Statistical inference for minimum inhibitory concentration data

The Minimum Inhibitory Concentration (MIC) is the smallest concentration of an anti-microbial agent that inhibits the growth of bacteria. The value is obtained in a highly mechanized fashion, but this procedure only provides interval censored reading. It is often of interest to use data collected from complex experiments to see how the mean MIC is affected by different factors. Because the MIC value is interval censored, ordinary least squares cannot be used. For models containing only fixed effects, maximum likelihood estimates (MLE) can be obtained. For models containing random effects, MLE methods are infeasible and Bayesian approaches are required. Model building, selection and diagnostic procedures are presented for selecting the appropriate model. In cases where several models seem to fit the data equally well, model averaging is also performed to get model averaged estimates. Four real data sets are analyzed using the methodology we developed.