Skip to main content
. 2017 Nov 9;20(1):1–16. doi: 10.1093/biostatistics/kxx058

Fig. 1.

Fig. 1.

The figure represents the developed EPS framework. The left hand side corresponds to (2.1) and (2.2): the area level confounders Inline graphic are estimated from the individual level confounders Inline graphic (2.1) and the EPS is estimated (2.2). Note that this model is only specified on the Inline graphic areas. EPS is represented by a circle as a latent quantity (not observed). The right hand side presents the EPS imputation and adjustment and is specified on Inline graphic: at the top the EPS is either obtained from (2.2) for Inline graphic and included as an observed quantity or missing for Inline graphic, imputed through the relationship with Inline graphic, Inline graphic, and the spatial structure (2.3) and included as a latent quantity. Thus the EPS is represented here as both a square and a circle as the combination of observed and latent quantities. At the bottom, the estimated and imputed EPS is included in the analysis model (corresponding to (2.4)) to provide confounder adjustment when assessing the effect of Inline graphic on Inline graphic. Note that the right hand side is separated from the left hand side as the EPS estimation is carried out disjointly from the imputation and adjustment (no feedback from the latter to the former). At the same time the imputation and adjustment are jointly carried out so that feedback is allowed from Inline graphic to influence the EPS imputation.