Measuring mimicry: methods for quantifying visual similarity

Abstract

The existence of substantial variation in the accuracy of resemblance of mimics to their models remains a fundamental evolutionary question, and the ability to quantify mimetic accuracy is key to addressing this question. Mimicry researchers have used a wide variety of methods to measure accuracy, which can be broadly grouped into (1) receiver responses, such as predator trials, and (2) trait measurements. While receiver responses are fundamental to mimicry research, logistical and ethical concerns mean that only a minority of studies include tests with relevant predator receivers. Trait measurement methods are therefore vital, both to supplement receiver response methods and as standalone methods for quantifying accuracy. Here, we collect and describe five methods for measuring the accuracy of visual mimicry. Three of the methods quantify the similarity of mimetic phenotypic traits: linear morphometrics, geometric morphometrics and biometric trait table. Two methods measure the responses of proxy receivers: machine learning and human assessment. These methods are applicable to any type of visual mimetic traits. We evaluate and compare the effectiveness and cost of these methods by assessing the accuracy of a variety of visual ant mimics to their ant models. The methods vary in the types and number of traits being measured, whether traits are measured directly (trait measurement methods) or indirectly (receiver responses) and in the effort and skills required for implementation. Despite the methods measuring different visual characteristics, the accuracy scores from all five methods were positively correlated when applied to ant mimics, suggesting some generality in how mimics rank in their accuracy. We provide practical advice on the use of these five methods and provide open-source implementations of the scripts used in our analysis. The scripts may be freely reused or used as the basis for new implementations. Our aim is to aid researchers to select and implement methods that are appropriate and suitable for future mimicry research.