Identifying health effects causally conferred by built environment exposures is challenging due to uncertainty about the spatial scale that is relevant for exposure assessment and confounding due to unmeasured person-level factors. We propose a difference-in-differences parameterization for the spatial temporal aggregated predictor (STAP) model to address the question of spatial scale and condition on unmeasured, time-invariant person-level confounders. As with STAP, the model uses the distances between study participants’ locations and environmental features (e.g., supermarkets) to define a weighted exposure count, where weights are a function of distance with parameters interpretable as the spatial scale. In addition, the predictor is written as the difference in exposure during the current visit from the person-level average exposure, such that the effect of interest is interpreted as the change in the outcome associated with person-level change in exposure, causal interpretation. Implemented using a custom No U-Turn Hamiltonian Monte Carlo Sampler in C++, the model is used to estimate the causal effect of healthy food availability and body mass index within an ageing cohort.