IAT LLM Associative Interference

This dataset contains trial-level responses from large language models evaluated using an adapted Implicit Association Test (IAT). Each row corresponds to a single forced-choice classification trial. Variables include the IAT domain (iat_type), experimental condition (block), stimulus item (item), response options (pairing_A, pairing_B), model identity, and raw model output (choice). Derived variables include valid_response (whether a valid A/B response was produced), choice_clean (parsed response), and task_consistent (whether the response matches the predefined mapping). Analyses are conducted in two stages: modeling response compliance and conditional task consistency to estimate associative interference.