Best Practices for Modeling Arthropod Lifetimes in a Bayesian Framework

Most arthropods are ectothermic, with multiple performance traits being constrained by environmental temperature. Through the effects on traits, temperature therefore constrains when, where, and how large are arthropod populations. Because many arthropods are of relevance to humans, for example as pests or disease vectors, biologists have spent substantial time trying to understand the mechanistic relationship between temperature and traits. For example, in mosquito and other vectors species, the lifespan of vectors - the number of days the adult survives - is important in determining both the size of the vector population and whether or not vectors will likely be able to transmit pathogens. Often thermal traits such as lifetime are modeled using Thermal Performance Curves (TPC) - functions that describe the relationship between temperature and traits mathematically. Many functional forms with many possible shapes have been proposed as TPCs. However, the effects of the distribution of data around these shapes and of common data transformations on how well we can infer the TCPs has been relatively ignored. In this paper we use simulated data on vector lifespans, inspired by mosquito data, to explore the reliability of inference under different assumptions about the data on our ability to accurately infer a known TPC. Using a Bayesian approach, we are also able to quantify the effects of data assumptions and transformations on uncertainty in estimates. Our results suggest that mismatches between a true and assumed distribution of data around a TPC can greatly increase uncertainty. Further, some transformations of the data before analysis are more likely to lead to biased results than others. Based on our results, we make suggestions for best practices in the analysis of arthropod thermal trait data such as lifespan and related traits.