EXHIBIT 4.

Risk-Selection Models Predicting Medical Care Spending In 2004

	Model 1a		Model 2b		Model 3c		Model 4d
Variable	Coeff.	SE	Coeff.	SE	Coeff.	SE	Coeff.	SE
PPO in 2004
PPO 2004–PPO 2005 (stay)	$10,002	$ 160	$11,533	$ 318	$7,385	$ 549	$5,452	$ 201
PPO 2004–HRA 2005 (switch)	7,067***	883	8,820***	933	$4,813***	1,042	3,684**	851
HRA in 2004
HRA 2004–HRA 2005 (stay)	5,135	448	6,690	534	2,859	728	1,896	439
HRA 2004–PPO 2005 (switch)	7,721**	1,139	9,350**	1,172	5,462**	1,247	4,057*	1,095
Includes potentially observable characteristics?	Model 1		Model 2		Model 3		Model 4
Age, sex only	No		Yes		Yes		Yes
Plus wage, job type, race, presence of chronic conditions	No		No		Yes		Yes
Plus fuller risk adjustment	No		No		No		Yes

SOURCES: Alcoa Inc. employee files and medical claims files (2003–2005).

NOTES: Significance levels indicate results from F-tests comparing 2004 expenditure differences between enrollees staying and switching health plans in 2005. N = 13,265 employees. PPO is preferred provider organization. HRA is health reimbursement arrangement. SE is standard error.

^aModel 1 regresses 2004 medical spending on the four plan-switching variables: PPO–PPO, PPO–HRA, HRA–HRA, and HRA–PPO.

^bModel 2 regresses 2004 medical spending on the four plan-switching variables as in Model 1, and also includes sex and three age categories (18–35, 36–50, 51–63).

^cModel 3 regresses 2004 medical spending on the four plan-switching variables, sex, and age as in Model 2, and also includes wage quartiles, job type (hourly versus salaried), race (white versus other), and presence of chronic conditions (diabetes, asthma, congestive heart failure/coronary artery disease).

^dModel 4 regresses 2004 medical spending on the four plan-switching variables, sex, age, wage, job type, race, presence of chronic conditions as in Model 3, and we also include the Hierarchical Condition Categories (HCC) risk adjustment from the Centers for Medicare and Medicaid Services.

^*p < 0.10

^**p < 0.05

^***p < 0.01