Plan Value-Added: Evaluating Medicaid Managed Care
Plans Using Random Assignment
Craig Garthwaite and Matthew J. Notowidigdo
∗
March 2019
[PRELIMINARY AND INCOMPLETE]
Abstract
We use the random assignment of more than 100,000 households in South Car-
olina to different Medicaid managed care plans to estimate the effect of health care
plan assignment on health care utilization, spending, and plan “effectiveness” evalu-
ated using industry-standard measures of preventive care and chronic disease man-
agement. We find large differences in plan effects across many different health care
categories, including ER visits, inpatient hospitalizations, well-child doctor visits, and
cancer screenings. We find that the estimated plan effects are generally positively cor-
related across health care categories, and with the length of time households remain in
a plan before switching plans or leaving Medicaid. We find evidence of a large amount
of selection bias when including households making “active” plan choices, with the con-
ventional (non-randomized) estimates often significantly overstating differences across
the plans. Given the conceptual similarity to School Value-Added estimates used in
school choice programs, we call our plan effect estimates “Plan Value-Added”, and
show using simulations calibrated to our experiment that incorporating randomization
into the evaluation of Medicaid managed care plans can help policymakers improve the
incentives for plans to increase quality and reduce costs, as well as help households
make more informed plan choices.
JEL codes: I13, H51, H75.
Keywords: Medicaid, Managed Care Organizations (MCOs), Health Care Utilization, Risk
Adjustment, Quality Rating, School Value-Added.
∗
Affiliations: Northwestern University Kellogg School of Management and the National Bureau of Eco-
nomic Research (Garthwaite); Northwestern University Department of Economics, Kellogg School of Man-
agement, and Institute for Policy Research; National Bureau of Economics Research; and J-PAL North
America (Notowidigdo). We are grateful to Nathaniel Downes, Eilidh Geddes, Pinchuan Ong, and Ting
Wang for excellent research assistance. We thank our excellent partners at South Carolina DHHS (Depart-
ment of Health and Human Services) and RFA (Revenue and Fiscal Affairs), especially Christina Galardi
and Bryan Amick at SC-DHHS for their tireless efforts and assistance throughout the project. We also thank
Chris Finney and Sarah Crawford at SC-RFA for assistance in preparing the data for analysis, and Joshua D.
Baker and Christian Soura for their feedback, vision, and willingness to partner with J-PAL North America.
We thank Josh Angrist, Amy Finkelstein, and Jon Gruber for helpful comments. The experiment reported in
this study is registered in the AEA RCT Registry and the unique identifying number is AEARCTR-0002762.
We gratefully acknowledge financial support from the J-PAL North America and the Laura and John Arnold
Foundation.
1
