Distance to Facility and Healthcare Utilization in Tanzania: Unbiased and Consistent Estimation Using Perturbed Location Data

Ryoko Sato , Harvard School of Public Health
David Canning, Harvard University
Mahesh Karra, Boston University
Bilikisu Elewonibi, Harvard School of Public Health
M. J. Mahande, Kilimanjaro Christian Medical University College
Sia Msuya, Killimanjaro Christian University College, Moshi
Iqbal H. Shah, World Health Organization (WHO)

Although access to health facilities is a major contributor to healthcare utilization, the accurate measurement of distance effects is difficult using Demographic and Health Surveys (DHS) since household location data is perturbed to protect respondents’ confidentiality. We show the attenuation bias due to perturbation by using a survey of 3950 women we conducted in Arusha, Tanzania where we can estimate distance effects both with accurate, and perturbed, data. Unbiased and consistent estimation using perturbed data is possible by numerical integration over all possible true locations, weighted by the probability the household is at that location. We show that for our Arusha sample this method produces estimates centered on those found with the true data. We then apply our method to DHS data showing that our estimates using numerical integration produce significantly larger effects than those found by naive regressions using perturbed data.

See paper

 Presented in Session 47. Spatial Methods