I am having problems understanding what is orthogonality, its relationship to sums of squares (thus it impact in the rest of the inferential statistics) and the benefit of orthogonal factors.
I don't understand either how the degrees of freedom are computed in Repeated Measures(seems to be K-1, N-1).
Finally, when you say in the review guide "Multiple Comparison Procedures", do you mean the epsilon adjustments of df used for correction of violation of sphericity?
I don't understand either how the degrees of freedom are computed in Repeated Measures(seems to be K-1, N-1).
Finally, when you say in the review guide "Multiple Comparison Procedures", do you mean the epsilon adjustments of df used for correction of violation of sphericity?
Labels: and multiple comparison procedures, df's, orthogonality
posted by Anonymous @ 8:03 PM
1 Comments:
At 10:07 PM,
Unknown said…
Orthogonal means statistically independent. If predictors are orthogonal, you are able to have an independent estimate of the effects of each predictor. If predictors are not orthogonal, the variance they are accounting for in the DV overlaps, and it is difficult to determine which amount of variance belongs to which predictor. (I believe that Venn diagrams were used to show this in class.) If there are equal cells, the predictors are more likely to be orthogonal because they are essentially as equally weighted as possible. Different types of sums of squares are used to account for this if factors are not orthogonal. The confidence in your inferential statistics may decrease if your predictors are not orthogonal (i.e. F is positively biased).
Yes, degrees of freedom in RM are k-1 (df-A in the handout) and N-1 (df-S) where k is number of times measured and N is the sample size. Df-AS is the df error term and is literally the other two multiplied together: (k-1)*(N-1). Df-total isn't really used, but it is df-A, df-S and df-AS added together.
Multiple comparison procedures refers to post-hoc tests (i.e. Scheffe, Tukey).
Post a Comment
<< Home