Among the strengths of this study are the similarities between the survey designs, the administration of the same validated structured clinical interview, and the representativeness of the APMS 2007 and the SELCoH samples. Specifically, the SELCoH sample was shown to be representative on most demographic and socioeconomic characteristics of the population in the study catchment area, the London boroughs of Southwark and Lambeth, according to the UK 2001 Census [16]. In terms of study limitations, non-response rates at the individual level in the national study and at the household level in the SELCoH sample may have resulted in participation bias, and the prevalence estimates should be considered with caution. A study linking a Norwegian household survey with disability pension registry data illustrated that mental and substance use disorders were strongly associated with non-participation, and conventional surveys may under-estimate prevalence of mental disorder [31]. Further, different weighting procedures were utilised and it is likely that there is residual confounding in the analysis examining differences between the two samples. In terms of the timing of the studies, APMS data were collected in the year leading up to the UK economic recession (2006–2007), while SELCoH data were collected after the recession period began (2008–2010). Common mental disorders, increased substance use and substance disorder are among the mental health problems more likely to increase during an economic downturn, primarily as a result of an increase in the poor socioeconomic conditions included in this analysis [32]–[34]. Finally, we acknowledge concerns about the validity of measures, such as the CIS-R [35]. However, as with the national study, we enlisted experienced and trained lay interviewers to administer the CIS-R. Given these strengths and limitations, it is worthwhile considering the extent to which differences in prevalence estimates may be real, or have resulted from the design and conduct of the two studies. Low participation rates may be a factor in any psychiatric epidemiological survey, and the lower participation rate in the SELCoH sample compared with the APMS sample may have led to an under-estimate of differences in prevalence estimates [31]. We used appropriate statistical methods to apply weights for each sample and to control for the effect of clustering by household within the SELCoH study in all analyses, so believe it unlikely that this contributed to the observed differences.

In theory these differences might have been a valid consequence of differences between the survey populations with respect to demographic profiles. The SELCoH sample was a somewhat younger and more female sample, with a much higher proportion of people with Black African or Caribbean ethnicity, and included more unemployed people and more manual workers, at the same time as having a better qualified population. Despite these differences in the demographic profile, there was a similar pattern of associations for the socio-demographic and socioeconomic indicators and the outcomes across the SELCoH and APMS samples. We argue that such differences in populations should not be seen as nuisance variables (i.e. confounders) but should be seen as plausible explanations for the differences observed. However, adjusting for these variables in analyses of combined datasets had no impact on the differences between the samples in frequency of disorders at the national and more local level. Thus, the differences are likely to be due to potential confounders not identified in this study. Further, it is notable that within virtually each stratum of the analyses presented, the prevalence of CMD and illicit drug use was higher and the prevalence of hazardous alcohol use was lower in the SELCoH sample than in the APMS sample.