Bayesian hierarchical piecewise regression models: a tool to detect trajectory divergence between groups in long-term observational studies

Table 3 Analyses of inter-cohort differences in BMI trajectories: assessment of Bayesian model complexity (effective number of parameters pD), and fit (deviance information criteria DIC) for each candidate model

	Model	Females	PP p-val	Males	PP p-val
Unconditional	A	26910 (2544)	0.72	197837(2223)	0.52
Birth cohort (int β₀)	B	26811 (2455)	0.70	19872 (2232)	0.70
Birth cohort (childhood slope β₁)	C	26759 (2489)	0.34	19849 (2175)	0.63
Birth cohort (Adulthood slope β₂)	D	26645 (2358)	0.67	19857 (2263)	0.68
Birth cohort (change point CP)	E	26395 (2599)	0.60	19862 (2211)	0.63
Birth cohort (CP and β ₂ )	F	26390 (2671)	0.49	19877 (2255)	0.43
Birth cohort (CP, β₂ and β₁)	G	26783 (2775)	0.48	19945 (2342)	0.53

Reported are: DIC (pD), and posterior predictive p-values (PP p-val). Best fitting models for each sex indicated in bold characters. (Convergence was not reached for the most complex model where all 4-trajectory parameters (i.e.β_0, β_1, β_2, and CP) were adjusted for birth cohort effects)

ISSN: 1471-2288