Reliability of Sampling Methods’ Representativeness in Decision-Making on the Use of OCR in Archives
Abstract This study examines whether commonly used sampling methods provide reliable and representative evidence for selecting optimal optical character recognition (OCR) configurations in archival digitization projects. Because OCR performance directly affects the informational value, authenticity, and long-term usability of digital records, reliable sampling is essential for evidence-based decision-making. The objective was to evaluate the influence of simple random, systematic random, and stratified random sampling on OCR accuracy and to determine whether sampling strategy affects the selection of OCR systems and digitization settings. A dataset of 3,634 Croatian-language archival pages (approximately 1.45 million words) was digitized and evaluated using three OCR systems. The results showed that different sampling methods produced substantially different OCR rankings and error rates, and that the effects of sampling method, OCR system, and their interaction were all statistically significant (ANOVA, p ≤ 0.05). These findings demonstrate that sampling strategy can influence OCR performance and may introduce bias into technology selection. Consequently, sampling should not be used as the sole basis for configuring large-scale digitization workflows but rather as one component of a broader, evidence-based evaluation framework, with implications extending to AI and machine learning applications that depend on representative training data.
Authors
- Željko Trbušić (ORCID: https://orcid.org/0000-0002-1555-7130)
- Hrvoje Stančić (ORCID: https://orcid.org/0000-0001-6925-4402)
Institutions
- University of Zagreb (HR)
Publication Details
- Journal
- Preservation Digital Technology & Culture
- Published
- 2026-11-14
- DOI
- https://doi.org/10.1515/pdtc-2026-0067
- Primary Topic
- Digital and Traditional Archives Management
- Type
- article
- Field-Weighted Citation Impact
- 0.00