Local Estimates of Internet Adoption

In 2023, the National Telecommunications and Information Administration and the Census Bureau began an experimental project to study the feasibility of—and ultimately to produce—estimates of Internet adoption for small, sub-state areas during a single year to address this knowledge gap and better serve the policymaking process. Using techniques that have been successfully employed in other data products, Census Bureau experts are combining existing data from key household surveys with auxiliary data that are known to correlate with Internet adoption rates. By using a predictive model, the Census Bureau team can produce estimates for less populous geographies or groups that have both smaller margins of error than equivalent estimates based on survey data alone, and reduced risk that such estimates can be used to identify individual respondents. Those two features of small area modeling make it possible to publish more granular estimates than would otherwise be permissible or recommended for estimates generated entirely from survey data. NTIA invites all suggestions for improvements to the initial experimental model. The following questions serve as a non-exhaustive guide to some of the issues commenters may wish to address:

1. Should NTIA be aware of any potential applications where Project LEIA (Local Estimates of Internet Adoption) could make a particularly substantial contribution to policy research or development? Would any future work on Project LEIA help improve or expand these contributions?
2. In the feasibility report, the Census Bureau describes the methodology it used in the experimental model, and lists a number of potential predictor variables it tested before selecting the ones used in these initial estimates. Are there additional variables or data sources that should be considered to improve the model’s predictive power? Should we consider any methodological refinements or modifications to this model to improve its performance?
3. While the current experimental model only produces estimates at the county level, the same principles can potentially be applied for other small geographies and populations. During the next phase of Project LEIA, NTIA and the Census Bureau intend to experiment with creating census tract-level estimates. Are there other small geographies or populations for which model-based estimates of Internet adoption might be beneficial? What relevant data sources at that level could be considered to help generate these estimates?
4. In this first phase, we decided to analyze the percentage of households subscribed to wired Internet services. We did this because (a) the variable is useful for policymaking and (b) sufficient data were available to accurately fit a model. However, this is not the only metric that possibly could be modeled through future work. In addition to considering other variables from the ACS questions on computer and Internet use, we are also interested in applying small area modeling to more detailed questions from the NTIA Internet Use Survey. What metrics from either survey could we prioritize for future work under Project LEIA?
5. Is there anything else NTIA should take into consideration when contemplating the further development of Project LEIA?