Effectiveness variation of different census outreach activities: An empirical analysis from the state of Illinois using machine learning and user interface technologies for participatory data collection

Abstract

Major social changes stem from policies that their developments heavily rely on public participation in the decennial U.S. Census. For individual states, the population counts determine representation in the U.S. House of Representatives and the amount of federal funding each state receives, leading to state outreach efforts to encourage participation in the census. The present empirical study uses advanced data collection technologies based on a user interface (UI) and machine learning techniques (k-means clustering, random forest, gradient boosting) to analyze the effectiveness of the outreach activities organized by the State of Illinois. As a result, we assess various types of outreach activities in different geographic areas with distinct socio-economic and demographic characteristics. Findings from the models in this study show that outreach activities classified as “direct engagement (1-on-1)” and “single events” bear the highest impact, especially in predominantly low-income minority communities in metropolitan areas. However, socio-demographic characteristics are found to be generally more influential on response rates than outreach activities performed in the area, and in many underperforming areas, a high number of activities does not correlate with an increase in response rates. The findings of this research could assist in structuring outreach efforts in different countries and the U.S.