Estimating the frequency of high microbial counts in commercial food products using various distribution functions

Industrial microbial count records usually form an irregular fluctuating time series. If the series is truly random or weakly autocorrelated, the fluctuations can be considered as the outcome of the interplay of numerous factors that promote or inhibit growth. These factors usually balance each other, although not perfectly, hence, the random fluctuations. If conditions are unchanged, then at least in principle the probability that they will produce a coherent effect, i.e., an unusually high (or low) count of a given magnitude, can be calculated from the count distribution. This theory was tested with miscellaneous industrial records (e.g., standard plate count, coliforms, yeasts) of various food products, including a dairy-based snack, frozen foods, and raw milk, using the normal, log normal, Laplace, log Laplace, Weibull, extreme value, beta, and log beta distribution functions. Comparing predicted frequencies of counts exceeding selected levels with those actually observed in fresh data assessed their efficacy. No single distribution was found to be inherently or consistently superior. It is, therefore, suggested that, when the probability of an excessive count is estimated, several distribution functions be used simultaneously and a conservative value be used as the measure of the risk.