Example 24.3 Split Plot

In some experiments, treatments can be applied only to groups of experimental observations rather than separately to each observation. When there are two nested groupings of the observations on the basis of treatment application, this is known as a split plot design. For example, in integrated circuit fabrication it is of interest to see how different manufacturing methods affect the characteristics of individual chips. However, much of the manufacturing process is applied to a relatively large wafer of material, from which many chips are made. Additionally, a chip’s position within a wafer might also affect chip performance. These two groupings of chips—by wafer and by position-within-wafer—might form the whole plots and the subplots, respectively, of a split plot design for integrated circuits.

The following statements produce an analysis for a split-plot design. The CLASS statement includes the variables Block, A, and B, where B defines subplots within BLOCK*A whole plots. The MODEL statement includes the independent effects Block, A, Block*A, B, and A*B. The TEST statement asks for an test of the A effect that uses the Block*A effect as the error term. The following statements produce Output 24.3.1 and Output 24.3.2:

title1 'Split Plot Design';
data Split;
   input Block 1 A 2 B 3 Response;
   datalines;
142 40.0
141 39.5
112 37.9
111 35.4
121 36.7
122 38.2
132 36.4
131 34.8
221 42.7
222 41.6
212 40.3
211 41.6
241 44.5
242 47.6
231 43.6
232 42.8
;
proc anova data=Split;
   class Block A B;
   model Response = Block A Block*A B A*B;
   test h=A e=Block*A;
run;

Output 24.3.1 Class Level Information and ANOVA Table
Split Plot Design

The ANOVA Procedure

Class Level Information
Class Levels Values
Block 2 1 2
A 4 1 2 3 4
B 2 1 2

Number of Observations Read 16
Number of Observations Used 16

Split Plot Design

The ANOVA Procedure
 
Dependent Variable: Response

Source DF Sum of Squares Mean Square F Value Pr > F
Model 11 182.0200000 16.5472727 7.85 0.0306
Error 4 8.4300000 2.1075000    
Corrected Total 15 190.4500000      

R-Square Coeff Var Root MSE Response Mean
0.955736 3.609007 1.451723 40.22500

First, notice that the overall test for the model is significant.

Output 24.3.2 Tests of Effects
Source DF Anova SS Mean Square F Value Pr > F
Block 1 131.1025000 131.1025000 62.21 0.0014
A 3 40.1900000 13.3966667 6.36 0.0530
Block*A 3 6.9275000 2.3091667 1.10 0.4476
B 1 2.2500000 2.2500000 1.07 0.3599
A*B 3 1.5500000 0.5166667 0.25 0.8612

Tests of Hypotheses Using the Anova MS for Block*A as an Error Term
Source DF Anova SS Mean Square F Value Pr > F
A 3 40.19000000 13.39666667 5.80 0.0914

The effect of Block is significant. The effect of A is not significant: look at the test produced by the TEST statement, not at the test produced by default. Neither the B nor A*B effects are significant. The test for Block*A is irrelevant, as this is simply the main-plot error.