Taking a data-driven approach to consumer outreach is at the core of Enroll America’s strategy for helping consumers gain coverage. A key way we use data to improve outreach is through a sophisticated model that helps predict at the individual level whether particular consumers are uninsured.
Predicting Where the Uninsured Are
Enroll America has a database drawn from a variety of public and commercial sources that includes basic contact and demographic information for more than 200 million U.S. adults. For each of the non-senior adults in our database, the uninsured model assigns a score between zero and 100 that represents the probability that an individual is uninsured. Enroll America worked with Civis Analytics, a technology and research firm, to build the first uninsured model in the spring of 2013 based on thousands of phone interviews with consumers combined with other information in the database.
Enroll America teamed up again with Civis Analytics over this past summer to survey and analyze data to update our uninsured model heading into the second open enrollment period. To rebuild the model, we surveyed 8,191 consumers about their current health insurance status, health insurance status in 2013, income, and employment status. The results of this survey were used in tandem with data in Enroll America’s database and from the previous uninsured model to create an updated algorithm that predicts the likelihood that a person is uninsured in 2014. You can read more about how the model was built in this New York Times article.
Identifying Changes in the Uninsured Population After OE1
Because Enroll America had an uninsured model in 2013 and now has an updated one in 2014, we can compare numbers of those who are likely uninsured from before and after the first open enrollment period. Below you can see how the distribution has shifted between 2013 and 2014.
The average uninsured score — which corresponds to the share of individuals in our nationwide database that we assume to be uninsured — decreased 5.1 percentage points, from 16.4 in 2013 to 11.3 in 2014. This shift indicates a drop in the uninsured rate, which is in line with public polling findings. The model also allows us to see more granular shifts in the uninsured population. Most subgroups moved at roughly the same rates, but we saw higher rates of enrollment among age group 26 to 35, Latinos, women, and those that lived in states that expanded Medicaid. If you are interested in learning more about the changes in the uninsured population, the New York Times has put some beautiful graphics based on the uninsured model.
Using the Uninsured Model to Improve Outreach
The model will serve as a tool for directing our outreach strategy in the months to come. We use it to prioritize ZIP codes and subgroups so that we focus our outreach conversations and events on consumers who are most likely to be uninsured. In 2013, among our 11 states with grassroots outreach staff, we were active in fewer than a quarter of the ZIP codes in the states, but we were still are able to reach nearly half of the uninsured (a two-fold increase in efficiency over targeting entire states).
We are also making this updated uninsured model available to partners via our data tool, Get Covered Data. If you are interested in finding out more about accessing Get Covered Data, please email PartnerData[at]EnrollAmerica.org.