Nonresponse in the analysis of the association between mental health and the urban environment in Brussels
(2021) European Colloquium on Theoretical and Quantitative Geography β Location: Manchester (online) (3.November.2021)
Files
abstract_ECTQG2021_mguyot.pdf
Extended Abstract
Open Access - Adobe PDF
- 335.15 KB
Details
- Authors
Thomas, Isabelle 
- Abstract
- 1. Introduction Understanding the relationship between urban environment and mental health is a challenging exercise (Pelgrims et al., 2021). The issue of nonresponse and incomplete data, that often confronts population studies such as health surveys, further complicates the matter (Berete et al., 2019; Boshuizen et al., 2006; Burzykowski et al., 1999; Volken, 2013). In the case of the Belgian Health Interview Survey (BHIS), three types of non-response can be highlighted: (i) initial nonresponse: refusal to participate in the survey (Van der Heyden et al., 2014), not addressed in this paper, (ii) self-administered questionnaire (SAQ) nonresponse: for those over 15 years of age, the survey consists in two parts: a face-to-face (F2F) interview and a SAQ, the latter of which is not filled out, and (iii) item nonresponse: at different moments during the survey some questions were not answered by the participants. In this paper, the determinants of non-response are analysed, taking into account socio-economic as well as urban environmental factors; the impact of these non-responses on the analysis of the association between environment and mental health are discussed. 2. Background For the BHIS, determinants of SAQ nonresponse have already been highlighted by Berete et al (2019): nonresponse is more frequent among youngsters, non-Belgians, lower educational levels and lower income, residents of Brussels and Wallonia, and people with poor perceived health. Non-response is also strongly associated with the interviewer. Other studies (Boshuizen et al., 2006; Volken, 2013) on partial non-response have additionally shown that non-response is greater for men and for unskilled workers. These studies also include urbanity indicators but the results do not converge. 3. Data and Method Data used here are from the 2008 and 2013 BHIS, and are spatially limited to the Brussels-Capital Region, because the Region was oversampled compared to the rest of Belgium and because of data consistency issues for the urban environment indicators. We limit ourselves to participants over 15 years old (minimum age to fill the SAQ) and living at place of residence for at least one year (n = 4 355). Logistic regressions with the nonresponse as dependent variable were computed: no missing data (0) vs. at least one nonresponse among six mental health indicators (1). As independent variables, socio-economic (reported household income, age, gender, family composition and highest educational level in the household) and urban environment (view of green, street canyon effect, noise, black carbon, street corridor effect, linear tree density, vegetation coverage and street visible vegetation coverage) indicators were selected (see Pelgrims et al., 2021 for more information). 4. Results In our sample of 4 355 individuals, the SAQ is not available for 36.5% of participants: for 19%, SAQ is required but not available and for 17.5%, SAQ is not required and not available (when the interview is done by a proxy such as a parent). For the available SAQ, non-response to mental health items ranged from 1% to 11%. If we analyse the subset of participants for whom the SAQ is available and who answered all mental health and socio-economic items, we have a set of 1 929 individuals (or 44% of the initial sample). Partial nonresponse is higher among low income, older person, low educational level and people with children [Figure 1]. At the exception of family composition, these socio-economic determinants are the same as those associated with poor mental health (Silva et al., 2016), so it is reasonable to assume that there will be more people with mental health problems in nonrespondents compared to the respondents. Partial nonresponse is higher in low vegetation, more polluted (black carbon) and more urbanized areas (street canyon and street corridor effect) when adjusted for socio-economic variables [Figure 2]. Figure 1. Odds ratios (and 95%CI) of nonresponse for socio-economic indicators. Figure 2. Odds ratios (and 95%CI) of nonresponse based on single-exposure models (Model 1). Model 2 is adjusted for sex, age, reported household income and year. 5. Discussion Table 3. Example of a contingency table With these results, we can extrapolate the effect of nonresponse on the association between mental health and the urban environment. Considering the association between vegetation and mental health, several studies on the topic have shown that vegetation is a protective factor for mental health (i.e. Gascon et al., 2018; Lee and Maheswaran, 2011). However, in a previous study using the same data as this paper (Pelgrims et al., 2021), no association could be found. With our results, it can be assumed that there would be proportionally a bigger pool of respondents in C [Table 3] than actually observed, i.e. πΆπ‘ππ’π>πΆπππ πππ£ππ. Indeed, the results show that the risk profiles for mental health have strong similarities with the risk profiles for nonresponse and significantly more nonresponses are found in places with lower vegetation cover. And considering the equation of an odd ratio: ππ = π΄β π·π΅β πΆ we may conclude that ππ π‘ππ’π<ππ πππ πππ£ππ. Therefore, the protective effect of green spaces on mental disorder would be underestimated, i.e. the odd ratio observed is greater than the true odd ratio. 6. Conclusion This paper deals with one of the major challenges encountered by research on the association between green space and health: the nonresponse in surveys. Because the spatial, as well as socio-economic, distribution of this bias is non-random, it is likely that it affects the research findings on the topic. We here show that the protective effect of green spaces on mental disorder may be underestimated.
- Affiliations
Citations
Guyot, M., Pelgrims, I., Aerts, R., De Clercq, E. M., Thomas, I., & Vanwambeke, S. (2021). Nonresponse in the analysis of the association between mental health and the urban environment in Brussels. European Colloquium on Theoretical and Quantitative Geography, Manchester (online). https://hdl.handle.net/2078.5/165610