Week 2 - Finkelstein, Amy et al. (2012) : The Oregon Health Insurance Experiment: Evidence from the First Year.

What is a summary of this paper?

• Aim: To measure the causal effect of taking up Medicaid on health care utilisation, out-of-pocket expenditures, medical debt, and self-reported physical and

  mental health.

• Empirical strategy: use a lottery in Oregon which determined whether uninsured low-income adults could apply for Medicaid (OHP Standard).

• Findings: findings suggest that health insurance access through Medicaid increases hospital utilisation and health outcomes, and decreases health-related spending

  and debt.

What is the purpose of Table II in the paper?

• It is a balance table (common in randomised experiments)

• Shows those that received the treatment (in this case, the ability to apply for Medicaid) and those that did not (the control group that was not selected by lottery to apply) are similar based on observable characteristics.

• This is important for the research design because the

  authors need to establish that the treatment and control groups were not differentially selected from the full sample in order to establish causality.

• The table suggests that the treatment and control groups here are indeed “balanced,” since columns (2) and (4)

  show that the differences in means are statistically insignificant.

Why is the coefficient in row (1) of column (2) in Table III only equal

to 0.256 and not equal to 1?

• The coefficient in row (1) of column (2) in Table III captures the effect of being selected in the lottery on getting insurance under Medicaid.

• The fact that this coefficient is 0.256 indicates that not all lottery winners ultimately signed up for Medicaid (and that some lottery losers also have gained access to Medicaid for reasons other than the lottery).

• In the language of instrumental variables, this coefficient represents the “first stage” effect of the instrument (lottery) on the endogenous independent variable of interest (insurance coverage).

(Table IV) In your own words, what is the difference (in terms of interpretation, i.e. what they are supposed to be capturing) between the estimates in columns (2) and (3) in Table IV? Why are the estimates in column (3) always larger? You can answer intuitively, no econometrics needed.

• The estimates in column (2) represent the effect of winning the lottery on hospital admissions.

  • This is known as the “intent to treat” effect (ITT) because it captures the effect of the instrument (lottery) on the outcome of interest, rather than the effect of the treatment itself (insurance coverage).

• The estimates in column (3) instead measure the effect of taking up Medicaid on hospital admissions, using lottery as an instrument for insurance access.

  • The estimates in column (3), which are called “local average treatment effects” (LATE), capture the average effect among those who take up insurance because they won the lottery. The LATE estimates are always higher than the ITT estimates because of the LATE theorem which says: LATE =ITT /FirstStage.

  • In words, the LATE of health insurance coverage is the effect of the lottery on health outcomes, divided by the probability of insurance take-up after winning the lottery.

In your own words, what is the main lesson from Figure I?

• Figure I shows that “winning” the Medicaid lottery is associated with lower out-of-pocket spending across all percentiles.

• Note that there are zero effects in the lower percentiles because many people simply don’t make out-of-pocket payments.

• Panel A shows that at any given percentile the corresponding payments are lower for the treated than for the control group.

• Panel B shows that across all percentiles where there are out-of-pocket payments, winning the lottery has a negative effect on such payments, although this effect is not statistically significant at very high percentiles.

What is the main takeaway from this paper?

The main takeaway is that expanding Medicaid led to overall positive health and financial outcomes in the context of the Oregon Health Insurance Experiment. Although people use more health resources when insured, we see an improvement in their health outcomes and reductions in their debt.

What is the main limitation of this paper?

The paper only measures short-term outcomes (within one year after the experiment concludes), and health gains may take time to materialise.

The study does not capture potential general equilibrium effects. In other words, the study implicitly assumes that the supply side of the health insurance market is not responding to the experiment. This could be important if, for instance, the influx of new Medicaid patients leads to differences in how physicians treat the overall population of patients.