Re: software for categorizing constructs

Devi Jankowicz (
Wed, 21 Apr 1999 21:26:04 +0000

Fionn Stevenson writes:

>I have now completed nearly 40 x2-3 hour interviews for my main
>survey on multicultural responses to common building materials .

Solid data!

>While I'm sorting out the grids, I've decided to try and categorize the
>constructs. I have four groups to sort (10 interviews each) with
>between 15-60 constructs per interview . So I'm getting about
>150-600 constructs to sort per group!

Just as a matter of interest: what is your basis of grouping? Do the
groupiongs reflect soem hypothesis that you have in mind? (See
alternative 1.2) below.)
>Does anyone know of any nice software that could help me by sorting
>constructs in terms of key words searches, key phrase searches etc.?
>How do folk usually "sort" their elicited constructs into categories?

By making judgements about which categories the constructs should be
coded into.
The categories are obtained in two ways:
1) suggested by some pre-existing analytic framework, derived from either
1.1) a strong theory which specifies what categories pertain given your
realm of discourse
1.2) a research hypothesis which you may be using yourself, based,
possibly, on 1.1, but not necessarily; (here the categories are
pre-defined for you); OR
2) suggested by your own judgement as you go through all the constructs
when you're being exploratory and have not started with any stronger
hypotheses or hunches.

Whether you use a piece of software to guide you, or make those
judgements directly, is only a matter of convenience (faster, perhaps, or
more swiftly indicative of interesting relationships which might suggest
category membership); it _isn't a matter of accuracy, "validity", or (as
we're all constructivists here, yes?) more effective understanding of
what's going on.

I say this for just one reason. Running some software over the data may,
if one's not careful, obscure the actuality of what you're doing: however
you do it, using some neat software or not, it's _you_ that make some
judgements which label the categories: yourself. Just your construing of
the categories and the membership of those categories. Just one person's
way of looking at it.

Now, if your survey is to communicate to other people (especially if it's
one of those constipated, rule-bound pieces of work called a Doctoral or
Masters Dissertation), it's important to demonstrate that your own
construing of your data isn't arbitrary; that it's reliable; that someone
else would agree with the way in which you have chosen categories and
assigned constructs to categories.

Under 1.1), your argument is based on the generality and reputation of
the theory you've chosen, and all you have to show is that your coding of
constructs to categories is reliable.
Under 1.2, your argument is based on the adequacy with which you inducted
your hypothesis: a matter of the internal consistency of your reasoning
from hypothesis to categories; and, again, it remains for you to show
that your coding of constructs to categories is reliable.
Under 2), you have to do both. i.e.

--demonstrate that at least one other person would have arrived at the
same categories as you
--demonstrate that at least one other person would have assigned the same
constructs to the categories being used.

So; however you do it, using software or not, you'll need to do your
analysis four times: once yourself; once by the other person (an
independent assessor briefed about your objectives and your intentions in
choosing categories); and once more, by yourself, and by your helper,
both of the latter being after you have both negotiated over the meaning
of your categories and thereby, hopefully, arrived at a more accurate
coding of constructs to categories.

It's then very useful to be able to demonstrate the improvement in
reliability between the first two occasions and the second two. The
percentage agreement over codings to categories; Cohen's Kappa calculated
for the agreement between the two sets of codings of constructs to
categories; or, (far better than either of these since it allows you to
calculate the sampling error of the statistic involved), the
Perrault-Leigh coefficient of agreement between two judges, would be the
statistics to compute.

I do hope that's helpful!

Kindest regards,

Devi Jankowicz

Cohen, J. "A coefficient of agreement for nominal scales", _Educational
and Psychological Measurement_, 1960, 20, 37-46.
Cohen, J. "Weighted Kappa: nominal scale agreement with provision for
scaled disagreement or partial credit", _ Psychological Bulletin_, 70,
October, 213-220.
Perrault, W.D. Jr.& Leigh, L.E. "Reliability of nominal data based on
qualitative judgements", _Journal of Marketing Research_ 1989, XXVI, May,