Date:         Thu, 31 Aug 2006 17:59:56 +0100
Reply-To:     Paul Mcgeoghan <mcgeoghan@Cardiff.ac.uk>
Sender:       "SPSSX(r) Discussion" <SPSSX-L@LISTSERV.UGA.EDU>
From:         Paul Mcgeoghan <mcgeoghan@Cardiff.ac.uk>
Subject:      choosing a covariance structure in Linear Mixed Models (repeated
              measures)
Content-Type: text/plain; charset=US-ASCII

Hi,

In the Linear Mixed Models repeated measures procedure, I am trying to choose a
residual covariance structure.

In the model, I specify Year and Site which have 2 levels each as Repeated.

The study involves looking at the number of reported items (TotalScore) of 74
individual companies on 2 different websites (Site) in 2 successive years so
Year and Site are repeated with 2 levels each.

I have a residual covariance structure which is Unstructured which seems to be
the best so far according to the BIC (Bayesian information criterion).

The residual covariance structure is below which does seem to follow a
pattern.
So the 1st diagonal variance and the 3rd diagonal variance are very similar
while the 2nd diagonal variance and 4 diagonal variance are fairly similar.
Which would be the best repeated covariance type to choose apart from
unstructured in this case?

        [year = 0]*[site = 1]   [year = 0]*[site = 2]   [year = 1]*[site = 1]
[year = 1]*[site = 2]
[year = 0]*[site =
1]      223.8530498     20.0855724      208.5294008     -.7649333
[year = 0]*[site =
2]      20.0855724      13.2969917      19.3039448      .2376745
[year = 1]*[site =
1]      208.5294008     19.3039448      224.0558073     -.7738841
[year = 1]*[site = 2]   -.7649333       .2376745
-.7738841         .0693235

Thanks in advance,
Paul


==================
Paul McGeoghan,
Application support specialist (Statistics and Databases),
University Infrastructure Group (UIG),
Information Services,
Cardiff University.
Tel. 02920 (875035).