The study analyzes 600,546 digitized newspaper pages from California and New York to measure suffrage coverage from 1890 to 1920. Researchers used 6 phrases to identify pages discussing women's suffrage, finding 14,971 matches in a strict query and 28,283 in a broad query. The analysis reveals that raw counts can be misleading due to archive size fluctuations. For instance, New York had the most matching pages in 1913 (1,227), but the normalized rate peaked in 1909. Similarly, California's 1914 data shows only 96 matching pages, but the archive contained only 2,248 pages that year, making 1914 appear high when normalized. The dataset comes from the Library of Congress' Chronicling America database and highlights the importance of normalization when studying historical archives where composition and quality change over time. The findings suggest that trends in suffrage coverage are influenced by both the phenomenon and the recording process, offering a methodological framework applicable beyond historical newspaper studies.
Normalization is crucial for accurate trend analysis in historical datasets. The raw data often conflates archive size with actual coverage rates. By normalizing for archive size, researchers can distinguish between genuine shifts in public attention and artifacts of changing dataset composition. This approach provides a more robust understanding of historical phenomena, demonstrating that apparent peaks in coverage may simply reflect increased digitization efforts rather than heightened social interest. The study serves as a case study for data scientists working with longitudinal historical archives.
Source: Towards Data Science · Summarized by HeadlinesBriefing