|Home | About | Journals | Submit | Contact Us | Français|
Editor—Kind et al reported the results of a survey which aimed to measure the health of a representative sample of the population of the United Kingdom by using the EuroQoL EQ-5D questionnaire.1 The authors concluded that the instrument “is a practical way of measuring the health of a population and of detecting differences in subgroups of the population” and that “the results are indicative of the average health status in the general population of the United Kingdom ... living in the community.” They do not, however, report the non-response rates of their survey.
We are informed that the respondents were representative of the general population in respect of age, sex, and social class. They may well have a similar distribution of these characteristics to the general population, but this does not make their views of their health status representative. Without knowing the non-response figures and the reasons for non-response, one can have no confidence in their conclusion.
The authors’ sample consisted of 6080 addresses, at each of which one adult aged 18 or over was sampled. Of the selected addresses, 12% were unproductive as they were non-residential, empty, or the locations were untraceable. The final sample comprised “3395 subjects.” If it was intended to sample one adult from each of the supposedly remaining 5351 addresses, then, by my calculations, 3395 represents a response rate of 63%. What happened to the other 1956, and who were they? The explanation is unlikely to be non-contact by the interviewer of at least someone in the household, given that figures from the Office for National Statistics for 1993 (the year the survey was undertaken) show that 63% of British households were occupied by two or more adults.2
These figures imply that the response rate was poor and that this was mainly because of people’s refusal to take part. A poor response rate could lead to unrepresentative findings. Maybe there is another explanation for the number of adults interviewed; if so, the authors should have provided it.
In addition, it is not possible to deduce the item non-response rates from their tables, as they only reported the numbers of people who reported problems. I assume that item response was good because this was an interview, not a postal, survey (and the responses in figure 1 can be totalled to make 3381). This information should have been provided, however, and commented on, for potential users of the instrument.
Editor—Bowling makes a helpful point in asking for information on the response rate in our national survey. Of the initial 6080 addresses identified from the postcode address file, 756 were found to be non-viable—for example, empty or derelict buildings, buildings not yet completed, business addresses, or simply untraceable locations. At the remaining 5324 viable addresses, completed interviews were achieved with 3395 respondents, yielding a 64% response rate. Reasons for non-response included a broken appointment, illness, no contact at the selected address or with the selected person, and refusal to provide any information (including a reason for this refusal).
All voluntary surveys are subject to non-response, and there can be no set definition of what constitutes a poor response rate or, indeed, a satisfactory one. There is clearly a relation between the proportion of responders who contribute to a study and the extent to which the data they generate can safely be regarded as being representative of the study population as a whole. Much depends on the magnitude of the non-response and, most importantly, on the degree to which the disposition (not) to respond is correlated with important survey variables—almost by definition an area of some uncertainty. Our sample was representative of the general population in terms of the distribution of respondents by age, sex, educational attainment, and housing tenure. Geographical distribution also closely matched local populations defined by regional health authorities and standard economic regions.
Findings like these can never provide a complete guarantee against non-response bias, but they do confirm our belief that the survey results are likely to be broadly representative of the population as a whole. Those with a residual concern about this might wish to consult table 5.24 in the health survey for England,1-1 which indicates rates of self reported health problems on the EQ-5D that are not dissimilar to those reported in our study.