DFA

What type of questions does DFA ask?

Can group membership be predicted reliably from a set of predictors?

What is a determinant

Generalised variance of a matrix

What assumption are classification procedures highly sensitive too?

Heterogeneity of variance-covariance matrices: Cases are more likely to be classified into the group with the highest dispersion, that is, the group for which the determinant of the within group covariance matrix is the greatest.

When is DFA robust to normality

If normality is caused by skewness rather than outliers

What does cross validation do?

It tests the utility of the coefficients in a new sample!

How many discriminations or dimensions can you have

One fewer than the number of groups or equal to the number of predictors.. whichever is smaller

What are eigenvalues?

Shows how much variance is extracted from a function- explain the percentage of variance explained in the DV

What is the requirement for sample size

Smallest group should not exceed the number of predictor variables

What is a canonical Correlation?

The correlation for that function with the discriminant scores

How many classification equations are required in DFA?

The number of groups = number of classification equations required!

What do the standardised canonical discriminant function coefficients show?

The relative contribution of each variable to the overall discrimination. They are partial coefficients and they only explain the UNIQUE explanation of each independent variable being explained

What is a centroid?

These are the mean discrimiant scores for each group on a function-

What does the structure matrix explain?

Within group correlations- how each variable is correlated with the discriminant function! These give meaning of a function!