Regression analysis
The results of the multiple regressions are shown in Table
2 (first column: total sample; second column: men; third column: women). R
2 values were .11, .13, and .09, respectively. After adjusting for various potential confounders (regression coefficients not shown here, but available upon request), linear regressions showed that while social exclusion was not associated with obesity in the total sample and in men, women with obesity were
less socially excluded than women without obesity (β = −.06,
p = .02). In addition, it was tested whether sex moderates the association between social exclusion and BMI. Actually, the corresponding interaction term (sex x obesity) was significant (
p = .03).
Table 2
Determinants of social exclusion (German Ageing Survey, 2014)
Potential confounders | ✓ | ✓ | ✓ |
Obesity (BMI ≥ 30 kg/m2; Ref.: Non-obese) | −0.03 (0.02) | 0.01 (0.02) | −0.06* (0.03) |
Constant | 2.50*** (0.06) | 2.39*** (0.08) | 2.642*** (0.09) |
Observations | 7041 | 3532 | 3509 |
R2 | 0.11 | 0.13 | 0.09 |
In sensitivity analysis (not shown here, but available upon request), survey weights were used in order to ensure the representativeness of the data. Results remained almost the same. For example, women with obesity were less socially excluded than women without obesity (β = −.07, p = .03). The interaction term remained significant (p = .04).
We also tested whether our findings were dependent on the statistical approach chosen. Therefore, linear regressions were replaced by ordered probit regressions. In terms of significance, our results remained virtually the same.
In further sensitivity analysis, it was tested whether the association between obesity and social exclusion varied by employment status, marital status, depression or age. However, none of the interaction terms achieved statistical significance.
In another sensitivity analysis, the definition of obesity was modified to only include individuals with BMI ≥ 30 kg/m2 and BMI < 40 kg/m2 in order to exclude Obese Class III individuals, which may strongly affect results. In terms of effect sizes and significance, findings remained virtually the same. In further sensitivity analysis, the definition of non-obesity was restricted to individuals with BMI > 18.5 kg/m2 and BMI < 30 kg/m2. Again, findings remained almost the same. Thus, findings were not driven by underweight individuals.
In another sensitivity analysis, BMI was classified as: (0) BMI < 30 kg/m2, (1) 30 kg/m2 ≤ BMI < 40 kg/m2 and BMI < 40 kg/m2 as well as (2) BMI ≥ 40 kg/m2 (63 men; 93 women). When compared with our main models, findings remained almost the same. Women with obesity (Obese class I and II) were less socially excluded than women without obesity (β = −.05, p = .04). Women with obesity (Obese class III) were not less socially excluded than women without obesity (β = −.13, p = .09).
We also tested whether differences exist between individuals with obesity and (i) individuals with normal weight as well as (ii) overweight individuals. Again, women with obesity were less socially excluded than women with normal weight (β = −.07, p = .02). However, women with obesity were not less socially excluded than women with overweight (β = −.05, p = .06). Apart from that, other significant differences between the groups were not observed.
Stratified by age (younger than 65 years; aged 65 years and above), women with obesity younger than 65 years were less socially excluded than women without obesity younger than 65 years (β = −.08, p = .01), whereas women with obesity aged 65 years and above were not less socially excluded than women without obesity aged 65 years and above (β = −.04, p = .35).