I was thinking of a paper I gave at the Australian PCP conference in 1984
(hardly new stuff!) where I used the ability of some multidimensional
scaling programs to 'fix' the location of some points and estimate the
rest. Run an mds at time 1 (say). At time 2 'fix' the locations of the
common elements/constructs, and estimate the locations of the other time 2
elements/constructs. Put the two solutions in the same picture. The common
ones will be in the same positions, and you should be able to see how the
time 1 unique constructs/elements relate to the time 2 unique ones.
Other possibilities:
1.Bill Rozeboom has a factor rotation package called 'hyball' which is a
bit idiosyncratic in operation but which, if memory serves me, allows you
to 'fix' some loadings.
2. Use confirmatory factor analysis (as a heuristic rather than a model
fitter) to do the same thing
Richard
At 09:59 13/06/99 +1000, you wrote:
>Richard,
>
>I was going to save my statistical questions about comparing grids/measuring
>change, until after Berlin, but the following interested me. If you mean
>something else other than factor analysis etc, could you elaborate upon what
>you mean by "fancy analysis?
>
>
>>Further to Ken's good ideas...another possibility is to carry forward some
>>of the elements & constructs from admin to admin. After eliciting the new
>>grid, review the old and find out what is and isn't relevant at the present
>>time, and perhaps re-rate those aspects that are still relevant. The
>>advantage of carrying forward part of the grid is that it can provide a
>>link through which (with some fancy analysis) you can relate the unrelated
>>aspects of the different grids.
>
>
>regards,
>
>bob
>
>
>
>
>
>
Richard C. Bell
Associate Professor
Department of Psychology
University of Melbourne
Parkville Vic 3052 Australia
ph: +61 3 9344 6364
fax: +61 3 9347 6618
%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%