Probit Analysis Models Used for Fitting Response Data

3 Probit Analysis Models Used for Fitting Response Data 

One of the following generie probit models (with or without some slight modifieation thereof) was used for fitting experimental response data via one-faetor MLE/probit analysis/multifactor probit analysis/etc. (see Table 9 2).36,38,48,55 binary or ordinal response (with a probit-link function) 

When fitting the equations in Table 9.2, all variability in the data will contribute to the estimate for PSs [kc, k and A n,), be it from variance due to individual susceptibilities (the true goal), batch effects, experimental error, differences in durations, compilation from many sources, etc. The heterogeneity introduced by outside sources of variance will artificially lower the PS, which is why combining response data from multiple independent studies is not recommended (as it introduces batch effects into the analysis). The precision of slope estimation typically improves as more experimental subjects are used. 

For studies where both C and t were varied, t was always treated as a factor [eqn (9.9) and (9.11)] rather than as a covariate [eqn (9.12)] in order to produce the best overall PS and median effective C or t estimates. This was because any curvature in the log(C) versus log(t) relationship will artificially lower the PS (insofar as the slope is a measure of the response variability among individuals exposed to a toxicant). Eqn (9.12) was used only for calculating the TLE. 

One-factor MLE [kc or fcp kept constant) or traditional probit analysis (nothing kept 

---