The Medicaid Counterfactual: How we ran the simulation

November 2, 2014 Civis Analytics

Last week, after the first article about our work on predicting the uninsured went out the Upshot posed an interesting question to Civis and Enroll: could we run a simulation wherein all states expanded Medicaid?

To understand how we did this, it helps to take a step back to understand a bit about our methodology. We interviewed thousands of people in 2013 and 2014 to ask them about their insurance status. In the last two years, we’ve analyzed hundreds of characteristics to confirm if there was or was not a mathematical relationship between possessing that characteristic and being uninsured. Out of that process, we have come up with the ~40 or so that gave us the best prediction. Many of these characteristics make immediate sense ““ age ““ while some of them are more surprising ““ such as whether or not the person had voted in a recent Midterm election, or the employment status of their spouse.

One of those important variables, not surprisingly, was Medicaid. By analyzing the survey responses, we found that individuals living in a Medicaid expansion state were statistically more likely to say that they had insurance, even when controlling for other factors such as income, age, and gender. The math also told us roughly the size of the difference.
To re-run the simulation, we just applied that ‘difference’ to people in every state, without changing anything else. In essence, what we are saying is: if everything about a person was exactly the same, but they happened to have Medicaid in the place they lived, how would that change the likelihood that that particular individual had insurance? We then rolled those numbers up from individuals to counties and states.

As the New York Times mentions, this is a thought experiment. We can only take past data on how Medicaid was expanded as our basis. But we know that Medicaid was not expanded randomly, and there are other things correlated with whether or not a state expanded Medicaid, and how successful that expansion might be, that we have not ““ and cannot ““ adjust for. So while the specifics are estimates the overall picture is clear: Medicaid made a significant difference.

Want to learn more about our work in predicting insurance? Contact us here. Please also take a look at our Work section to learn more about individual-level predictions we’ve done in other industries. And, of course, we are always hiring: Apply here you’d like to work with us solving interesting problems with data and analytics.

The post The Medicaid Counterfactual: How we ran the simulation appeared first on Civis Analytics.

Previous Article
On building stuff at AWS re:Invent 2014
On building stuff at AWS re:Invent 2014

Our founder, Dan Wagner, was recently featured at AWS re:Invent in a Founders Fireside Chat with Amazon CTO...

Next Article
Obamacare: Who Was Helped Most
Obamacare: Who Was Helped Most

“Now a large set of data “” from Enroll America, the group trying to sign up people for the program, and fr...