TABLE 1. Analysis of Deviance for nested Models in Men and Women.
Residual df | Residual Deviance | Nested Model comparisons | Tested effect | Df | Deviance Difference | P-values | |||||
---|---|---|---|---|---|---|---|---|---|---|---|
men | women | men | women | men | women | ||||||
model 1 | Age | 57 | 131.5 | 249.6 | |||||||
model 1a | Age-drifta | 56 | 51.9 | 62.5 | 1 and 1a | Linear effect of cohort or period adjusted for age | 1 | 79.5 | 187.1 | <0.0001 | <0.0001 |
model 2a | Age-cohort | 38 | 25.8 | 26.7 | 1a and 2a | Nonlinear cohort effect adjusted for age | 18 | 26.1 | 35.7 | 0.09 | 0.008 |
model 3 | Age-period-cohort | 36 | 23.5 | 25.7 | 2a and 3 | Nonlinear period effect adjusted for age and cohort | 2 | 2.3 | 1.0 | 0.32 | 0.59 |
model 2b | Age-period | 54 | 49.3 | 61.7 | 2b and 3 | Nonlinear cohort effect adjusted for age and period | 18 | 25.8 | 35.9 | 0.10 | 0.007 |
model 1b | Age-drift | 56 | 51.9 | 62.5 | 1b and 2b | Nonlinear period effect adjusted for age | 2 | 2.6 | 0.8 | 0.27 | 0.66 |
The linear evolution of obesity prevalence observed during the survey period, cannot be specifically attributed to cohort or to period, and is called the “drift.” This model includes the period effect, or the cohort effect, as a quantitative variable. Whatever the variable (period of cohort), the fit and the estimation of the drift effect are identical.