CHS: Small: Protecting Election Integrity Via Automated Ballot Usability Evaluation

点击翻译按钮获取中文摘要

Ballots used in elections are not always well-designed. Flaws in ballot design can lead voters to make errors, and this can affect the outcome in a election. While we know how to design better ballots, there are too many jurisdictions and too many different ballots for human usability experts to manually evaluate or conduct usability studies on all of them. A usability evaluation method that scales to a problem this large does not yet exist. This project will do the basic science to support the initial development of a computational model of human performance that is able to automatically analyze a ballot and flag areas where there are potential design flaws. The ultimate goal is to develop a web-based tool that local election officials could use to check their ballots before they are used in a real election, thus preventing errors before they happen. This has the potential to improve election integrity by ensuring that what is recorded on each ballot is actually what each voter intended. In addition, this research will contribute to our understanding of how errors emerge in other contexts where fillable forms are used, such as use of electronic health records. The project aims to address this problem by developing a tool that, when given a ballot as input, produces an assessment of whether or not that ballot is likely to lead to voter error, and if so, where on the ballot these errors are most likely to occur. This tool will be based on computational human performance models developed with the ACT-R cognitive architecture, which has been successfully applied to other usability problems by numerous researchers. The model will be developed based on an exploration of the space of ballot completion and visual search strategies available to voters and informed by eye-tracking data. ACT-R's current visual capabilities in terms of scene analysis are limited, and the architecture will be extended with capabilities to support detection of visual groups, which are critical in understanding how voters visually navigate a ballot. A preliminary version has already been developed and incorporated into preliminary models of voting; this project will further develop and evaluate both the visual grouping extension and the voting models. The models will be validated using ballots with known flaws as well as with new behavioral data.This award reflects NSF's statutory mission and has been deemed worthy of support through evaluation using the Foundation's intellectual merit and broader impacts review criteria.