i. Assumptions for ANOVA
|
|
Normality test |
Jarque-Bera test. Residuals are previously corrected by the block effect in RBD or by plots. |
Equality of variance test |
Hartley, Bartlett, Levene and Levene(Med). Levene(Med) when r > 3 in CRD and with removal of structural zeros. |
Non-additivity test |
Tukey test for non-additivity. Only for RBD. |
Scan by transformations |
Algorithm to find the type that satisfies both assumptions. |
ii. Analysis of Variance
|
|
ANOVA for CRD |
For balanced and unbalanced data, in single schema or in double or triple factorials. With or without up to six additional treatments to the factorial. |
ANOVA for RBD |
With Yates estimate for missing data when in RBD. Correction of SS for treatments when unbalanced. |
- Split-plot ANOVA |
With estimated values when unbalanced (limited). |
- Split-block ANOVA |
With estimated values when unbalanced (limited). |
Analysis for repeated measures |
Greenhouse-Geisser correction for DF (most conservative condition, ɛ=1/(p-1), editable). |
Nested ANOVA (mixed model) |
Hierarchical model only for two factors (with B random) |
Non-parametric ANOVA |
Simple rank and block rank (RT-2). Aligned rank (ART) to estimate the interaction in double factorials. |
iii. Means tests
|
|
Tukey test |
For balanced and unbalanced data. |
Student-Newman-Keuls test |
For balanced and unbalanced data. |
Dunnett test |
For balanced and unbalanced data (also for factorial). |
Scott-Knott test |
For balanced and unbalanced data. Only approximate for unbalanced data. |
t test |
For balanced and unbalanced data (indicated for orthogonal contrasts and independent samples). |
Bonferroni mod. by Conagin test |
For balanced and unbalanced data (P(F) for treatments). |
Dunn-Sidak test |
For balanced and unbalanced data. |
iv. Regression and other procedures
|
|
Regression ANOVA |
For balanced and unbalanced data. With or without 1 additional treatment. Only for nine select models. |
Descriptive statistics |
Standard error, standard deviation, margin of error. |
Outliers test |
ESD test (L+1=10). Residuals are previously corrected by the block effect in RBD. Chauvenet criterion. |
Size effect statistics |
d-Cohen. |
Others |
complex contrasts, graphs, tables, information from the regression models, verification of correlated errors between successive treatments and other analyzes (since it allows to insert external values of experimental error and DF of error) |