It is important to establish the extent to which sampling bias explains survey variation in prevalence of child health outcomes in these surveys. We reported a significant increase in prevalence of maternal asthma with survey year from 1991 to 2006 (6.6% to 13.4%), p<0.001)  with a lower childhood prevalence of doctor diagnosed asthma in 2006 than 1998 (29.8% and 19.4%). There was an increasing proportion of hospital admissions between 1991 and 1998 (5.5% to 11.3%, p<0.001), which decreased in 2006 (9.7%). Childhood obesity prevalence was higher in 2006 (14.9%) compared to 1991 (9.2%, p=0.039, linear trend) . Over these periods the pattern of questionnaire compliance showed a linear reduction across all schools with the largest difference between the 1998 and 2006 surveys. The trends were comparable between the ten Liverpool and five Wallasey schools and it is unlikely that non-compliance was a specific problem to Liverpool schools, although these had had an early history of airborne dust exposure from the Liverpool docks. It is likely that the very high survey compliance in 1991 and 1993 was influenced by community concerns about childhood health and dust pollution. The much lower response rate in 2006 is unlikely to be related entirely to amelioration of these environmental exposures which occurred in the interim period.
It is unlikely that questionnaire and sampling variation explained changes in compliance between the four surveys as essentially the same survey instrument, methodology and schools were used for these sequential assessments. The surveys were comparable in terms of questionnaire design and the same core questions were used which comprised the majority of listed questions.
Response rates to general population surveys have been in decline over recent decades, with compliance ranging from 30% to 70% [15, 20–27] and the much lower compliance in the 2006 Merseyside survey was consistent with this pattern. This was a major concern, as marked prevalence changes in the survey population were concurrently observed for both childhood and parental asthma , and for childhood obesity . The response rates in these school surveys are critical for assessing the validity of the study findings. Response representativeness is more important than the actual response rate in this type of survey . A low response rate of 30% has been proposed as acceptable for patient satisfaction surveys, providing the sample is representative . Supplemental analysis is helpful to confirm that respondents are in fact representative of the population . Ideally analysis of representativeness in relation to non-response bias should compare data from responders who participated in the survey, with non responders who didnot participate in order to measure the potential bias resulting from low response rates . In the present survey this was not possible, and an alternative approach of comparing socio-economic characteristics of responders and non-responders to specific questions was adopted.
Without information on the household characteristics of parents who did not return the questionnaires, it is not possible to determine if these comprised households with different socio-economic profiles. Non-responders have often been shown to differ from responders in terms of a number of socio-demographic and economic variables which are linked to lifestyles, attitudes and beliefs [24, 31]. The pattern of non-response bias is difficult to assess because characteristics of those who are contacted, but refuse to participate, will not be available . There were no mean differences in socio-economic deprivation indices between respondents across the four surveys. This population was predominantly Caucasian (>94%)  throughout these survey years, and there were no major shifts in ethnic composition, indicating there were no substantial ethnic demographic changes.
The pattern of parental non-responses to specific questions varied across the four surveys indicating some influence of survey year on response characteristics. Non-responses to some core questions were higher in 2006, when survey compliance was also lowest, indicating a greater reluctance to provide answers to specific questions in the 2006 survey. This was not the case for the response to the question on dampness in the home, for which the response rate had improved in 2006, concurrent with social improvements in housing in the later survey years. A significant change in the 2006 survey was the higher proportion of employed mothers. As mothers were mostly responsible for completing questionnaires, a busier working lifestyle might explain increased reluctance to respond to time consuming school-questionnaires. The estimated time for completing the school questionnaire was approximately 20 minutes. The proportion of single parent households, and the mean Townsend scores, did not vary between surveys. Similarly, duration of residence at the present address for more than three years did not significantly differ across surveys. This suggests that despite a reduction in compliance in 2006, there were no major changes in household socio-demographic profiles between surveys for parents who returned questionnaires. This is re-assuring in terms of response bias related to social deprivation. Examining this in further detail in relation to mean Townsend scores for responders or non-responders to specific questions, showed scores were slightly lower in 2006 than 1991, but this occurred for both these groups. There was no pattern towards lower scores for households with non-responses to specific core questions.