On Sun, 9 May 2004, DG wrote:
> I have a 3x3x3 within subjects design.
Do you have a 3x3x3 analysis to show for it? What does the ANOVA table
show?
> I am doing a test of linear contrast to compare performance between
> cells. I want to compare 3 cells with the rest of the 24 cells. I am
> using SAS to perform the linear contrast however I end up with no
> values for F and p in the output.
Why do you think you need a formal test of the linear contrast? It is
evident, just eyeballing the data, that cells {1,2,3} are not different,
on the average, from the other 24 cells. Which is, of course, what SAS
tells you: the SS for the linear contrast is 0.00050112, which is
surely not detectably different from zero. From this, with the output
from your (I presume!) 3-way AOV, you can calculate the corresponding
formal values of F and p. SAS cannot compute this for you, because you
did not supply to SAS any information about the error term (that is, the
error mean square) pertinent to the desired contrast analysis.
> Can someone please check my SAS code and tell me where I am going
> wrong. I have pasted both my input and output.
See below.
> Thanks in advance.
> DG
>
> INPUT
> options pageno=1 formdlim='-';
> DATA distancehypothesis;
> *cell 1, 2, 3 are matched conditions (control conditions);
Which implies that they have a common value on two of your 3-level
factors, which in turn implies that the contrast you seek to analyze is
part of the two-way interaction involving those factors. If that
interaction was not significant in your 3-way analysis, the contrast
will not differ significantly from zero.
> INPUT cell completion accuracy;
"completion" apparently is your dependent variable. What is "accuracy",
and why did you supply it if you weren't going to use it in the
analysis?
> CARDS;
> 1 4.042 3.604
<snip, 26 other values>
> ;
> proc glm;
> CLASS cell;
> MODEL completion = cell / ss3;
So you've specified a one-way analysis with 27 levels, and (from your
data list) one observation at each level. So, as Rich pointed out,
there are zero degrees of freedom for error; and consequently no test
of hypothesis is possible.
> contrast 'control vs treatment' cell 8 8 8 -1 -1 -1 -1
> -1 -1 -1 -1 -1 -1 -1 -1 -1 -1 -1 -1 -1 -1 -1 -1 -1 -1 -1 -1;
> run;
> OUTPUT
>
> The GLM Procedure
> Class Level Information
> Class Levels Values
> cell 27 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23
> 24 25 26 27
>
> Number of observations 27
>
> Dependent Variable: completion
>
> Sum of
> Source DF Squares Mean Square F Value Pr > F
> Model 26 10.22504874 0.39327111 . .
> Error 0 0.00000000 .
> Corrected Total 26 10.22504874
>
> R-Square Coeff Var Root MSE completion Mean
> 1.000000 . . 3.637519
>
> Source DF Type III SS Mean Square F Value Pr > F
> cell 26 10.22504874 0.39327111 . .
>
> Contrast DF Contrast SS Mean Square F Value Pr > F
> control vs treatment 1 0.00050112 0.00050112 . .
------------------------------------------------------------
Donald F. Burrill [EMAIL PROTECTED]
56 Sebbins Pond Drive, Bedford, NH 03110 (603) 626-0816
.
.
=================================================================
Instructions for joining and leaving this list, remarks about the
problem of INAPPROPRIATE MESSAGES, and archives are available at:
. http://jse.stat.ncsu.edu/ .
=================================================================
