Just got back from overseas and have seen your most interesting query.
Bob Green and Mike Donovan have given you some useful pointers, some of
them beyond my own level of competence; Mike's comments about being
careful to keep your supervisor and extenrals happy is particularly
important. (Infuriating that things should be this way, but true!)
My own two-pennorth, if you haven't yet decided on a completely
quantitiative design, wouild be as follows.
a) Don't use supplied constructs (one tends to, to simplify subesquent
analysis, when a quantitative approach is chosen).
b) Make the individual constructs, and not the individual respondents,
your unit of analysis. A sample of 50 people will then be ample for the
subsequent steps.
c) Try and elicit 8 or 9 constructs from each respondent in a series of
_individual_, not group, grid interviews. (By doing it individually,
taking trouble to clarify terms, laddering down for greater operational
specificity, etc., you get more useful constructs than in a mass-produced
group grid session. Okay, so that's a minimum of 50 hours of
interviewing: so be it. Oy, oy: you're young, the world is your oyster,
and time is on your side. "GIGO" is as true of grid work as it is of
computer work, IMHO.)
d) You'll get a massive set of over 400 constructs, many of them looking
as though they're conveying similar meanings. Now do a Honey (1979)
content analysis, seeking to arrive at 8 to 14 or so different
categories. (His "match-against-a-single-supplied-construct" procedure is
crucial to this form of content analysis, by the way.) Code all the
constructs across those categories and arrange all the constructs under
these category headings.
e) Now look to see to what extent the constructs which were provided by
particular respondent groupings in which you're interested
(males-females, hi-lo aggressives, or whatever) are spread differentially
across the categories. (So if you had, say, 12 categories, you'd do a 2 x
12 table to see whether the males tended to have more of one sort of
category than the females, less of another, and so on. Then another N x
12 table for any other N-valued crosstabulation variable; and so on.)
This simple counting procedure will surely not offend your supervisor's
sensibilities regarding mixing qualitative and quantitative approaches,
and will serve to familiarise you with the meanings being used by your
respondents.
And so, if you subsequently decide to go all quantitiative, you'll have
derived some hypotheses already.
(A first step towards keeping your supervisor happy would be to do a
reliability check of your content analysis coding: Perrault and Leigh's
(1989) measure in place of Cohen's Kappa or a simple percentage
reliability score would surely satisfy the most numerically- obsessed
supervisor that the initial content analysis (call it a "pilot study")
was "really" quantitative in spirit.
Or, if you subsequently decide to continue your investigations
qualitatively, you might want to explore the personal values of your
respondents categorised by the various groupings in which you're
interested. That means a resistance-to-change procedure, and re-accessing
your original respondents for another half-hour each of their, and your,
own time. (Unless of coure you do both the original grid and the
resistance-to-change grid in a single sitting. but then you'll have had
to decide between qual and quant at the outset. Infuriating, but if
that's the way your supervisor sees things and you can't convince him/her
of the folly involved in that way of construing the world of research...
I guess it has to be so.)
Shout if you think I can say anything else that's useful.
Kind regards,
Devi Jankowicz
******
References
Honey P. (1977-or may be 79:check) "The repertory grid in action"
_Industrial and Commercial Training_ vol. 11, no. 11, pp. 452-459, and
look at his articles in nos. 10 and 12 for that year as well: useful
source of ideas.
Perrault W.D. & Leigh L.E. (1989) "Reliability of nominal data based on
qualitative judgements" _Journal of Marketing Research_ vol. XXVI, May,
pp. 135-148 but don't calculate the formulae themselves: once you've read
the article to understand their rationale for the sampling distribution
which underlies the statistic they advocate (always a good idea!) just
use the handy nomogram they provide, instead: saves time.
The Honey, Perault & Leigh, laddering, and resistance-to-change grid
procedures were described in a series of very user-friendly illustrated
procedural guides in the _Newsletter_ of the European Personal Construct
Association. If the current editor is reading this, perhaps he'll contact
you if you require these particulars; if he's not, you could contact him
as follows:
John Fisher, e-mail address
<John_M_Fisher@compuserve.com>
or
<101515.501@compuserve.com>
%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%