Analysis of variance

Page 1

Univariate = one way

Multivariate = two or more ways (two or more separate variables which can each divide sample into two groups)

ANOVA can be used instead of T Test, even if only 2 levels, but should always be used for 3 or more groups

Interaction = does the effect of one variable affect the other?

On plots = the steeper the line, the bigger the effect

If lines are parallel = no significant interaction

Plots show where interaction is (in what direction); test figures only say how big it is

Main effect (eg time or weather) = difference between average time/weather of one one variable compared to other

Interaction = difference between differences

To run a contrast = insert following before /method; /contrast(variable name) = special (.....) - contrast weights

Scheffe (post-hoc) - use same procedure as normal, inc contrasts but ignore significance.

Scheffe criterion = (K-1) x Tabled value of Fk-1, error df

where K = number of levels in factor being examined

error df = degrees of freedom of demonimator from table

If F = higher than scheffe criterion answer it is significant

New Page