Recognise Self-Selection Bias in a Thesis Online Survey

Ask who could and would respond

Access and motivation jointly shape an open online survey. A link in a student support forum reaches different people from a university register. Those with strong experiences may be more likely to click. The resulting distribution can be shifted even when the questionnaire and statistical calculations are technically correct.

Example: You post a survey on exam anxiety in a stress-support community. Many affected students respond. Their answers do not estimate anxiety prevalence among all students at the university. The data can still inform a narrower question about the people reached if the selection process is explained.

Describe recruitment in enough detail

Record channels, dates, invitation wording, eligibility and incentives. Could the link be forwarded freely, or did only a defined group receive it? Could one person submit more than once? Without a known number of invitees, a conventional response rate often cannot be calculated reliably.

Set exclusions before analysing results. Look for overrepresentation by field, study year or age. These differences do not reveal every bias, but show which target-group segments are scarcely represented. Mention accessibility barriers, mobile usability and survey language when relevant.

Use comparison data cautiously

If credible population data exist, compare a few important characteristics such as field and year of study. Large differences make straightforward generalisation doubtful. Weighting may help under specific assumptions, but it does not create a random sample or fix unknown differences in motivation or experience.

Phrase findings as statements about respondents. Avoid writing “students believe” when a self-selected subgroup answered. Give the analysed sample size for each result; screening questions and missing answers may produce different denominators across tables.

Make the limitation part of interpretation

Self-selection is especially serious for prevalence estimates. A statistically significant association does not remove recruitment bias. Consider whether willingness to participate could be related to both the explanatory variable and outcome, potentially altering the association itself. Discuss the actual recruitment route rather than adding a generic limitation sentence.

Future work might use several channels, directed invitations or a defined sampling frame. For this thesis, the important question is which claims the collected responses support. Transparent limits strengthen a small student study because they prevent the results from saying more than the evidence permits.

Frequently asked questions