Keywords: File Linking, Causal Effects, Multiple Imputation, Bayesian Analysis
Home-delivered meals offered by Meals on Wheels (MOW) programs provide critical support to community-dwelling, vulnerable older adults. Little is known about the health and healthcare utilization of clients who receive services from MOW. In the absence of unique identifiers, linking MOW clients file to Medicare Claims Data using exact matching on common variables identified only 61% of the individuals. Relaxing the linking criteria increases the proportions of individuals that are matched, but it also increases the number of erroneously matched individuals. We developed a two-step procedure to estimate the effects of MOW programs on healthcare utilization. First, we create multiple datasets in which MOW beneficiaries are linked to Medicare beneficiaries. Second, we matched Medicare beneficiaries that did not receive MOW to those that did. Using these matches we multiply imputed for each MOW beneficiary their counterfactual healthcare utilization if they did not receive MOW. This procedure properly propagates the errors in the linking process and the matching process, and can be used to estimate effects of interventions in other linked datasets when there are no unique identifiers.