Stratified sampling for content analysis: a python-based solution

Manual stratified sampling in content analysis is time-intensive and difficult to reproduce. While stratified sampling produces more efficient estimates than simple random sampling, implementation remains predominantly manual, creating reproducibility challenges and substantial researcher burdens. This study develops and validates an automated stratified sampling tool using Python and configuration-driven design. The tool enables researchers to specify sampling parameters through a configuration file without modifying code, supporting adaptation across diverse content analysis contexts. Validation using a set of 1,888 news articles about U.S. Border Patrol coverage (2012-2025) stratified by outlet political bias and presidential administration demonstrates two key findings. First, reproducibility testing across 100 independent runs with identical seeds shows 100% agreement in item selection, confirming exact replicability. Second, implementation verification reveals zero duplicate items, correct stratum allocation, and valid temporal boundaries, demonstrating the elimination of common manual sampling errors. The tool is released as open-source software, extending computational methods to the sampling phase of content analysis research and addressing reproducibility challenges in communication and social science.