Structured to Fail: Gender Bias in Large Language Models Across Text and Visual Modalities Through Data Feminism and Intersectionality in the Indian Context

This repository contains the dataset associated with the study "Structured to Fail: Gender Bias in Large Language Models Across Text and Visual Modalities Through Data Feminism and Intersectionality in the Indian Context". The deposit includes a representative subset of the full study data, comprising the following components: Sample Design: Documentation of the sampling framework and run structure across the three LLM platforms examined Prompt Battery: The complete set of prompts administered across text and image generation tasks Model-Generated Text Outputs: Textual responses generated (1,020) by the models under the study Model-Generated Image Outputs: Visual outputs generated (480) in response through image-generation prompts Codebook: Provides complete instructions for all independent coders participating in the inter-rater reliability (IRR) study Inter-rater Reliability Dataset: The double-coded subset used to establish coding agreement Reported Reliability Scores: Cohen's Kappa values and percentage agreement statistics as reported in the manuscript