Is shape in the eye of the beholder? Assessing landmarking error in geometric morphometric analyses on live fish

• Published in: PeerJ
• Main Authors: Paolo Moccetti, Jessica R. Rodger, Jonathan D. Bolland, Phoebe Kaiser-Wilks, Rowan Smith, Andy D. Nunn, Colin E. Adams, Jen A. Bright, Hannele M. Honkanen, Angus J. Lothian, Matthew Newton, Domino A. Joyce
• Format: Article in Journal/Newspaper
• Language: English
• Published: PeerJ Inc. 2023
• Subjects: Measurement error; Bias; Landmarks; Morphometrics; Replication crisis; Reproducibility; Medicine; R; Atlantic salmon; Salmo salar
• Online Access: https://doi.org/10.7717/peerj.15545; https://doaj.org/article/6152a6e8ec5f40aeb3716041724eb169

Geometric morphometrics is widely used to quantify morphological variation between biological specimens, but the fundamental influence of operator bias on data reproducibility is rarely considered, particularly in studies using photographs of live animals taken under field conditions. We examined this using four independent operators that applied an identical landmarking scheme to replicate photographs of 291 live Atlantic salmon (Salmo salar L.) from two rivers. Using repeated measures tests, we found significant inter-operator differences in mean body shape, suggesting that the operators introduced a systematic error despite following the same landmarking scheme. No significant differences were detected when the landmarking process was repeated by the same operator on a random subset of photographs. Importantly, in spite of significant operator bias, small but statistically significant morphological differences between fish from the two rivers were found consistently by all operators. Pairwise tests of angles of vectors of shape change showed that these between-river differences in body shape were analogous across operator datasets, suggesting a general reproducibility of findings obtained by geometric morphometric studies. In contrast, merging landmark data when fish from each river are digitised by different operators had a significant impact on downstream analyses, highlighting an intrinsic risk of bias. Overall, we show that, even when significant inter-operator error is introduced during digitisation, following an identical landmarking scheme can identify morphological differences between populations. This study indicates that operators digitising at least a sub-set of all data groups of interest may be an effective way of mitigating inter-operator error and potentially enabling data sharing.