Sample size etc.

Robin Hill (BSRAH@TWP.AC.NZ)
Tue, 13 Feb 1996 10:09:28 +1300

A day or two ago, I recall reading a posting to this list about ways
of handling grid elicitation with larger numbers of subjects (thirty
per group was cited). This was followed by a response from Devi
regarding the debatable notion of 30 subjects per group and advising
against trying Grid elicitation by mail. I thought I'd add my
two-pennies worth.

To the questions "How big should my sample be?" "How many subjects
should I have per group?" my reading of the literature suggests that
the _most_ correct answer is "big enough" or "sufficient." While
this might seem flippant, among the many answers various textbooks
might give, this answer appears to be more correct than others. The
reading then suggests that "big enough" sample size means "as big as
my budget will allow." That includes not only the budget of
financial resourses but also the budget or limitations in all other
resources -- time, assistance, colleague support, equipment, software
and so on. Don't forget to include time to analyse the data, as well
as time taken to elicit grid data. Remember what Bell (1988) says
(repeated in my own recent paper in Vol. 3 of "Advances in PCP.")
Grids produce a deceptivley large amount of data in a very compact
form. In a 10 by 10 grid there are 130 pieces of information to be
digested for just one individual. If you have thirty subjects per
group and 2 groups, then you're facing analysis of 7800 pieces of

The question of sample size seems to be a dilemma of similar proportions
to some ethical dilemmas. Is it better for the research to continue
with smallish sample sizes or to be abandoned? Is it better to do
some research rather than none at all? In some disciplines you would
be laughed off the podium at a conference if you did not use the
mathematically hard-nosed prickly rules for determining sample size.
I suggest that when faced by those who subscribe to this list and
who attend PCP conferences you will get a more sympathetic hearing.
Hence, I'm trying to support what Devi was saying, and suggest that
smaller sample size might be acceptable if you consider PCP
subscribers as your academic peers. Given the
enormity of the data set produced by grid based research, you may
need to compromise mathematical hard-nosed scientific precision, in
the interests of actually doing some research that is useful and
practical in the long run.

Devi alluded that Grid method doesn't work by mail-out survey. While
not used in mail-out form a colleague & I, here in New Zealand, have
used grid elicitation by booklet form, as a more remote method when
face-to-face interviewing has not been practical. I'm sure its
common practice elsewhere round the globe, and hence those interested
should be able to find something close to home. I have also wondered
about the role that interactive grid elicitation packages (such as GPack)
might play in circumstances where large numbers of face-to-face
interviews are unmanageable.

Sorry I can't remeber the name of the person who originally sparked
this discussion. Hope my thoughts reinforce those of others and give
you some comfort.

Dr. Robin Hill

Senior Lecturer & Research Leader
Department of Business Studies
The Waikato Polytechnic
Private Bag 3036
Hamilton 2020
New Zealand