48 SPSS UPDATE 7-9,

There are several possible strategies for analysis of repeated measures designs. Both
univariate and multivariate solutions can be obtained. Selection of a strategy should be based on
the appropriateness of the necessary assumptions as well as power considerations.

1.46 An Example

Data from a repeated measures design found in Winer (1971, p. 546) are shown in Table 1.46. They
consist of accuracy scores obtained by adjusting three dials (DIAL) under two levels of background
noise (NOISE) during three consecutive ten-minute periods (PERIOD). Each subject is observed
nine times, once at each combination of period and dial type. PERIOD and DIAL are called
within- subjects factors, while NOISE is called a between-subjects factor. If subject is considered a
factor. then the subject factor is crossed with PERIOD and DIAL, but nested under NOISE level.

Table 1.48

rT—
Periods: 1 2
Noise Subject

2 3]1 2 3

ki 1

1 45 53 60) (40 52 57
1 2 35 41 50 | 30 37 47
60 65 75 | 58 54 70

50 48 61
42 45 55
b dab //

25 34 Sl
)) 37 43
odo /

16 23 35
22,27 3]
ou I6

1.47 Obtaining a Univariate Analysis the Hard Way

The univariate analysis of the repeated measures design displayed in Table 1.46 is obtained by

treating subject as a random effect nested under the NOISE factor. The model is called a mixed- |

effects model, and the resulting analysis is a mixed-model analysis of the repeated measures design. |
The technigue described in Section 1.25 can be used to determine the appropriate error terms |

for testing the various effects. Table 1.47 summarizes the effects and corresponding error terms for

this example.

Table 1.47

Subject within NOISE
PERIOD PERIOD x Subject within NOISE

NOISE x PERIOD

DIAL DIAL x Subject within NOISE
NOISE x DIAL

PERIOD x DIAL PERIOD x DIAL x SUBJECT within NOISE
NOISE x PERIOD x DIAL

a omo ERA EEEEE ROSE ER WERE VERE ROE WERE ER REEHH

Figure 1.47a shows an SPSS command file that can be used to perform a univariate analysis of
the repeated measures design for the data in Table 1.46. The resulting ANOVA table is presented
in Figure 1.47b. A somewhat complicated DESIGN specification is needed because of the multiple
error terms in the model. In the next section, a much easier approach to the same problem is given.

Na OR RMENI