40 / 100 40%

The 48% figure is not an estimated population employment rate. It counts only the known employed people against the whole eligible population. Calling every unknown person unemployed would turn an information gap into a claim about their activity.

Under the exercise's assumptions, the total employed population could be as low as 48 if none of the 40 nonrespondents had the outcome. It could be as high as 88 if all 40 did. The corresponding bounds are 48% to 88%. These are logical limits under the stated assumptions, not a confidence interval or a claim that the endpoints are equally likely.

Show what an extra assumption would do

Let q be the unknown fraction of nonrespondents who had the outcome. The whole-population proportion would be 0.48 plus 0.40 multiplied by q. If q were 0.50, the result would be 68%. If q were 0.80, it would be 80%.

The second scenario reproduces the respondent percentage because it assumes the missing group has the same employment rate as the observed group. The arithmetic does not supply evidence for that assumption. Write it as a scenario instead of silently treating 80% as the measured population result.

Nor does a lower response rate prove that a published result is biased in a particular direction. The missing group's outcomes determine that direction in the example. A participation percentage alone does not reveal them. Two surveys with similar participation can therefore have different coverage problems, and a high participation rate is not a complete quality assessment.

Read the methodology before comparing providers

Real releases may use weights, combine activities, restrict populations or apply publication thresholds. A weighted estimate cannot be reproduced by substituting a raw respondent count into a different formula. Use the source's own published methodology and keep its rounding, missing-value and suppression rules intact.

Ask whether the eligible cohort, question coverage, response definition and collection period match between the results you are comparing. Look for separately published response information for relevant groups. Do not infer a particular nonrespondent's activity or try to reconstruct suppressed small cells. The missing-data guide provides a related method for preserving unknown states.

Record both the outcome result and its coverage evidence in your shortlist. A careful conclusion states whose answers supplied the figure, what the source did with missing responses and what remains unknown. That makes a survey useful without presenting the observed respondents as automatically identical to every graduate.