Performance of the interaction models was evaluated by lack-of- fit F-tests as suggested by Zwietering et al. (1990). The residual sum of squares (RSS) of growth curves simulated by the interaction models was compared to RSS of the growth curves fitted individ- ually to the log-transformed Logistic model with delay for temperatures at 11.9 C and above and to the log-linear model for 9.4 and 10.4 C, representing the pure error. In addition, and inspired by the acceptable prediction zone method suggested by Oscar (2005), an acceptable simulation zone was defined and applied to evaluate simulation performance. The acceptable simu- lation zone (ASZ) was defined as `0.5 log10-units from the simu- lated Salmonella spp. and natural microbiota counts, as also used in the study of Velugoti et al. (2011) for evaluation of predictions during dynamic temperatures. The simulations were considered acceptable when at least 70% of the observed values were inside the acceptable simulation zone (Oscar, 2005).
2.4. Curve fittings and simulations
Curve fittings and simulations were done in five steps. In the first step, individual bacterial growth curves of Salmonella spp. in sterile ground pork were fitted to the log-transformed Logistic with delay model by minimizing the RSS using the solver function in Microsoft Excel. In the second step, the PROC NLIN in SAS Enter- prise was applied for fitting the secondary models, Equations (1) and (2), describing the effect of temperature on growth parame- ters for Salmonella spp. in sterile ground pork and for the natural pork microbiota, respectively. In the third step, the parameter estimates from Equations (1) and (2) were used in the Equations (4), and (5) for the estimation of the coefficients of interaction g and aS,NB, respectively. This was done for each of the challenge tests performed in ground pork with a natural microbiota. The obtained growth curve of Salmonella was fitted to Equations (4), and (5) by an iteration procedure where g and aS,NB, respectively, was changed until the minimum RSS was found. The solver function in Microsoft Excel was used for this purpose. In the fourth step, the non-linear SAS computation procedure was applied for fitting the secondary
2.3.4. Evaluation of interaction models