Molly,
This is an important question. Randomization might be feasible but it's expensive because of course medical care is expensive. Still the differential per patient between medicaid and private payer might not be so large. Could you do an effective study by simply masking providers to the insurance status of the patient?
Maria