Estimating stage durations from samples of cohorts

Background/Question/Methods

Many organisms develop through a series of clearly distinguishable life stages. Estimating the time individuals spend within each of these stages is important for biological understanding of population dynamics and in applications. Many animals with stage-structured development are small, and following individuals across their life cycle is impractical or impossible. Instead, stage structured data may be collected by following a cohort and sampling or counting the number of individuals in each stage over time. Current methods for stage-structured data employ several restrictive assumptions, such as assuming no among individual variation in development or development following a distribution with a known shape. We formulate a more general model for stage-structured cohort development and estimate it using Monte Carlo methods.

Results/Conclusions

For situations where stage count data are independent across time, we show that Monte Carlo approximation of stage probabilities combined with MCMC is a flexible, albeit computationally demanding, approach to parameter estimation. This type of data may arise when stage frequencies are independent samples from a large population or under destructive sampling schemes. We illustrate the approach by analyzing classical stage frequency data on grasshoppers and spittlebugs sampled in the field. A different data collection method that appears in laboratory studies is to repeatedly census a cohort over time, leading to stage count data that are dependent. Analyzing this type of data rigorously is challenging and we develop an MCMC algorithm specifically targeted at this problem. The algorithm is applied to replicated data on spider mite cohorts for illustration.