Assessing Grading Bias from Artificial Intelligence Disclosure at UPLB: An Experimental Study

Date

5-2026

Degree

Bachelor of Science in Economics

College

College of Economics and Management (CEM)

Adviser/Committee Chair

Richard B. Daite

Committee Member

Jefferson A. Arapoc, Jaimie Kim B. Arias, Maria Angeles O. Catelo

Request Access

For non-UP researchers, requests for access to this material may be directed to the CEM Library at [email protected] or to the UKDR administrator at [email protected]

Abstract

This study examines whether an artificial intelligence (AI) disclosure label is associated with differences in academic essay evaluation. As generative AI becomes more common in education, disclosure is increasingly discussed as a transparency measure, yet its grading effects remain unclear. The study used a 2 × 2 within-rater factorial vignette experiment in which undergraduate BS Economics students evaluated essays varying by quality and AI disclosure status. Four essay stimuli were developed across two economics topics. Two pilot waves validated the manipulation, and a faculty benchmark exercise constructed a calibration control. Student respondents were recruited from eligible BS Economics undergraduates at] UPLB through a college-wide email invitation, with participation based on voluntary self- selection. The final dataset consisted of 408 essay-level observations from 102 respondents. OLS regression with respondent-clustered standard errors showed that high- quality essays received higher grades than low-quality essays (β = 13.55, p < 0.001). The AI disclosure label was associated with lower grades conditionally: modestly in the low- quality condition (β = -3.58, p = 0.064) and more strongly in the high-quality condition, where the penalty reached 9.27 points. The findings suggest label-induced differential devaluation. AI disclosure did not merely lower evaluation uniformly; it reduced the grade advantage normally given to high-quality essays. This suggests that strong writing may be discounted when framed as AI-assisted, possibly because evaluators attribute part of its quality to the tool rather than to the student.

Language

English

Location

UPLB College of Economics and Management (CEM)

Call Number

LG 993.5 2026 E2 L67

Notes

Viewing access to electronic resources is restricted solely to UP Gmail accounts. Any access and share requests from external organizations and personal email accounts will be promptly declined.

Award: Best Undegraduate Thesis in Economics

Document Type

Thesis

Link to Full Text

Share

COinS