More than 80% of eligible SHS participants were willing to complete and return the additional mail-back questionnaire.
Subjects who participated in the mail survey had a higher socio-economic status (higher education, white-collar workers), reported a better subjective health and were more likely to be Swiss nationals. Generally, these finding are in accordance with previous studies that investigated initial non-response [3
]. Unfortunately, a rigorous comparison of initial and second-stage SHS/MSHS non-response is not feasible because systematic data on non-response in the SHS are unavailable. However, it is plausible that the mechanisms of initial non-response in the SHS are not very different from those found in previous HIS studies, at least when it comes to important sociodemographic and socioeconomic factors such as gender, education, employment and nationality. Moreover, available population data in comparison with SHS/MSHS data (Table ) show that the proportion of men, the proportion of subjects with primary education, the proportion of subjects with full-time employment, and the proportion of non-Swiss is lower in the SHS and MSHS samples. At least partially, these differences may be attributed to non-response. Together, evidence from previous HIS non-response studies and available population data suggest that there seems to be no fundamental difference between the mechanisms of initial non-response and second-stage non-response. Consequently, the systematic under-representation of the same group or groups of non-respondents is reinforced in the second stage. In the MSHS, this may be specifically the case for non-Swiss nationals with low educational levels. While crude MSHS response rates for Swiss nationals were 73.4% for subjects with primary education, 82.2% for subjects with secondary education, and 86.7% for subjects with tertiary education, the respective response rates for non-Swiss nationals were substantially lower (58.3%, 66.4%, and 78.3%). Furthermore, the differences between Swiss and non-Swiss nationals were much more pronounced in the lower educational levels. Unfortunately, no information is available on the initial SHS response rates by nationality and educational level. However, it is very likely that non-Swiss nationals had lower SHS response rates since interviews were conducted in German, French or Italian which potentially favoured non-Swiss nationals who are well-integrated, well-educated and have been living in Switzerland for a long period of time. Given that in 2009, non-Swiss nationals accounted for 1.7 million people or roughly 22% of the permanent resident population of Switzerland, the implications of lower response rates in this group should not be taken lightly.
Proportions of selected demographic variables in the Swiss permanent resident population and in the SHS/MSHS
Subjects without paid work and subjects who worked part-time were more likely to participate in the MSHS than those who worked full-time (all other things being equal). Although this finding is not in accordance with previous evidence [22
], one plausible explanation is that subjects who work full-time have lower time budgets and are therefore less willing to spend time on completing and returning the questionnaire.
Religious affiliation was also found to be associated with MSHS non-response. Especially subjects who were members of relatively small religious denominations were less likely to return the mail-back questionnaire while the response rates of members of the large Swiss State Churches (Protestant Church, Roman Catholic Church) and subjects with no religious affiliation did not significantly differ from each other. These finding are in partial agreement with one previous study of second-stage non-response which observed that Muslims had lower response rates than members of other religious denominations [19
MSHS response rates were not different between rural and urban areas and between different geographic regions. Previous evidence is mixed: some studies observed that response rates are higher in rural areas [13
], others found that response rates are higher in urban areas [22
]. In Switzerland, the presence of similar response rates in rural and urban areas may be related to geographic proximity. Switzerland has a very dense network of post offices. Therefore, travel time to the nearest post office to mail back the questionnaire is hardly an issue, if at all.
Finally, numerous previous studies report that females are more likely to participate in surveys [16
]. While the current study also found higher female participation, the current study did not find a uniform pattern of higher female participation in the MSHS. Rather, participation rates of men and women were linked with age, i.e. at younger ages, the participation rates of women were higher than those of men whereas participation rates of men were higher at older ages. The fact that higher male participation rates approximately coincide with retirement age may suggest that men are more willing than women to invest their additional time resources in the completion of surveys. This hypothesis should be examined in further studies.
In a second step, the current study attempted to quantify the magnitude of the bias due to selective non-response by estimating population average values of eight major MSHS outcome variables. Although overall non-participation rates were relatively low, non-participation was clearly associated with socio-economic status, non-Swiss nationality, and health status. Hence bias is potentially induced by the under-representation of non-Swiss nationals and lower socio-economic strata. Bias could be demonstrated for all crude average outcomes, except for the lifetime prevalence of osteoporosis and myocardial infarction. Furthermore, bias could also be shown for all age-standardized and sex-specific average outcomes, except for myocardial infarction. The predicted age-standardized average mastery and SOC scores of male and female MSHS non-respondents were below the respective values for respondents while the predicted lifetime prevalence of depression was higher in non-respondents. These results support the notion that bias is potentially induced by the under-representation of lower socio-economic strata: earlier studies find that lower mastery and SOC scores are associated with lower socio-economic status [49
] and higher SOC and mastery scores are associated with a better mental health status [50
]. There is also evidence that some immigrant groups in Switzerland have higher hospitalization rates due to affective disorders as compared to Swiss nationals [53
]. Similarly, the Swiss Migrant Health Survey 2010 reports better health for Swiss nationals as compared to nationals from Portugal, Serbia, Kosovo, and Turkey. More specifically, Swiss nationals were less likely to suffer from depression, high blood pressure, arthrosis, osteoporosis, sick headache, and allergy than most non-Swiss nationals [54
Bias was also observed for age-standardized and sex-specific average lifetime prevalence of high blood pressure, arthrosis, and osteoporosis. However, the predicted prevalence rates for female non-participants were above those for female MSHS participants while the predicted prevalence rates for male non-participants were below those for male participants. This may be due to a gender-specific self-selection mechanism: subjects with high socio-economic status (especially tertiary education and high and medium level non-manual workers) were much more common in male than in female non-participants.
Although bias could be demonstrated for several MSHS outcomes, the magnitude of the bias is at most moderate, i.e. the differences between age-standardized and sex-specific average prevalence rates of MSHS respondents and predicted prevalence rates of non-respondents are below 2% and the respective differences in average mastery and SOC scores cannot be considered substantial. Yet, the SHS response rate amounted to little more than 66% which suggests that SHS participants may be a selective sample of the general population of Switzerland. Moreover, the SHS sampling frame excluded subjects living in institutions and included only subjects fluent in German, French or Italian. Were these subjects part of the SHS, their non-participation may further increase bias because they may have high MSHS non-response rates and poorer health.
The current study has important limitations. Firstly, prevalence calculations for non-respondents are based on parameter estimates for respondents. This implies that the selected prevalence models for respondents also yield valid results for non-respondents. However, given the various documented differences between respondents and non-respondents, this may not necessarily be the case. That is, the use of models suited for respondents may be a potential source of bias when used to estimate the prevalence rates for non-respondents. Secondly, systematic data on initial non-response in the SHS were unavailable. Hence, the discussion on initial and MSHS second-stage non-response had to be based on previous evidence on initial HIS non-response and the available population data.