Enhancing Statistical Power in Animal Behavior Research: A Case Study on Enrichment in Dairy Cows Using Data Augmentation Methods

Authors: Marjorie Cellier and Thomas Gisiger and Abdoulaye Baniré Diallo and Elsa Vasseur

Date: 2025-05-01

Status: Published

DOI: 10.1038/s41598-025-89891-4

View External Publication

Research on animals often faces a critical limitation: small sample sizes limit statistical power, making it challenging to detect behavioral differences. Artificial data augmentation offers a promising solution by increasing dataset size. This study explores the potential of data augmentation in behavioral research through a case study on dairy cow enrichment in tie-stall housing.
Ten Holstein dairy cows were tested using a rossover design over two weeks, with a four-day washout
period. Two enrichment types were evaluated: a standard Kong and a modified version with additional visual (white tape) and tactile (chains) elements. Each device was suspended in the stall. Behavioral interactions with the enrichment were recorded using GoPro cameras for 45-minute sessions and annotated using The Observer XT software. A linear mixed model was applied, with interaction type, day, and treatment as fixed effects, and individuals nested within day as a random effect. Preliminary results showed no significant difference in the percentage of time spent interacting with the two Kong types (standard: 6.9±5.3% of their time; modified: 6.6±6.6% of their time