Table 6.

Summary of simulation study results.

Manipulated variable	Result
Method	CART had lowest misclassification rates; MIXDA had second lowest misclassification rates; LDA, GAM, and LR had highest misclassification rates
Overlap	More overlap led to higher misclassification
N	Larger N generally led to lower misclassification rates for LDA, GAM, and LR. Larger N led to higher misclassification rates for CART and MIXDA
Nratio	Known group size inequality led to lower misclassification rates for all methods except MIXDA
Subgroup ratio	Subgroup size inequality led to higher misclassification rates for all methods
Group separation	Greater known group separation led to lower misclassification rates for all methods