UNCERTAINTY ASSESSMENT IN PREDICTIVE MICROBIOLOGY. PART 1: COMPARISON OF STATISTICAL METHODS
Predictive microbiology, which combines microbiological knowledge and mathematical techniques in order to develop models for the prediction of microbial evolution in foods, has become a useful tool to complement laborious analyses to verify the microbiological safety of food products. Reliable application of the models requires proper understanding of model uncertainty. In order to have a quantitative measure for the uncertainty on model parameters (obtained by using common Least Sum of Squared Errors (LSSE)) and model predictions, different statistical methods can be implemented (i.e., Asymptotic Standard Errors, Joint Confidence Regions and Monte Carlo analysis). The LSSE-criterion implicitly assumes that (i) the model is (close to) linear in the model parameters, (ii) the independent variable (time t in the case study) is exactly measured and (iii) the experimental noise is characterised by an additive Gaussian distribution with zero-mean and constant variance. The underlying assumptions, properties and results of the 3 previously mentioned statistical methods are thoroughly investigated. Comparison is based on a data set of Escherichia coli K12 at constant temperature, which is described by the non-linear Baranyi and Roberts growth model. The uncertainty on model parameters as well as the uncertainty on model predictions is assessed.
Poschet, F., Bernaerts, K., Geeraerd, A.H. and Van Impe, J.F. (2001). UNCERTAINTY ASSESSMENT IN PREDICTIVE MICROBIOLOGY. PART 1: COMPARISON OF STATISTICAL METHODS. Acta Hortic. 566, 351-356
Asymptotic Standard Errors, Joint Confidence Regions, Monte Carlo analysis, Parameter estimation, Baranyi and Roberts growth model