"Visiting scientist effect"? Exploring the impact of time-lags in the digitization of 2D landmark data.

Abstract

Measurement error (ME) in geometric morphometrics has been the subject of countless articles, but none specific to the effect of time lags on landmark digitization error. Yet, especially for visiting scientists working on museum collections, it is not uncommon to collect data in multiple rounds, with interruptions of weeks or years. To explore the impact of time lags on Procrustes shape analysis, I repeatedly digitized the same landmarks, on photographs of crania of adult yellow-bellied marmots, at progressively longer time intervals, ranging from a few hours to days, weeks and, in one case, many years. Using a battery of methods, I found that there is indeed a time-related systematic ME suggesting the possibility of a "visiting scientist effect" biasing shape patterns. However, the relationship between time lags and the magnitude of the bias is not simple and linear, but complex. Interestingly, the impact of the bias on the results of tests of sexual dimorphism and allometry is modest, and mostly negligible, unless the design of the data collection is highly unbalanced. When this happens, as in a simulated case where females are digitized first and males only later (or vice versa), the effect of the bias on tests of biological variation becomes important and can even lead to opposite conclusions on group differences. I will discuss when systematic ME in landmark data is more problematic and how to try to mitigate the impact of a potential "visiting scientist effect" on shape analyses.