Re: Grid Stability

Jim Legg (
Fri, 08 Nov 1996 00:42:57 +1300


I'm a new member to the PCP group. My name is Jim Legg and I am a
self-employed computer programmer with 17 years experience in writing,
testing and using grid software. My website under construction has some
interesting points on grid stability, namely the use of second order
correlates called 'Slater Distances' to indicate enough element
separation. Normal correlations can't tell what are enough constructs.
Also the component table measures sufficient three dimensional stability
-- this figure slides downwards on larger grids. Although a grid's
components can be stable, the order of the components can change in a
predictable manner as new constructs are added. I haven't researched its
use in PCP but this elevator effect of components is useful in spotting
trends with financial data, etc.

My site below is also mirrored on Chris Evans' site in London. (+200k)


Jim Legg

W Ramsay wrote:
> Two problems with construct elicitation seem to be eliciting the constructs

Maybe the problem is not construct elicitation but the choice of
elements for a chosen freeform subject matter. It is after-all
completely measured by its constructs. Predefined published grids can
however suffer from ill-fitting constructs if incorrectly re-used.

> in the first place and avoiding duplications or trivisl paraphrases in the

I don't think duplications affect the order of the principal components.
For example, if several constructs say the same thing and they all show
only slight variation they will not make it to the top.

> second. Getting a decent sample of construing behaviour should be of
> paramount importance for our understanding and these problems inhibit this.
> (Many published grids seem to be pretty small, even for quite restricted
> domains of enquiry.) How do we know when a sample of constructs is big
> enough and representative enough? In discussion with a colleague the other
> day we wondered about grid stability as an indicator. Presumably if we do
> successive analyses as we successively elicit constructs these should
> converge on a stable structure, i.e. the order of components' contributions
> to variance, and the number and construct loadings of components, should
> become increasingly stable, with only some redistribution of small amounts
> of variance as newly elicited constructs are added to the analysis.
> If that makes sense, could anyone point us at any work on this? If it
> doesn't, any contribution to our re-education would be welcomed!
> Thanks.
> Bill.
> Bill Ramsay,
> Dept. of Educational Studies,
> University of Strathclyde,
> Jordanhill Campus,
> G13 1PP,
> Scotland.
> 'phone: +44 (0)141 950 3364
> 'fax: +44 (0)141 950 3367
> e-mail: