This research investigates the degree that estimates of the magnitude
of small area variations in hospitalization rates depend on both the e
stimation method and the number of years of data used. Hospital discha
rge abstracts for patients 65 and older from acute care hospitals in M
assachusetts from 1982 to 1987 were analyzed. The SCV statistic, the a
pproach used in many current small area variation studies, and empiric
al Bayes (EB), an approach that adjusts more fully for the effect of r
andom variation, were compared. EB estimates based on 3 years of data
were best able to predict future area-specific hospitalization rates.
Compared to EB estimates using 3 years of data, the SCV statistic with
1 year of data overestimated the median amount of systematic variatio
n by over 70% for the 68 conditions studied; with 3 years of data, the
SCV overestimated the median by 55%. Regardless of method, the same c
onditions were identified as relatively more variable and the same geo
graphic areas were found to have higher than expected hospitalization
rates. The magnitude of differences in hospitalization rates depends o
n how the data are analyzed and how many years of data are used. Hospi
talization rates across small geographic. areas may vary substantially
less than reported previously.