Application of a statistical bootstrapping technique to calculate growth rate variance for modelling psychrotrophic pathogen growth.

The inherent variability or 'variance' of growth rate measurements is critical to the development of accurate predictive models in food microbiology. A large number of measurements are typically needed to estimate variance. To make these measurements requires a significant investment of time and effort. If a single growth rate determination is based on a series of independent measurements, then a statistical bootstrapping technique can be used to simulate multiple growth rate measurements from a single set of experiments. Growth rate variances were calculated for three large datasets (Listeria monocytogenes, Listeria innocua, and Yersinia enterocolitica) from our laboratory using this technique. This analysis revealed that the population of growth rate measurements at any given condition are not normally distributed, but instead follow a distribution that is between normal and Poisson. The relationship between growth rate and temperature was modeled by response surface models using generalized linear regression. It was found that the assumed distribution (i.e. normal, Poisson, gamma or inverse normal) of the growth rates influenced the prediction of each of the models used. This research demonstrates the importance of variance and assumptions about the statistical distribution of growth rates on the results of predictive microbiological models.