The ACP Enrollment Performance Tool is based on the premise that enrollment decisions involve both individual and community-wide circumstances. A household’s decision to enroll is tied to its income or other household characteristics, but other factors such as housing costs, share of occupied housing units, the presence of anchor institutions like public libraries, and population density (i.e., whether a place is urban or rural) can play a role.
The ACP Enrollment Performance Tool examines these individual and community-wide issues by fitting a regression model that predicts ACP enrollment at the 5-digit zip code geography using variables that capture eligibility, as well as socio-economic characteristics (SES) of the community, and adoption of digital tools. More specifically, the model relies on:
- ACP eligibility. The model uses this estimation approach for characterizing the number of households in a given area that are eligible to enroll in the benefit. The approach captures variation across states for those eligible for the program who qualify through participation in such programs as the Supplemental Nutrition Assistance Program (SNAP) or Medicaid.
- SES characteristics. The model uses American Community Survey (ACS) 5-year estimates (2017-2021) for information on the share of household heads that are not in the labor force, not employed, with incomes under $15,000 annually, under the age of 18, over the age of 65, with college degrees or more, high school graduates or less, foreign born, rent burdened (i.e., they pay more than 30% of income for rent), and moved from a different county in the past year. It also includes the share of households with teens not in school and not working, as well as the share of occupied homes in an area.
- Digital tools. The model uses ACS 5-year data on the share of households with wireline broadband subscriptions at home, the share of households that rely only on cellular data for home internet, and the share of households with no computers.
- Libraries. Data from the Institute for Museum and Library Services shows whether the main branch of a public library is located in a particular zip code.
- Other. The model includes data on the degree of “ruralness” for a given zip code, the share of population that is Native American, Black, and Hispanic, and the number of Lifeline service providers in a zip code. It also controls for state fixed effects, that is, it includes a dummy variable for each state that captures unobserved characteristics specific to each state.
With this data, an ordinary least squares (OLS) regression model is run for 25,281 zip code areas in the United States and captures how well these variables predict household ACP enrollment. The model explains 63% of the variation in ACP enrollment across these zip codes. The model excludes zip codes with fewer than 150 households because small sample sizes of many of the independent variables may make these zip codes unreliable for model inputs. Many of the more than 40,000 zip codes in the United States are excluded either because they do not contain ACP households or have fewer than 150 households.
The model’s parameters are used to predict ACP household enrollment levels for each 5-digit zip code used in the analysis. The difference between this prediction and actual number of ACP enrollees is the basis for the ACP performance metric. We place each zip code into one of five categories:
- Highest zip code areas whose actual ACP enrollment exceeds predicted enrollment by 40% (e.g., there are 100 households enrolled in ACP in a zip code area compared with the model’s prediction of 60).
- This group is 17% of all zip code areas.
- High zip code areas whose actual ACP enrollment exceeds predicted enrollment between 10% and 40%.
- This group is 21% of all zip code areas.
- Medium zip code areas where the difference between actual and predicted enrollment is between 10% and -10%.
- This group is 14% of all zip code areas.
- Low zip code areas are ones where actual and predicted enrollment is between -10% and -40%.
- This group is 17% of all zip code areas.
- Lowest zip code areas where the difference between actual and predicted enrollment is less than -40%.
- This final group is 31% of all zip code areas.