Measurement Error in Discrete Health Facility Choice Models:an Example from Urban Senegal

A number of authors have utilized health facility choice models to determine how individuals in developing countries evaluate the tradeoff between the price, quality, and indirect costs of obtaining medical care. A common problem in this literature is that researchers only observe the type of facility that individuals report visiting (e.g., public or private hospital, health center or dispensary, or traditional healer) and, therefore, must assume individuals visit the nearest facility of the type they report. This matching procedure creates measurement error in the choice variable, which may introduce bias in parameter estimates. In this research, we use a data set from urban Senegal that allows for a precise individual-health facility match to estimate consumer preferences for health facility characteristics related to maternal health and family planning services. Using actual rather than imputed choices, we find that consumers prefer high quality health facilities that are nearby. Given the preference for quality, our findings indicate that in contrast to the typical assumption in the literature, individuals frequently bypass the facility nearest their home. When we estimate models using the mismeasured choice variable, the results show a significant bias in preference estimates; most notably, these models systematically overestimate distaste for travel. To highlight a potential consequence of biased preference estimates in this setting, we conclude by simulating the impact of an actual policy that took place in 2014; namely, the opening of a new facility in a previously underserved area of Dakar, Senegal.