Forensic analysis of dominant versus non-dominant handwriting: Statistical and machine learning insights

Systematic, paired quantitative evidence on how handwriting changes when the non-dominant hand is used remains limited in forensic document examination, despite its frequent relevance in cases involving disguise and authorship concealment. This study presents the first large-scale within-writer analysis examining general and individual handwriting characteristics using statistical testing and predictive modeling. Handwriting samples were collected from 94 right-handed participants, each of whom produced the same standardized text with both the dominant and non-dominant hand. Thirteen general and 19 individual handwriting characteristics were evaluated following established forensic criteria. Statistically significant differences were identified in 61.5% of general and 57.9% of individual handwriting characteristics. General features most affected by non-dominant writing included writing speed, graphic maturity, slant stability, letter connectivity, neatness, and legibility, whereas many structural letter forms exhibited high concordance across hands. Individual differences were primarily observed in fine motor features such as loop formation and diacritic construction, while core letter styles remained largely stable, supporting the persistence of handwriting individuality despite altered hand use. To assess the discriminative value of handwriting characteristics, a supervised logistic regression model incorporating cross-validation and feature selection was applied. A model using 10 handwriting characteristics achieved a test accuracy of 92.98%, while a reduced model retained high accuracy (91.23%), indicating that execution-quality features carry the strongest diagnostic value for distinguishing hand use. By establishing paired quantitative benchmarks for stable and degradable handwriting characteristics, this study provides practical guidance for evaluating suspected off-hand writing and demonstrates how statistical modeling can support expert forensic judgment.