Table 3.
Weighted KHB decomposition of nested logistic regression models of material hardship types and health outcomes, add health wave IV
| Reduced (effect without mediator) | Full (effect with mediator) | Total difference | Δ (%) due to stress† | ||||
|---|---|---|---|---|---|---|---|
| Outcome: Poor health | b | OR | b | OR | b | OR (95% CI) | |
| Food hardship | 0.35∗∗∗ | 1.41 | 0.24 | 1.27 | 0.11∗∗∗ | 1.12 (0.06, 0.16) | 31% |
| Bill-paying hardship | 0.24∗∗∗ | 1.27 | 0.14 | 1.15 | 0.10∗∗∗ | 1.11 (0.06, 0.15) | 43% |
| Health-resource hardship | 0.54∗∗∗ | 1.71 | 0.48∗∗∗ | 1.62 | 0.06∗∗∗ | 2.01 (0.03, 0.10) | 12% |
| Outcome: Depression | |||||||
| Food hardship | 0.45∗∗∗ | 1.47 | 0.11 | 1.12 | 0.34∗∗∗ | 1.36 (0.22, 0.41) | 73% |
| Bill-paying hardship | 0.27∗∗∗ | 1.30 | −0.01 | 0.99 | 0.28∗∗∗ | 1.34 (0.19, 0.38) | 103% |
| Health resource-hardship | 0.42∗∗∗ | 1.52 | 0.14∗∗ | 1.15 | 0.28∗∗∗ | 1.32 (0.10, 0.27) | 42% |
| Outcome: Sleep problem | |||||||
| Food hardship | 0.28∗∗ | 1.32 | 0.15 | 1.16 | 0.13∗∗∗ | 1.14 (0.08, 0.18) | 46% |
| Bill-paying hardship | 0.26∗∗ | 1.30 | 0.15 | 1.16 | 0.11∗∗∗ | 1.13 (0.08, 0.16) | 44% |
| Health resource-hardship | 0.37∗∗∗ | 1.44 | 0.29∗∗∗ | 1.34 | 0.08∗∗∗ | 1.08 (0.04, 0.12) | 21% |
| Outcome: Suicidal thoughts | |||||||
| Food hardship | 0.79∗∗∗ | 2.20 | 0.52∗∗ | 1.68 | 0.27∗∗∗ | 1.40 (0.23, 0.45) | 40% |
| Bill-paying hardship | 0.32∗ | 1.35 | 0.01 | 0.99 | 0.31∗∗∗ | 1.36 (0.21, 0.42) | 103% |
| Health resource-hardship | 0.49∗∗∗ | 1.63 | 0.30∗ | 1.35 | 0.20∗∗∗ | 1.22 (0.10, 0.30) | 40% |
∗P < .05, ∗∗P < .01, ∗∗∗P < .001 (two tailed tests).
Control variables include age, sex, race/ethnicity, immigration status, educational attainment, family income, homeownership status, recent job loss, number of kids in the household, receipt of public assistance, as well as health behaviors including physical inactivity and smoking.
Complete tables listing all coefficients are shown in Appendix Table S1.
n = 13,313.
KHB = Karlson-Holm-Breen.
Δ(%) is the percentage reduction in the logit coefficient between the reduced and full models attributable to perceived stress, net of rescaling.