Comparing Human and AI Rater Effects Using the Many-Facet Rasch Model

TL;DR

This study evaluates the reliability and rater effects of ten large language models versus human raters in scoring writing tasks, finding some models perform comparably or better in accuracy and consistency.

Contribution

It provides a comparative analysis of multiple LLMs and human raters using the Many-Facet Rasch model, highlighting models with minimal rater effects and high scoring reliability.

Findings

ChatGPT 4o, Gemini 1.5 Pro, and Claude 3.5 Sonnet perform best.

Some LLMs show comparable accuracy to human raters.

Certain models exhibit less rater bias and higher reliability.

Abstract

Large language models (LLMs) have been widely explored for automated scoring in low-stakes assessment to facilitate learning and instruction. Empirical evidence related to which LLM produces the most reliable scores and induces least rater effects needs to be collected before the use of LLMs for automated scoring in practice. This study compared ten LLMs (ChatGPT 3.5, ChatGPT 4, ChatGPT 4o, OpenAI o1, Claude 3.5 Sonnet, Gemini 1.5, Gemini 1.5 Pro, Gemini 2.0, as well as DeepSeek V3, and DeepSeek R1) with human expert raters in scoring two types of writing tasks. The accuracy of the holistic and analytic scores from LLMs compared with human raters was evaluated in terms of Quadratic Weighted Kappa. Intra-rater consistency across prompts was compared in terms of Cronbach Alpha. Rater effects of LLMs were evaluated and compared with human raters using the Many-Facet Rasch model. The results in general supported the use of ChatGPT 4o, Gemini 1.5 Pro, and Claude 3.5 Sonnet with high scoring accuracy, better rater reliability, and less rater effects.