Background A major aim in public health research is to prevent harm. Accurate data is needed to inform harm reduction policies, practices and research. Therefore samples used in observational studies need to represent the population from which they were drawn but response rates are declining in the UK.

Nonresponse has been addressed in various ways, including ‘boost’ samples and statistical adjustment. However, these may introduce bias and have limited applicability to smaller studies not of a national scale.

If response rates vary by sociodemographic factors then this could contribute to poor understanding of health-risk behaviours and inequality in health care provision. The aim of this review was to describe patterns of response in UK population surveys and cohort studies across a range of sociodemographic attributes to evaluate the extent to which women respondents were representative of sampled populations. While these concerns relate to both genders, the focus is upon women because they spend more years in ill-health. Their engagement in risk behaviours is also known to affect their children’s health.

Methods Six national UK health surveys (examined between 2009 and 2013) and eight cohort studies (1993 to 2009 and funded by the Medical Research Council) were included. Proportions of women respondents by age, socioeconomic status (as defined by income/occupation), education and ethnicity were recorded. These distributions were compared against population estimates for similar time-periods and geographical areas.

Results There was a pattern of underrepresentation of women aged <34 years. In four of nine studies the oldest age band was also underrepresented. Percentage differences ranged from 10% underrepresentation (16–24 year olds) to 9% overrepresentation (55–64 year olds). At the highest deprivation levels there was 4% underrepresentation to 43% overrepresentation while at the lowest levels the range was 6% under- to 21% overrepresentation. The highest underrepresentation in education was 16% for individuals with no qualifications, while the highest overrepresentation was 28% at level 4 (degree equivalent). Only four included studies reported response by ethnicity and only one had any overrepresentation of non-white groups. This same study underrepresented white women by 9%.

Conclusion This review suggests that patterns of response are varied and influenced by topic, sampling frame, recruitment and fieldwork strategies. The work was limited by variability in proxy measures for socioeconomic status and by availability of some data. Future health-risk behaviour research should focus on how to effectively reach and engage women from all social groups.