Hi everybody:
I have one of those "let's take a look at an old statistics book with
SPSS eyes" Friday evenings... I have been trying to turn into simple
SPSS code Ching Chun Li's example of randomized blocks split-plot
example (chapter 25 of his out of print masterpiece "Introduction to
Experimental Statistics" book). I have used UNIANOVA and MIXED to get
the same ANOVA table he presents in table 25.2 (and I shuddered when I
remembered that I used to be able, as a student, to obtain it with the
only aid of a calculator).
Here is the code for the dataset and the UNIANOVA solution:
* Ching Chun Li's Randomized blocks Split-plot example
(Fig 25.1 & Table 25.1 from: Introduction to Experimental Statistics) *.
DATA LIST LIST/plots blocks TreatA TreatB yield (5 F8).
BEGIN DATA
1 1 3 3 33
1 1 3 1 26
1 1 3 2 21
1 1 3 4 31
2 1 2 4 24
2 1 2 1 13
2 1 2 3 13
2 1 2 2 28
3 1 1 2 8
3 1 1 4 14
3 1 1 1 12
3 1 1 3 11
1 2 2 1 11
1 2 2 4 30
1 2 2 2 20
1 2 2 3 22
2 2 1 2 26
2 2 1 3 18
2 2 1 4 16
2 2 1 1 8
3 2 3 4 24
3 2 3 3 28
3 2 3 2 27
3 2 3 1 28
1 3 3 1 24
1 3 3 2 24
1 3 3 4 17
1 3 3 3 23
2 3 1 3 10
2 3 1 1 19
2 3 1 2 24
2 3 1 4 20
3 3 2 3 14
3 3 2 4 18
3 3 2 1 17
3 3 2 2 18
END DATA.
VAR LEVEL plots TO Treatb (NOMINAL).
VAR LABEL TreatA 'Plot treatments (A)' / TreatB 'Sub-plot treatments (B)'.
VAL LABEL TreatA 1'A1' 2'A2' 3'A3' / TreatB 1'B1' 2'B2' 3'B3' 4'B4'.
* Goal: to obtain the following ANOVA table (Table 25.2 of the book):
* Source SS df MS F
*------------------------------------
* blocks 42.0 2 21.0 0.47
* TreatA 618.0 2 309.0 6.93
* Error1 178.5 4 44.6
* TreatB 111.11 3 37.0 1.44
* A*B 216.22 6 36.0 1.40
* Error2 462.17 18 25.7
*-------------------------------------
* Error1: whole plot error
* Error2: sub-plot error
* Using UNIANOVA (error terms OK) .
UNIANOVA
yield BY TreatA TreatB blocks plots
/RANDOM = blocks plots
/INTERCEPT = EXCLUDE
/DESIGN = blocks TreatA plots(blocks*TreatA) TreatB TreatA*TreatB .
And here is the problem:
* Using MIXED (1): this doesn't work (setting 'blocks' as RANDOM: wrong
error term for TreatA).
MIXED
yield BY TreatA TreatB blocks plots
/FIXED = TreatA TreatB TreatA*TreatB | NOINT
/RANDOM= blocks plots(TreatA*blocks).
* Using MIXED (2): this works (setting 'blocks' as FIXED: correct error
terms).
MIXED
yield BY TreatA TreatB blocks plots
/FIXED = blocks TreatA TreatB TreatA*TreatB | NOINT
/RANDOM= plots(TreatA*blocks).
If I set blocks as a random factor (as I did with UNIANOVA, and as I
usually treat blocking factors), then MIXED considers it redundant and
computes the wrong error term for TreatA. Why?
Happy weekend,
Marta García-Granero
--
For miscellaneous statistical stuff, visit:
http://gjyp.nl/marta/=====================
To manage your subscription to SPSSX-L, send a message to
[hidden email] (not to SPSSX-L), with no body text except the
command. To leave the list, send the command
SIGNOFF SPSSX-L
For a list of commands to manage subscriptions, send the command
INFO REFCARD
Locked
