Survival analysis approach to insect life table analysis and hypothesis testing: with particular reference to Russian wheat aphid (Diuraphis noxia (Mordvilko)) populations

Abstract

The goal of this paper is to examine and demonstrate that survival analysis, which has been a de facto standard in biomedical research since the 1990s but has not been widely adopted in entomology yet, should possess similar potential in entomological research. The following three objectives are set to achieve this goal: (i) addressing a fundamental issue - censoring or incomplete observations; (ii) demonstrating the application of survival analysis to analyze insect life tables; and (iii) applying survival analysis for hypothesis testing. The data used to demonstrate the applications is from our laboratory experiments, which recorded the development, survival and reproduction of 1800 Russian wheat aphids (Diuraphis noxia (Mordvilko), RWA) under 25 treatments of temperature and plant-growth stage. With regard to the first two objectives, besides examining the near ubiquitous existence of censoring in insect population research, we constructed and analyzed life tables of 1800 RWA individuals with survival analysis. We further demonstrate that there could be very significant differences in life table parameters, such as median development times with and without considering censoring. To the best of our knowledge, this is the first recognition in entomology that censoring, which is hardly avoidable, can cause significant systematical bias (ranging between 4-25%; table 1) in insect development data analysis. As for the third objective, the study shows that four statistics from survival analysis can be applied to testing the effects of covariates, such as temperature and plant-growth stage, on development and survival of the Russian wheat aphid. The advantages of survival analysis include the handling of censored observations, survival probabilities in the form of rigorous survivor function vs. simple survival rates, dynamic modeling of covariates effects on development and survival with a unified model structure, etc. The methods demonstrated in this article should also be useful for entomological research beyond insect demography, such as bioassay, assessment of natural enemies, the studies of insect behaviors, etc.