Table 1.

Impact of splitting covariates into separate within- and between-subject covariates. Ages for full 817 subjects ADNI dataset were used to construct 4 models: (1) Intercept and Age, (2) Intercept and centred Age, (3) Intercept, mean age per subject ${\displaystyle {\overline{\mathit{Age}}}_{\mathit{i}}}$, and intra-subject-centred age $\mathit{Age}-{\displaystyle {\overline{\mathit{Age}}}_{\mathit{i}}}$, and (4) Intercept, centred mean age per subject ${\displaystyle {\overline{\mathit{Age}}}_{\mathit{i}}}-\overline{\mathit{Age}}$, and intra-subject-centred age $\mathit{Age}-{\displaystyle {\overline{\mathit{Age}}}_{\mathit{i}}}$. The relative efficiency is shown for each model for 3 possible values of ρ, the common intra-visit correlation. Here, we define relative efficiency as the ratio between the variance of the GLS estimate and the variance of the SwE estimate.

		Relative efficiency
Model	Covariate	ρ = 0	ρ = 0.5	ρ = 0.95
1	Intercept	1	0.88	0.40
	Age	1	0.88	0.40
2	Intercept	1	0.94	0.89
	$\mathit{Age}-\overline{\mathit{Age}}$	1	0.88	0.40
3	Intercept	1	0.92	0.87
	${\displaystyle {\overline{\mathit{Age}}}_{\mathit{i}}}$	1	0.92	0.87
	$\mathit{Age}-{\displaystyle {\overline{\mathit{Age}}}_{\mathit{i}}}$	1	1.00	1.00
4	Intercept	1	0.94	0.89
	${\displaystyle {\overline{\mathit{Age}}}_{\mathit{i}}}-\overline{\mathit{Age}}$	1	0.92	0.87
	$\mathit{Age}-{\displaystyle {\overline{\mathit{Age}}}_{\mathit{i}}}$	1	1.00	1.00