Skip to main content

Table 2 Mediated effects of sex on obesity via the selected variables

From: Do routinely measured risk factors for obesity explain the sex gap in its prevalence? Observations from Saudi Arabia

 

Effect of sex on mediator M (path a in causal diagram)1

Effect of mediator M on obesity (path b in causal diagram)2

Effect of sex on obesity (path c’ in causal diagram)2

Indirect Effect3

Direct Effect3

Total Effect3

Prop. of Total Effect mediated

Mediator (M)

EST (CI)

EST (CI)

EST (CI)

Mean (CI)

Mean (CI)

Mean (CI)

 

Education

−3.62 (−3.99,–3.25)

−0.003 (−0.02,0.01)

0.66 (0.48,0.87)

0.002 (−0.01, 0.01)

0.136 (0.10, 0.17)

0.139 (0.11, 0.17)

0.015 (0.01,0.02)

Household Income

−0.57 (−0.73,-0.40)

0.20 (0.06, 0.34)

0.69 (0.54, 0.84)

−0.005 (−0.01,–0.001)

0.143 (0.11, 0.17)

0.137 (0.11, 0.17)

−0.04 (−0.05,0.03)

Physical Activity Levels

−0.65 (−0.91,–0.38)

−0.02 (−0.17, 0.13)

0.66 (0.46, 0.86)

0.0004 (−0.003, 0.004)

0.131 (0.10, 0.16)

0.132 (0.10, 0.16)

0.003 (0.002, 0.004)

Time spent sedentary

−0.11 (−0.32, 0.11)

0.004 (−0.15, 0.14)

0.71 (0.54, 0.88)

−0.00005 (−0.001, 0.002)

0.1469 (0.12,0.18)

0.1468 (0.12, 0.18)

−0.0003 (−0.0004,-0.0002)

Fruit/Veg consumption

−0.19 (−0.47, 0.08)

0.23 (0.0, 0.41)

0.66 (0.46, 0.86)

−0.002 (−0.005,–0.0007)

0.13 (0.10, 0.16)

0.128 (0.10, 0.16)

−0.02 (−0.02,-0.01)

Smoking

−3.24 (−3.83,–2.64)

−0.16 (−0.37, 0.06)

0.67 (0.49, 0.85)

0.007 (−0.004, 0.02)

0.139 (0.11, 0.17)

0.145 (0.12, 0.17)

0.05 (0.04,0.06)

  1. Table provides estimates obtained from Stata’s medeff function. Estimates in the first three columns are obtained from the initial regression models that the function fits (described in Methods).
  2. (1) The effect of sex on the mediator is obtained from a model regressing the mediator (as outcome) on sex and age. The coefficient and 95% confidence intervals (CI) for sex is shown in the first column. EST (estimate) is either beta coefficient from linear regression (in the case of education variable as the outcome) or log odds (all other variables. For significance at the 5% level, the 95% CI should not cross over 0.
  3. (2) The effect of the mediator on obesity and the effect of sex on obesity are obtained from a single model regressing the outcome (obesity) on sex, the mediator, age, region, and mediator-outcome confounders. The log odds (CI) estimate for the mediator and sex are shown in the second and third column, respectively.
  4. (3) The effects in columns 4–7 are derived by medeff based on the parameters in columns 1–3: Total Effect estimate is expressed as a proportion of the change in the probability of obesity. Similarly, indirect effect of sex on obesity via each of the mediators and direct effect of sex on obesity are also expressed as proportions. The direct effect is equivalent to c’ (third column) transformed on a probability scale. The last column reports the ratio of indirect effect to the total effect. A negative proportion of total effect mediated reflects inconsistent mediation.
Back to article page

BMC Public Health

ISSN: 1471-2458

Contact us