Robustness

Change the Number, Change the Answer: Assessing Rating Scale Sensitivity in LLM-Based Text Scoring

A systematic study of how rating scale design (format, range, and sign) shifts LLM-generated scores and their interpretation in computational social science text scoring