all3anova(y, a, b, c [,s2:val]), REAL vector y, factors a, b, c of same length as y, REAL scalar val > 0. |

all3anova(y, a, b, c) fits all hierarchical 3 factor ANOVA models. y must be a REAL vector and a, b and c must be factors (created by factor() or makefactor()) of the same length as y. For each of the 18 models fit, all3anova() prints the following, in order of increasing C(p). p Number of degrees of freedom in the model, including the constant term C(p) Mallow's Cp statistic Adj R^2 Adjusted R^2 R^2 Unadjusted R^2 Model The right hand side of the fitted model using the symbols a, b and c for the 3 factors The estimate of variance used in computing C(p) is the error mean square from the model fitting all factors and their interactions ("y=a*b*c"). Thus this model must not fit perfectly (have zero residual sums of squares). all3anova(y, a, b, c, s2:v), where v > 0 is a REAL scalar does the same, except v is used in computing C(p) instead of the error mean square from the model fitting all factors and their interactions. all3anova(y, a, b, c [, s2:v], keep:T) does the same except nothing is printed. Instead a structure with the following component is returned: Component Contents p Vector of integers containing p for all the models fit cp REAL vector containing C(p) for all the models fit adjrsq REAL vector containing adjusted R^2 for all the models fit rsq REAL vector containing unadjusted R^2 for all the models fit model CHARACTER vector containing right hand side of each model fit using a, b and c as factor names. The models are in increasing order of C(p). all3anova(y, a, b, c, [, s2:v], keep:T, print:T) returns a structure and prints results. See also all4anova() and screen().

Gary Oehlert 2003-01-15