There have been many articles over recent years that have included questionnaires in the study methods, most recently the National Biometry Audit II.1 Much has been published both in text books and in the peer-reviewed literature concerning best practice in questionnaire design and utilisation,2, 3, 4, 5 and while questionnaires are clearly an invaluable research tool when used appropriately, poorly constructed questionnaires produce meaningless or misleading responses, and lack of rigour in analysis leads to the formation of invalid conclusions.6

The National Biometry Audit II utilised a telephone-administered questionnaire, recruiting 94 biometrists from 178 potential interviewees (53%);1 there would be no reason to assume that the 47% who did not respond were similar to those who did. The risk of systematic bias being introduced is clearly a problem that should have been discussed in the article; nonresponders might well have been found to be less rigorous in their biometry practice than responders, hence the true picture of biometry practice in the UK may be less healthy than that reported.

Maximising response rate is one way in which the effect of systematic response bias can be reduced, and some journals have employed response thresholds below which they will automatically reject questionnaire-based studies.7 There are well-established guidelines on how to maximise response rate in surveys, despite which many studies are published that have failed to take advantage of these and often suffer as a result.8

A hand search of all articles published in Eye over the past 10 years found 26 studies involving questionnaires. We compared the published methodologies of these studies to the points of best practice identified from a literature review.2, 3, 4, 5 Although some studies were extremely rigorous, pre-testing of questionnaires, maximising response rates, and taking into account the possible bias introduced by under-attainment in the analysis or even making active attempts to characterise non-responders, there were more studies that did not.

We would encourage all authors undertaking questionnaire-based research to pay close heed to their study design. Where the number of potential interviewees is large, random sampling might be a better way to reduce the overall numbers rather than self-exclusion by nonresponse.