Direct and stable weight adjustment in non-experimental studies with multivalued treatments: analysis of the effect of an earthquake on post-traumatic stress

In February 2010, a massive earthquake struck Chile, causing devastation in certain parts of the country, affecting other areas, and leaving territories untouched. 2 months after the earthquake, Chile's Ministry of Social Development reinterviewed a representative subsample of its National Socioeconomic Characterization Survey, which had been completed 2 months before the earthquake, thereby creating a prospective longitudinal survey with detailed information of the same individuals before and after the earthquake. We use a new weighting method for non-experimental studies with multivalued treatments to estimate the effect of levels of exposure to the earthquake on post-traumatic stress. Unlike common weighting approaches for multivalued treatments, this new method does not require explicit modelling of the generalized propensity score and instead focuses on directly balancing the covariates across the multivalued treatments with weights that have minimum variance. As a result, the weighting estimator is stable and approximately unbiased. Furthermore, the weights are constrained to avoid model extrapolation. We illustrate this new method in a simulation study, with both categorical and continuous treatments. The results show that directly targeting balance instead of explicitly modelling the treatment assignment probabilities tends to provide the best results in terms of bias and root-mean-square error. Using this method, we estimate the effect of the intensity of the earthquake on post-traumatic stress. We implement this method in the new package msbw for R.