A policy oriented model of child growth and mortality is developed in the context of a stochastic state space model. The model incorporates unobserved heterogeneity as an unmeasured covariate which affects both mortality and an observed time varying covariate. It is demonstrated using Monte Carlo simulations that a model ignoring this unobserved heterogeneity will give biased parameter estimates; parameters are found to be unbiased if a model which allows for an unobserved variable is estimated. Monte Carlo simulations are then used to test the robustness of the model to misspecification of the distribution of the unobserved covariate. Estimates of the change in child survival are obtained using dynamic equations derived from a Kolmogorov-Fokker-Planck (KFP) equation. It is shown that the model which ignores unobserved heterogeneity produces incorrect estimates of the change in mortality that would result if certain types of mortality intervention programs were implemented.