re: Counsellor constructs
Thu, 10 Aug 1995 00:17:10 +0000

Jim Mancuso writes:

> As a personal construct theorist, I would hold that "The Big Five"
>represents the category systems which we use as we attempt to categorize
>>those events which we know as persons. Further, I would relate the idea of a
>>"Big Five" to Miller's magic number seven, plus or minus two!! Seven minus
>>two, being five, as we know (???).

Um. Isn't this to confuse structure with content? Let's unpack it a moment
and try it on for size. The "big five" arise because:
a)- respondents respond to inventory items, attending to the personal
meanings that are suggested
b)- these responses are then analysed by principal component / factor
analysis techniques which identify different variance components
(essentially, patterns of different arrangements of numbers across a set of
raw answer sheets giving rise to correlations between sets of scores the
researcher calls variables, yup?)
c)- the researcher then decides to ignore all but the first five, seven, or
whatever components, bearing the total variance they account for in mind
d)- s/he then applies a label, meaningful to him/her, to each, depending on
the items which s/he judges to be loaded on each component.

Now, the sample of respondents with whose help the inventory is constructed
have each no particular reason to be structuring their responses into the
same groups as the researcher (okay, we all know that respondents seek to
be personally consistent in working with inventory material; what I'm
saying is that, with a personal way of construing the world, they are
unlikely to be systematically using the same ways of structuring it as the

Also, remember that the analysis at step b) is a blind statistical
procedure; of the many different permutations in which the inventory items
could have been grouped into 5 to 7 variance components, the analysis
technique simply chooses the one which optimises variance explained,
according to some (optimax, minimax, whatever, criterion).

And there's utterly no reason why the one which is optimal statistically
should be the one which conveys the pattern(s) of meaning with which the
respondents were working, (remember, they were potentially working with
many, since there were many different respondents!). (Though it just might
be the one to which the researcher is habituated by his/her reading about
the Osgoodian 3, the big 5, or whatever, because the researcher may have
constructed the item pool with this sort of predilection in mind.)

Unless of course the researcher sets his/her statistical criteria with
sufficient insight into the various permutations present in the data, to
deliberately select just that permutation which coincides with his/her,
_and the respondent's_, predilection for 5 or 7 factors.) As an
information-processing task, this seems unlikely to me.

Or have I misunderstood something terribly obvious?

(Entirely possible, mind you; I've yet to find a psychologist-friendly
statistician who can really, _really_ explain eigenvalues to me in a way
which I can understand! "A property of a matrix" is a bit too abstract for
my tastes...)

Kind regards,

Devi Jankowicz