Variance-Aware LLM Annotation for Strategy Research: Sources, Diagnostics, and a Protocol for Reliable Measurement

TL;DR

This paper identifies sources of variance in LLM-generated annotations for strategy research, demonstrating how design choices impact results and proposing a protocol for reliable, reproducible measurement using LLMs.

Contribution

It introduces a variance-aware protocol for LLM annotation, addressing instability issues and establishing standards for reliable measurement in strategy research.

Findings

Minor design choices can shift outcomes by 12-85 percentage points

Variance sources threaten reproducibility and bias econometric estimates

Proposes a protocol with sampling, aggregation, and reporting standards

Abstract

Large language models (LLMs) offer strategy researchers powerful tools for annotating text at scale, but treating LLM-generated labels as deterministic overlooks substantial instability. Grounded in content analysis and generalizability theory, we diagnose five variance sources: construct specification, interface effects, model preferences, output extraction, and system-level aggregation. Empirical demonstrations show that minor design choices-prompt phrasing, model selection-can shift outcomes by 12-85 percentage points. Such variance threatens not only reproducibility but econometric identification: annotation errors correlated with covariates bias parameter estimates regardless of average accuracy. We develop a variance-aware protocol specifying sampling budgets, aggregation rules, and reporting standards, and delineate scope conditions where LLM annotation should not be used. These contributions transform LLM-based annotation from ad hoc practice into auditable measurement infrastructure.