Skip to main content

Table 2 Decomposition of the total time effect on good / poor SRH into direct and indirect effects via leisure time physical activity (LTPA) – using KHB-method for GEE logistic regression

From: Trends in self-rated health among the elderly population in Germany from 1995 to 2015 – the influence of temporal change in leisure time physical activity

 Good SRH
 MenWomen
 Model 1Model 2Model 1Model 2
 OR95% CIOR95% CIOR95% CIOR95% CI
Total time effect1.38***1.25–1.521.13*1.04–1.291.77***1.63–1.981.34***1.20–1.41
Direct time effect1.14*1.03–1.251.000.93–1.151.38***1.26–1.521.15**1.03–1.28
Indirect time effect1.21***1.19–1.241.13***1.11–1.151.28***1.24–1.311.17***1.14–1.19
 APE
(%-points)
95% CIAPE
(%-points)
95% CIAPE
(%-points)
95% CIAPE
(%-points)
95% CI
Total time effect7.35.1–9.42.70.9–5.712.410.4–14.46.84.5–9.1
Direct time effect2.90.7–5.10.0−1.6-3.17.15.0–9.23.51.2–5.9
Indirect time effect4.4---a2.7--- a5.3--- a3.3--- a
Conf_ratio2.5--- b1.72.1
Conf_pct60.1100.042.853.4
 Poor SRH
 MenWomen
 Model 1Model 2Model 1Model 2
 OR95% CIOR95% CIOR95% CIOR95% CI
Total time effect0.75***0.67–0.841.040.91–1.180.72***0.65–0.800.990.88–1.12
Direct time effect0.960.86–1.081.21**1.07–1.380.920.83–1.031.18**1.04–1.33
Indirect time effect0.79***0.77–0.810.86***0.84–0.880.78***0.76–0.800.85***0.83–0.87
 APE
(%-points)
95% CIAPE
(%-points)
95% CIAPE
(%-points)
95% CIAPE
(%-points)
95% CI
Total time effect−4.7−6.6-2.80.6−1.4-2.6−6.0−8.2-0.4−0.1−0.22-0.20
Direct time effect−0.7−2.6-1.23.01.0–5.0−1.4−3.5-0.52.80.70–4.94
Indirect time effect−4.0--- a−2.4--- a−4.6--- a−2.9--- a
Conf_ratio6.9--- c4.2--- c
Conf_pct85.5--- c76.3--- c
  1. OR Odds ratio, APE Average partial effects (change in average probability of good / poor health over time in percentage points), model 1 adjusted for age, model 2 additionally adjusted for school education, income and occupational position. Conf_ratio: Confounding ratio, gives information on the total effect size relative to the direct effect size, calculated by: total effect / direct effect. Conf_pct: Confounding percentage measures the percentage change of effect attributable to confounding net of rescaling, calculated by: indirect effect / total effect. *p < 0.05, **p < 0.01, ***p < 0.001,--- a95% Confidence intervall cannot be calculated since standard errors of indirect effects are not known for APE method, −--bConf_ratio cannot be calculated in case of direct effect = 0, −--cConf_ratio and Conf_pct cannot be meaningfully interpreted in case of non-significant total effect
\