Coding Automation in Narrative-Identity Research: Human Versus AI Coding of Self-Defining Memory Constructs

We compared pretrained large language model vs. human-generated quantitative coding for 8 narrative-identity variables in 1000 self-defining memories (100 participants). Cross-method reliability was fair-to-substantial (percent agreement; nested concordance coefficients). RV coefficients suggested comparable validity against external variables. Cost-efficiency favored A.I. Both coding methods yielded errors, whose differences are discussed.