This study examines risk selection among nine health plans competing f
or 16,182 employees of one large firm in 1989: one conventional fee-fo
r-service plan, one group-model health maintenance organization (HMO),
and seven network and independent practice model HMOs. We develop and
compare measures of risk using weights based on HMO and fee-for-servi
ce expenditure data, respectively. We use a multiequation statistical
model to develop two sets of utilization and expenditure weights for e
nrollees in each plan. One set of weights, based on discharge abstract
s and outpatient records from the large group-model HMO, measures how
much each of the nine groups of employees and dependents would have sp
ent, had they been enrolled in a stringently managed plan with no cons
umer cost sharing. The other set of weights, based on fee-for-service
claims data, measures how much each group would have spent, had it bee
n enrolled in an unmanaged health plan with significant coinsurance an
d deductibles. Predicted annual expenditures per enrollee exhibit a 23
% range from lowest (favorable selection) to highest (adverse selectio
n) risk plans using the HMO weights and a 17% range using fee-for-serv
ice weights. The fee-for-service plan and group-model HMO with large e
nrollments have risk mixes near the center of the spectrum. Smaller HM
Os exhibit the extreme forms of both favorable and adverse selection.
The statistical methods adopted in this study can be used to risk-adju
st capitation payments to competing health plans. As mergers among HMO
s and group purchasing arrangements among employers increase the avera
ge enrollment in each plan from each payor, however, risk differences
among plans will be attenuated and the need to risk-adjust payments wi
ll be less severe.