Browsing by Author "Byrne, Michael D"

Now showing 1 - 6 of 6
  • Thumbnail Image
    Item
    Checking Your Answers: An Investigation of Alternative Review Screen Design for Electronic Voting Systems
    (2015-04-21) Holmes, Danae V; Kortum, Philip T; Byrne, Michael D; Lane, David M; Wallach, Dan S
    Verifying a ballot for correctness in an election is a critical task considering the large, negative repercussions of an incorrect ballot. Studies have shown weaknesses in the ballot review process in electronic voting systems, allowing up to 30% of the ballot to be changed without being noticed by 68% of voters (Everett, 2007). There is also a noted lack of research on the effect of navigation style on electronic voting system usability and review screen performance. In response to these issues, this study evaluated the usability and viability of alternative ballot verification and navigation methods in an electronic voting medium, specifically direct recording electronic (DRE) voting systems. Currently, most DRE’s employ an end-of-ballot review where all selections are confirmed at once at the end of the ballot, which has been proven to be ineffective. Several studies (Holmes and Kortum, 2013; Selker 2007) have also indicated that in-line confirmation, confirming each selection immediately after making it, and a combination of the two confirmation methods (Ghandi et al., 2005; Cohen et al., 1996) may prove to be a suitable alternatives. The current study tested these methods of verification in terms of performance and usability to determine whether they are viable methods of verification as well as to provide a benchmark for review screen performance in a DRE. The method of navigation through the ballot, the ability to move backwards through the ballot or not after selecting a candidate, was also tested for its impact on usability and performance. The verification methods were evaluated on three metrics of usability as defined by ISO 9241 part 11; efficiency (time to complete a ballot), effectiveness (errors), and satisfaction (subjective usability). Participants cast their ballot in a mock national election using a custom DRE interface. Results indicate that in-line and dual confirmation methods prove to be viable alternatives for DRE review screens. In-line and dual confirmation perform similarly to end-of-ballot confirmation in terms of effectiveness, but differed in other usability and performance aspects, though not necessarily in a negative way. The most efficient method is end-of-ballot review, and dual confirmation produced the longest time spent on the review screen. End-of-ballot confirmation produced the highest satisfaction ratings, though survey results indicated that dual confirmation may be the most appropriate method in terms of voting. Based on the results from this study, further studies should be conducted to determine which confirmation method performs best as an error prevention tool.
  • Thumbnail Image
    Item
    Comparing vector-based and ACT-R memory models using large-scale datasets: User-customized hashtag and tag prediction on Twitter and StackOverflow
    (2014-12-02) Stanley, Clayton; Byrne, Michael D; Kortum, Phillip; Subramanian, Devika
    The growth of social media and user-created content on online sites provides unique opportunities to study models of declarative memory. The tasks of choosing a hashtag for a tweet and tagging a post on StackOverflow were framed as declarative memory retrieval problems. Two state-of-the-art cognitively-plausible declarative memory models were evaluated on how accurately they predict a user’s chosen tags: an ACT-R based Bayesian model and a random permutation vector-based model. Millions of posts and tweets were collected, and both declarative memory models were used to predict Twitter hashtags and StackOverflow tags. The results show that past user behavior of tag use is a strong predictor of future behavior. Furthermore, past behavior was successfully incorporated into the random permutation model that previously used only context. Also, ACT-R’s attentional weight term was linked to a common entropy-weighting natural language processing method used to attenuate low-predictor words. Word order was not found to be strong predictor of tag use, and the random permutation model performed comparably to the Bayesian model without including word order. This shows that the strength of the random permutation model is not in the ability to represent word order, but rather in the way in which context information is successfully compressed. Finally, model accuracy was moderate to high for the tasks, which supports the theory that choosing tags on StackOverflow and Twitter is primarily a declarative memory retrieval process. The results of the large-scale exploration show how the architecture of the two memory models can be modified to significantly improve accuracy, and may suggest task-independent general modifications that can help improve model fit to human data in a much wider range of domains.
  • Thumbnail Image
    Item
    Computational Modeling of Voters' Checking Behavior & Checking Performance
    (2024-04-19) Chavez, Fabrizio; Byrne, Michael D
    In order to preserve election integrity, it is crucial for voters to check their ballots for errors and other potential anomalies. Previous research has explored voters’ ability to detect anomalous changes to their ballots and the factors affecting this capability. However, the specific strategies voters use when checking their ballots remain unexplored. This study aims to fill this gap by utilizing eye-tracking methods and computational modeling to examine voters’ checking behavior. To this end, an experiment was designed where participants took part in a fictitious election by interacting with a paper ballot layout displayed on a computer screen. The ballot interface was programmed to alter the voters’ selections. The findings reveal that voters’ ability to detect these anomalous selections varies and is influenced by their visual and checking strategies, along with the availability of a slate or voter guide. More specifically, these elements collectively impact voters’ performance, resulting in different anomaly detection rates depending on the combination of strategies used by a voter. Based on these data, ACT-R models were constructed that replicated the observed strategies. These models were validated by matching their anomaly detection performance to that of the voters and were then used to make predictions about voters’ ability to detect anomalies and their own errors. Key suggestions from this study are that voter guides should always be provided in voting booths, voters should avoid using a random search strategy, and voters should be reminded to check for both undervotes and wrong candidate selections.
  • Thumbnail Image
    Item
    Factors Influencing Speed-Accuracy Tradeoffs
    (2014-11-13) Zemla, Jeffrey Clark; Byrne, Michael D; Kortum, Philip T; Schnur, Tatiana; Batsell, Richard R
    Many simple decisions allow us to trade o between speed and accuracy. When time is critical, decisions can be made quickly but accuracy su ers. Conversely, one may spend more time making a decision which often results in more accurate decisions. Speed-accuracy tradeo s have been studied in a number of domains including motor control (Fitts, 1954), perception (Usher & McClelland, 2001), and higher order reasoning (Kahneman & Frederick, 2002). Recent research has examined a set of normative models for how one should trade o speed and accuracy (Bogacz, Brown, Moehlis, Holmes, & Cohen, 2006); that is, how long someone should spend deliberating prior to action in order to maximize some reward. However, empirical work has shown haphazard adherence to these normative models (e.g., Zacksenhouse, Bogacz, & Holmes, 2010). While some subjects behave optimally, many do not. In two experiments, several factors that a ect speed-accuracy tradeo s in a perceptual decision-making task are investigated. In one experiment, it was found that feedback and shorter blocks not only improved participants’ task ability, but also resulted in more optimal speed-accuracy tradeo s. In a second experiment, manipulating trial di culty and subjects’ awareness of di culty level a ected task performance. However, despite predictions from a normative theory, participants did not engage in an optimal speed-accuracy tradeo policy.
  • Thumbnail Image
    Item
    Modeling Password Entry on Mobile Devices: Please Check Your Password and Try Again
    (2015-04-21) Gallagher, Melissa Ann; Byrne, Michael D; Kortum, Philip; O'Malley, Marcia; Lane, David M
    Despite being recognized as a fundamentally flawed system, password authentication is a widely deployed security feature on desktop and mobile systems. Inputting complex passwords on mobile devices can be an onerous task. The composition of the passwords creates a unique challenge for people to input as not all characters are displayed on the keyboard at the same time, forcing the user to switch between multiple screens. While previous studies of text input on mobile devices have focused on typing words and phrases, little work has been done examining the effects screen switching has on text input. Three experiments were conducted in which subjects typed strings similar to secure passwords. Subjects were considerably slower typing password-like strings than typing standard text. Uncertainty about the location of symbols was a key factor in this slowdown. One of the largest contributors to the number of errors made was the size of the keyboard keys. This source of error suggests technologies that may aid error prevention. The results from these studies informed an ACT-R model of the task. The timing data generated from the model fits the experimental results well. The strategy that the model employs depends on the type of character it is trying to input providing further evidence that finding and inputting symbols decreases speed. Validated models of password input on mobile devices can aid designers in usability testing new password policies. The results have implications for both usability and security of password input on mobile devices.
  • Thumbnail Image
    Item
    System Usability and User Mental Models of Three Verifiable, End-to-end Voting Methods: Helios, Prêt à Voter, and Scantegrity II
    (2014-07-11) Acemyan, Claudia Ziegler; Kortum, Philip T; Byrne, Michael D; Lane, David; Wallach, Dan S.
    There are many ways voting systems can be maliciously attacked so that election outcomes are altered. In response, voting security experts developed end-to-end (e2e), verifiable voting methods. These systems were intended to be secure, accurate, reliable, and transparent, while still preserving voter anonymity. What is not clear is if these complex, novel systems, which allow voters to check on their ballots after voting, will be usable by every voter. If voting methods are unusable, negative ramifications like disenfranchisement and altered election outcomes could occur. For this reason, system usability and voter mental models of e2e systems must be understood. To address this lacuna in voting research, three e2e methods representative of voter verifiable technologies were studied: Helios, Prêt à Voter, and Scantegrity II. Four studies were conducted. In the first study, baseline usability data was collected. By having participants vote with each system in a mock election, it was found that the systems were difficult, if not impossible, to use. Only 58% of voters were able to cast a ballot, and fewer were able to verify their vote. In the second study, the behavioral errors that led to ballot casting and vote verification event failures were identified, and potential contributing system design deficiencies were discussed. This study revealed that a few design details were driving most of the observed failures, of which all can be fixed. In the third study, voters’ mental models for each voting system were explored. The data supported the claim that voters did not have comprehensive mental models accounting for how the systems work; rather their models emphasized how-to-vote procedures, which were not always correct. In the fourth study it was asked if voters even wanted to use the verification systems, and if they did, what form of verification they would expect. Sixty-five percent of voters indicated that they would be interested in checking that their ballot was cast. As for the preferred form of verification, there was not a consensus—indicating that a diverse set of expectations will need to be accounted for when developing the systems. In conclusion, the tested e2e systems were not easily usable by voters, fully understood by them, or in a form that voters might have expected. Yet the system problems observed can be fixed, and voters seem to support the idea of auditable voting systems—meaning future effort should be spent improving upon the next generations of e2e systems so that all voters can use secure, accurate, transparent, and reliable voting systems.