Working Paper: NBER ID: w19450
Authors: Charles Courtemanche; Samir Soneji; Rusty Tchernis
Abstract: We propose a Bayesian factor analysis model to rank the health of localities. Mortality and morbidity variables empirically contribute to the resulting rank, and population and spatial correlation are incorporated into a measure of uncertainty. We use county-level data from Texas and Wisconsin to compare our approach to conventional rankings that assign deterministic factor weights and ignore uncertainty. Greater discrepancies in rankings emerge for Texas than Wisconsin since the differences between the empirically-derived and deterministic weights are more substantial. Uncertainty is evident in both states but becomes especially large in Texas after incorporating noise from imputing its considerable missing data.
Keywords: Bayesian Factor Analysis; Health Rankings; Mortality; Morbidity
JEL Codes: C11; I14
Edges that are evidenced by causal inference methods are in orange, and the rest are in light blue.
| Cause | Effect |
|---|---|
| Bayesian factor analysis model (C11) | more accurate ranking of health outcomes (I14) |
| data-derived factor weights differ significantly from deterministic weights (C69) | discrepancies in health rankings (I14) |
| incorporating uncertainty into rankings (D81) | misleading geographic disparities (R23) |
| health rankings should not be interpreted without considering uncertainty (I14) | meaningful differences in rank (C46) |
| model used (C52) | identification of least healthy counties (I14) |
| method identifies counties with higher probability of being in least healthy quintile (I14) | conventional rankings do not account for uncertainty (D81) |