Statistical Methods Available in CETIS

Point Estimate Methods

  • The linear regression (MLE) module offers four models, including the Log-Cumulative Normal (Probit Analysis), Log-Logit, Log-Gompertz and Log-Angle. This implementation emulates the U.S. EPA Probit software, as well as the SAS Probit procedure;
  • The linear regression module includes the Pearson chi-square heterogeneity test for fit and application of the heterogeneity factor in cases where a significant chi-square test is declared. CETIS also calculates the likelihood ratio test and an ANOVA based lack-of fit test for replicated regressions;
  • The non-linear regression module offers 41 different models including Logistic, Log-Logistic, Log-Logistic with hormesis factor (van Ewijk and Hoekstra, 1993), Log-Logistic with threshold factor (Chèvre et al., 2002), Cumulative Normal, Log-Cumulative Normal (Bruce and Versteeg, 2002), Gompertz, Log-Gompertz, Weibull, Morgan-Mercer-Flodin, Bragg, Holliday, single exponential (SFO), bi-exponential (DFOP), Hockey Stick, Gustafson-Holden (FOMC), MicroTox Gamma, and five OECD models. All models can be setup to run with the dependent or independent variable transformed, providing additional forms of the models. A binomial, beta-binomial, Poisson, negative binomial, gamma, inverse gaussian and a generalized Box-Cox weighting scheme can be applied at the user's discretion for special data types such as quantal, count or continuous data. The module includes an ANOVA based lack-of fit test for replicated test group regressions;
  • All non-linear model parameterizations were carefully selected for their desirable close-to-linear characteristics as presented by Ratkowsky (Handbook of Nonlinear Regression Models, 1990). Proprietary linearization techniques used by the nonlinear regression module typically insures rapid convergence with no additional input from the user, such as providing initial estimates of the parameters;
  • Both the linear and non-linear regression modules provide a thorough residual analysis to insure appropriate model selection. The residual analysis uses Pearson chi-square, likelihood ratio, ANOVA lack of fit, Shapiro-Wilk, Anderson-Darling, Bartlett, Brown-Forsythe Levene, Grubbs (maximum normalized residual test), and Mann-Kendall Trend auxilliary tests;
  • The linear interpolation with bootstrapping method emulates the U.S. EPA ICPIN method. The user can specify any number of resamples in multiples of 40. When desired, a log transform can be applied to the response or predictor variable;
  • The Trimmed and Untrimmed Spearman-Kärber module emulates the U.S. EPA TSK method. The Trimmed Spearman-Kärber module includes an auto-trim feature, which automatically calculates the minimum trim and requires no additional user intervention. In cases of all or none datasets, CETIS provides the binomial method.

Hypothesis Test Methods

  • A total of 41 different parametric and non-parametric hypothesis test methods including the U.S. EPA bioequivalence TST-Welch's t-Test, the OECD recommended Jonkeheere-Terpstra Step-Down test and Bonferroni-Holm adjusted Fisher Exact STP Tests;
  • Parametric hypothesis test methods include the Homoscesdastic t-Test, Heteroscedastic t, TST-Welch's t-Test, Paired Sample t, Bonferroni t, Dunn-Sidák t, Balanced Dunnett, Unbalanced Dunnett, Dunnett T3, Tamhane-Dunnett Step Down, Williams, Fisher LSD, Tukey-Kramer, Student-Newman-Kuels, Hochberg-GT2, Gabriel and Games and Howell Tests;
  • Non-parametric hypothesis test methods include Mann-Whitney U, Wilcoxon Two-Sample, Wilcoxon Rank Sum, Steel Many-One Rank, Shirley-Williams, Nemenyi-Damico-Wolfe, Dwass-Steel-Critchlow-Fligner, Hayter-Stone, Dunn Mean Rank, Cochran-Armitage Linear Trend (with and without Rao-Scott adjustment) and Fisher Exact Tests;
  • Simultaneous test procedures (STP) including Bonferroni-Holm, Bonferroni-Hommel, and Bonferroni-Hochberg adjustments can be applied to Fisher's Exact Test, Dunn's Mean Rank Test and Wilcoxon Rank Sum Test to maintain a consistent familywise or experimentwise error rate;
  • Many of the non-parametric methods yield exact probabilites, however, when exact probabilites are time prohibitive to calculate, or in the case of extensive ties, CETIS offers a Monte Carlo alternative;
  • Single classification ANOVA, randomized block ANOVA, Kruskal-Wallis, Fligner-Wolfe, Jonckheere-Terpstra Trend, Fisher-Freeman-Halton, Shapiro-Wilk W, Anderson-Darling A2, D'Agostino-Pearson K2, Bartlett, Brown-Forsythe modified Levene, Grubbs Maximum Normalized Residual test, Tarone C(α) Test for Binary Overdispersion, and Mann-Kendall Trend tests are all available as auxiliary tests;
  • The angular (arcsin square-root), log, power, Freeman-Tukey, Anscombe, rank and rankit data transforms are available and can be specified for any endpoint as a default transform;
  • Compare primary control to secondary controls, primary control to treatments, or conduct all pair-wise comparisons. Controls can also be pooled by just selecting the desired controls from a listbox.
  • A modified box plot is displayed during data analysis and on the printed reports. The box plot displays the mean, median, maximum and minimum values.





"CETIS is a complete toxicity testing database. It sets up the tests, creates the paper work, analyzes the data, prepares the reports, creates control charts, and identifies whether the test meets acceptability criteria or not. I can't imagine anyone doing toxicity testing without CETIS."

-Stan Asato, Supervisor for EMD Toxicity Testing Unit, City of Los Angeles