Partially Identifying Treatment Effects with an Application to Covering the Uninsured

Abstract

We extend the nonparametric literature on partially identified probability distributions and use our analytical results to provide sharp bounds on the impact of universal health insurance on provider visits and medical expenditures. Our approach accounts for uncertainty about the reliability of self-reported insurance status as well as uncertainty created by unknown counterfactuals. We construct health insurance validation data using detailed information from the Medical Expenditure Panel Survey. Imposing relatively weak nonparametric assumptions, we estimate that under universal coverage monthly per capita provider visits and expenditures would rise by less than 8 percent and 16 percent, respectively, across the nonelderly population. s t r a c tWe extend the nonparametric literature on partially identified probability distributions and use our analytical results to provide sharp bounds on the impact of universal health insurance on provider visits and medical expenditures. Our approach accounts for uncertainty about the reliability of self-reported insurance status as well as uncertainty created by unknown counterfactuals. We construct health insurance validation data using detailed information from the Medical Expenditure Panel Survey. Imposing relatively weak nonparametric assumptions, we estimate that under universal coverage monthly per capita provider visits and expenditures would rise by less than 8 percent and 16 percent, respectively, across the nonelderly population.

Covering The Uninsured In 2008: Current Costs, Sources Of Payment, And Incremental Costs

“…According to the CPS, 47.0 million Americans were uninsured in 2006, which we 25 They found that even though there is uncertainty about the number of people lacking insurance, under reasonable nonparametric assumptions, estimates from MEPS of the maximum cost of covering the uninsured are not much affected by this uncertainty. We estimate that government spends nearly $43 billion-roughly 75 percent of total uncompensated care costs-through Medicaid DSH and supplemental payment programs, Medicare DSH and IME payments, various direct care programs, and state and local tax appropriations.…”

HIV estimation using population‐based surveys with non‐response: A partial identification approach

“…In practice, more than one instrument is usually used in a particular study, see, for example, Lawlor et al 16 and Kreider 20 . Suppose there are $$$ L $$$ candidate instruments, and all we can assume is at least one of the $$$ L $$$ candidates is valid.…”

Concave-Monotone Treatment Response and Monotone Treatment Selection: With an Application to the Returns to Schooling

“…Some of these relationships can be estimated by using our bounds because they represent concave functions (e.g., diminishing marginal returns). For example, our bounds are applicable to the relationships studied in González (2005), Gerfin and Schellhorn (2006), Blundell, Gosling, Ichimura, and Meghir (2007), Pepper (2007, 2008), Gundersen and Kreider (2009), Kreider and Hill (2009), de Haan (2011), Kang (2011, Kreider, Pepper, Gundersen, and Jolliffe (2012), and Huang, Maassen van den Brink, and Groot (2012).…”