Statistical Methods for Psychology

of correlations that decrease regularly the more removed the observations are from each

other. That sounds like a logical expectation for what we would find when we measure

depression over time. For now we are going to consider the autoregressive covariance

structure.

Having decided on a correlational (or covariance) structure we simply need to tell

SPSS to use that structure and solve the problem as before. The only change we will

make is to the repeatedcommand, where we will replace covtype(cs) with covtype(AR1).

MIXED

dv BY Group Time

/FIXED 5 Group Time Group * Time | SSTYPE(3)

/METHOD 5 REML

/PRINT 5 DESCRIPTIVES SOLUTION

/REPEATED 5 Time | SUBJECT(Subj) COVTYPE(AR1)

/EMMEANS 5 TABLES(Group)

/EMMEANS 5 TABLES(Time)

/EMMEANS 5 TABLES(Group * Time).

Information Criteriaa

504 Chapter 14 Repeated-Measures Designs

2 2 Restricted Log Likelihood 895.066

Akaike’s Information Criterion (AIC) 899.066

Hurvich and Tsai’s Criterion (AICC) 899.224

Bozdogan’s Criterion (CAIC) 905.805

Schwarz’s Bayesian Criterion (BIC) 903.805

The information criteria are displayed in smaller-is-better

forms.

aDependent Variable: dv

Fixed Effects

Type III Tests of Fixed Effectsa

Source Numerator df Denominator df F Sig.

Intecept 1 26.462 270.516 .000

Group 1 26.462 17.324 .000

Time 3 57.499 30.821 .000

Group * Time 3 57.499 7.721 .000

aDependent Variable: dv

Covariance Parameters

Estimates of Covariance Parametersa

Parameter Estimate Std. Error

Repeated Measures AR1 diagonal 5349.879 1060.035

AR1 rho .618198 .084130

aDependentVariable: dv

Here we see that all effects are still significant, which is encouraging. But which of

these two models (one assuming a compound symmetry structure to the covariance matrix

and the other assuming a first order autoregressive structure) is the better choice. We are

going to come to the same conclusion with either model in this case, but that is often not