A plot of observed versus predicted log δ values ( Fig. 5a) as well as a plot of observed versus predicted β values ( Fig. 5b) including their corresponding correlation coefficients is presented in Fig. 5. Observed versus predicted Salmonella count values for all data are presented in Fig. 6. Additionally, Table 4 shows the correlation (R), % discrepancy (% Df) and % bias (% Bf) values for predicted

versus observed time required for GSI-IX first decimal reduction (δ), shape factor values (β) and Salmonella counts in the different food products used. Data presented in Fig. 5a and the results in Table 4 (all data) indicate that the secondary model (Eq. (19)) provides a high correlation between observed versus predicted times required for first decimal reductions (R = 0.97, p < 0.001). The correlation of observed versus predicted shape factor values was not as satisfactory (R = 0.03, p = 0.915), with Eq.  (20) both over and under predicting β values ( Fig. 5b). Still,

as seen in Fig. 6 and Table 4, a significant correlation (R = 0.94, p < 0.001) of observed versus predicted CFU values was obtained when using the developed secondary models selleck kinase inhibitor to predict the survival of Salmonella in all tested food types. The degrees of discrepancy and bias found between the secondary predictive models and the data used to develop these models was found to be 16% discrepancy and − 2% bias. A negative percent bias is indicative of a tendency of the models to underestimate

survival Sitaxentan numbers (even when using the data that derived the model). This underestimation followed from the degree to which the shape parameter (in Eq.  (20)) deviated from the observed values and was more prominent at the lower CFU values. The extent to which the models underestimated the survival of Salmonella in the validation data is illustrated in Fig. 6. Data points which appear below the equivalence line are CFU values that have been underestimated and are consistent with the shape factor results in Fig. 5b. As seen in Table 4, the % bias and % accuracy factors showed a discrepancy of 41% and a bias of − 7% for all validation data collected. These discrepancy and bias values differ from those inherent to the models (16% and − 2%). However, the data collected in non-fat products including wheat flour, non-fat dry milk and whey protein powder ( Table 4) gave 12% discrepancy and − 3% bias. The bias and accuracy percentage results in non-fat food are within the error margin inherent to the models, and are an example of the consistency of the models in predicting survival data in non-fat foods. The higher discrepancy and bias percentages obtained for the whole dataset are the result of the higher discrepancy and bias percentages found for data in low-fat food products (which contain 12% fat). Table 4 shows low-fat products to have 50% discrepancy and − 9% bias.