<p>Spatial disaggregation models may employ a cross-scale approach, namely a regression predicting values of a variable at small area level, although observations are at an aggregated region level. These constitute target and source areas in dasymetric mapping terminology. Existing cross-scale models consider single target variables only, with limited role for ancillary indicators. We consider here more general causally oriented approaches with disaggregation modelling as a constituent feature. The first involves multiple target indicators of a latent neighbourhood health risk factor to predict observed health outcomes. The second, a multiple outcome framework, considers prediction of multiple target indicators from observed health indicators. We use Bayesian inference which adapts straightforwardly to multiple likelihoods and sharing of spatial information via random effects. Two case studies are included. The first considers four adverse environmental characteristics as target indicators for 1002 neighbourhoods (substance use, sexual infection rates, violent crime and drug crime), though they are only observed for 33 London local authorities. Modelled adverse environmental measures are combined in a single neighbourhood latent construct to predict psychosis prevalence. The second case study considers impacts of neighbourhood crime on indicators of neighbourhood social cohesion, so is a social indicator application rather than health risk application. We assess impacts of crime—expected, and confirmed, to be negative—on feelings of belonging to the neighbourhood, neighbourly interactions, neighbourhood trust, and formal volunteering.</p>