Roman M. Sheremeta
Current Organizations / Institutes
The organizations/institutes from the scholar's latest article
Previous Organizations / Institutes
The organizations / institutes from the scholar's other articles in the past ten years
Research Subjects
Roman M. Sheremeta has a total of 64 co-authors. Here are the top co-authors by number of shared publications:
- Timothy W. Shields
Chapman University - George L. Argyros School of Business & Economics; Chapman University - Economic Science Institute
9 shared publications - Eric Schniter
Chapman University
6 shared publications - Erik O. Kimbrough
Chapman University
6 shared publications Shakun D. Mago
University of Richmond
6 shared publications- Subhasish M. Chowdhury
University of Bath - Department of Economics
6 shared publications - Timothy N. Cason
Department of Economics, Purdue University, West Lafayette, USA
5 shared publications
Latest Publications
- Open Access
- 11 October 2026
- Published by Elsevier in SSRN Electronic Journal
A Turing Test for Grading in Economics: Detection, Grades, and Student Satisfaction in a Blinded Field Experiment
Writing assignments are valuable in economics courses but costly to grade, and large language models can now grade written work and draft feedback cheaply. In a semester-long blinded field experiment in two undergraduate economics courses, 37 students completed 24 short written assignments (846 submissions). Three AI graders and three human teaching assistants independently graded every submission on the same five-dimension rubric. Students saw one randomly selected grade and its feedback, without being told its source, then rated their satisfaction and guessed the grader type. AI feedback was easy to tell apart from human feedback: it was more than twice as long and hedged more than seven times as often, and a classifier identified its source with 99.4% accuracy. Students could not: they named the grader type correctly in 52.1% of cases, which is chance (p = 0.229), and did not improve over the semester. The human graders graded 0.68 points higher on a five-point scale, a difference in grading standards whose direction the design does not identify. Human-graded submissions received higher satisfaction ratings, but holding the grade fixed cut the coefficient on the actual grader to nearly zero, and a higher grade raised satisfaction equally from either grader type. Students who believed a human had graded them reported higher satisfaction whichever grader actually had. Instructors can close the grade gap by calibrating grading standards; because satisfaction is associated with the believed rather than the actual grader, disclosure may matter through beliefs.
- 11 October 2026
- Published by Elsevier in SSRN Electronic Journal
Open AccessA Turing Test for Grading in Economics: Detection, Grades, and Student Satisfaction in a Blinded Field Experiment
Writing assignments are valuable in economics courses but costly to grade, and large language models can now grade written work and draft feedback cheaply. In a semester-long blinded field experiment in two undergraduate economics courses, 37 students completed 24 short written assignments (846 submissions). Three AI graders and three human teaching assistants independently graded every submission on the same five-dimension rubric. Students saw one randomly selected grade and its feedback, without being told its source, then rated their satisfaction and guessed the grader type. AI feedback was easy to tell apart from human feedback: it was more than twice as long and hedged more than seven times as often, and a classifier identified its source with 99.4% accuracy. Students could not: they named the grader type correctly in 52.1% of cases, which is chance (p = 0.229), and did not improve over the semester. The human graders graded 0.68 points higher on a five-point scale, a difference in grading standards whose direction the design does not identify. Human-graded submissions received higher satisfaction ratings, but holding the grade fixed cut the coefficient on the actual grader to nearly zero, and a higher grade raised satisfaction equally from either grader type. Students who believed a human had graded them reported higher satisfaction whichever grader actually had. Instructors can close the grade gap by calibrating grading standards; because satisfaction is associated with the believed rather than the actual grader, disclosure may matter through beliefs.
- Open Access
- 10 April 2026
- Published by Elsevier in SSRN Electronic Journal
A Turing Test for Academic Evaluation: Detection, Grades, and Student Satisfaction in a Field Experiment
We study whether AI evaluation can pass a behavioral Turing test in a consequential academic setting. In a semester-long field experiment, undergraduates in two economics courses complete multiple essay assignments, each randomly graded by one of three AI systems or one of three human teaching assistants, with students blinded to grader identity. After receiving their grade and written feedback, students report satisfaction and guess whether their grader was human or AI. We find that students cannot identify their grader type above chance (52.1% accuracy; p = 0.229 using a two-sided binomial test), and detection failure is uninformative: neither the grade received nor any observable feature of the evaluation predicts students’ guesses. A naive comparison shows that human-graded assignments produce higher satisfaction, but this gap decomposes entirely into two channels: the grade received and students’ beliefs about who graded them. Actual grader identity has no independent effect on satisfaction once these channels are controlled, and the grade–satisfaction slope is identical across grader types. These results establish behavioral substitutability under source uncertainty: AI evaluation is not merely undetected—it is functionally equivalent to human evaluation in the eyes of those being evaluated. The apparent satisfaction cost of AI grading reflects grade differences and attribution, not grader identity itself.
- 10 April 2026
- Published by Elsevier in SSRN Electronic Journal
Open AccessA Turing Test for Academic Evaluation: Detection, Grades, and Student Satisfaction in a Field Experiment
We study whether AI evaluation can pass a behavioral Turing test in a consequential academic setting. In a semester-long field experiment, undergraduates in two economics courses complete multiple essay assignments, each randomly graded by one of three AI systems or one of three human teaching assistants, with students blinded to grader identity. After receiving their grade and written feedback, students report satisfaction and guess whether their grader was human or AI. We find that students cannot identify their grader type above chance (52.1% accuracy; p = 0.229 using a two-sided binomial test), and detection failure is uninformative: neither the grade received nor any observable feature of the evaluation predicts students’ guesses. A naive comparison shows that human-graded assignments produce higher satisfaction, but this gap decomposes entirely into two channels: the grade received and students’ beliefs about who graded them. Actual grader identity has no independent effect on satisfaction once these channels are controlled, and the grade–satisfaction slope is identical across grader types. These results establish behavioral substitutability under source uncertainty: AI evaluation is not merely undetected—it is functionally equivalent to human evaluation in the eyes of those being evaluated. The apparent satisfaction cost of AI grading reflects grade differences and attribution, not grader identity itself.
- Open Access
- 1 January 2026
- Published by Elsevier in SSRN Electronic Journal
AI Grading, Grade Inflation, and Student Satisfaction
This paper studies whether artificial intelligence (AI) grading can reduce grading costs and grade inflation without generating independent student dissatisfaction. Using a randomized controlled trial in which identical undergraduate assignments are graded by either human graders or an AI system, with students blinded to grader identity, we document three main results. First, AI graders assign significantly lower grades than human graders to the same work. Second, students are unable to detect whether grading is performed by AI or humans. Third, conditional on the grade received, student satisfaction does not differ by grader type. These findings are robust across specifications with student fixed effects, assignment controls, and interaction tests. The results suggest that AI grading changes grading outcomes by altering evaluators’ incentive constraints rather than through differences in pedagogy or student perceptions.
- 1 January 2026
- Published by Elsevier in SSRN Electronic Journal
Open AccessAI Grading, Grade Inflation, and Student Satisfaction
This paper studies whether artificial intelligence (AI) grading can reduce grading costs and grade inflation without generating independent student dissatisfaction. Using a randomized controlled trial in which identical undergraduate assignments are graded by either human graders or an AI system, with students blinded to grader identity, we document three main results. First, AI graders assign significantly lower grades than human graders to the same work. Second, students are unable to detect whether grading is performed by AI or humans. Third, conditional on the grade received, student satisfaction does not differ by grader type. These findings are robust across specifications with student fixed effects, assignment controls, and interaction tests. The results suggest that AI grading changes grading outcomes by altering evaluators’ incentive constraints rather than through differences in pedagogy or student perceptions.
- Open Access
- 14 September 2023
- Published by Elsevier in Journal of Economic Behavior & Organization
- Vol. 215, 60-73
Conflict in the pool: A field experiment
- 14 September 2023
- Published by Elsevier in Journal of Economic Behavior & Organization
- Vol. 215, 60-73
Open AccessConflict in the pool: A field experiment
- Open Access
- 21 July 2023
- Published by Institute of Higher Education NAES of Ukraine in Filosofiya osvity. Philosophy of Education
- Vol. 29 (1) , 324-330
American University Kyiv: American standards of education in Ukraine. To the official opening of the campus in Kyiv
An overview of a landmark event in the space of higher education of Ukraine – the grand opening of the American University Kyiv (AUK) campus on May 19, 2023 in Kyiv – was carried out. The goals of creation, the mission of the university, its main tasks and advantages are presented. The main ideas of the speeches of the key speakers of the event, including the founder-rector of the AUK Roman Sheremeta and the new president of the AUK Dan Rice, are provided.
- 21 July 2023
- Published by Institute of Higher Education NAES of Ukraine in Filosofiya osvity. Philosophy of Education
- Vol. 29 (1) , 324-330
Open AccessAmerican University Kyiv: American standards of education in Ukraine. To the official opening of the campus in Kyiv
An overview of a landmark event in the space of higher education of Ukraine – the grand opening of the American University Kyiv (AUK) campus on May 19, 2023 in Kyiv – was carried out. The goals of creation, the mission of the university, its main tasks and advantages are presented. The main ideas of the speeches of the key speakers of the event, including the founder-rector of the AUK Roman Sheremeta and the new president of the AUK Dan Rice, are provided.
- Open Access
- 30 November 2022
- Published by Elsevier in SSRN Electronic Journal
Nudging Civilian Evacuation During War: Evidence from Ukraine
In times of war, evacuating civilians from conflict zones is of critical importance for their survival and well-being. However, many people are hesitant to evacuate. Text-based nudges are a promising, yet unexplored, venue to increase the willingness to evacuate. We conduct a controlled survey experiment in Ukraine, manipulating the framing of automated alert messages. Our findings suggest that providing individuals with an evacuation plan by the authorities is crucial. The specific framing of the message itself does not seem to play a role in the perceived effectiveness of the messages. Heterogeneity analysis shows that women respond more strongly to information about a provided evacuation plan. Furthermore, having a pre-existing personal evacuation plan and being provided with one by the authorities act as substitutes.
- 30 November 2022
- Published by Elsevier in SSRN Electronic Journal
Open AccessNudging Civilian Evacuation During War: Evidence from Ukraine
In times of war, evacuating civilians from conflict zones is of critical importance for their survival and well-being. However, many people are hesitant to evacuate. Text-based nudges are a promising, yet unexplored, venue to increase the willingness to evacuate. We conduct a controlled survey experiment in Ukraine, manipulating the framing of automated alert messages. Our findings suggest that providing individuals with an evacuation plan by the authorities is crucial. The specific framing of the message itself does not seem to play a role in the perceived effectiveness of the messages. Heterogeneity analysis shows that women respond more strongly to information about a provided evacuation plan. Furthermore, having a pre-existing personal evacuation plan and being provided with one by the authorities act as substitutes.
- 1 July 2020
- Published by Elsevier in Journal of Economic Behavior & Organization
- Vol. 175, 314-327
Winner-take-all and proportional-prize contests: Theory and experimental results
- 1 July 2020
- Published by Elsevier in Journal of Economic Behavior & Organization
- Vol. 175, 314-327
Winner-take-all and proportional-prize contests: Theory and experimental results
- 1 December 2019
- Published by Elsevier in European Journal of Political Economy
- Vol. 60, 101778
The tug-of-war in the laboratory
- 1 December 2019
- Published by Elsevier in European Journal of Political Economy
- Vol. 60, 101778
The tug-of-war in the laboratory
- 1 October 2019
- Published by Elsevier in Journal of Economic Behavior & Organization
- Vol. 166, 1-11
Adding tournament to tournament: Combining between-team and within-team incentives
- 1 October 2019
- Published by Elsevier in Journal of Economic Behavior & Organization
- Vol. 166, 1-11
Adding tournament to tournament: Combining between-team and within-team incentives
- 13 August 2019
- Published by Cambridge University Press (CUP) in Behavioral and Brain Sciences
- Vol. 42, e140
The attack and defense games
The attack-and-defense game is a game in which an attacker (a group of attackers) has an incentive to revise the status quo and a defender (a group of defenders) wants to protect it. The asymmetry in objectives creates incompatible interests and results in a mixed-strategy Nash equilibrium. However, this equilibrium could be heavily impacted by behavioral considerations.
- 13 August 2019
- Published by Cambridge University Press (CUP) in Behavioral and Brain Sciences
- Vol. 42, e140
The attack and defense games
The attack-and-defense game is a game in which an attacker (a group of attackers) has an incentive to revise the status quo and a defender (a group of defenders) wants to protect it. The asymmetry in objectives creates incompatible interests and results in a mixed-strategy Nash equilibrium. However, this equilibrium could be heavily impacted by behavioral considerations.