I have a bit of a methodological quandary about sampling.
I want to analyse both existing interview data and future
interviews, using a grounded theory approach. The
interviewees can be selected from a population of
participants in a wider study (N= 120), which is looking at
2 distinct groups -
a.)asthma patients who have been using the emergency
services a lot, and
b.) matched controls who have the same age, gender and
(doctor-rated) asthma severity, but have not used the
emergency services for years.
The interviews consist of a fixed core of questionnaires
and categorical questions (necessary for other parts of the
study), as well as a more open, loosely-structured
qualitative section, where I can ask whatever I want.
So far 60 subjects have been interviewed, and another 60
are to be done. (so I still have chances to ask about
evolving categories / concepts)
The sampling restrictions on the main-study population are
not necessarily a problem for the qualitative work, as I am
interested in sampling to maximise the similarities and
differences related to the effectiveness of patients'
asthma-control / coping strategies. Having a sample which
is polarised in terms of an asthma-control outcome enables
me to do this.
My first round of interviewees for open coding will be
selected from the existing (already interviewed)
population, and then further theoretical sampling and
questioning will be based on categories and dimensions
elicited from this first round (I am doing the interviews
myself, and can change the content of the qualitative
section at will / as issues arise).
The question is - when sampling for open coding, should I
pick pairs of subjects (already matched for age, gender and
asthma severity), or just pick from the high-emergency-use
group and the control group at random?
If I pick matched pairs, then this will mean the pairs will
not differ in terms of asthma severity, which is clearly an
imortant determinant of ability to control asthma. However,
the whole point of the study is to find out what other
factors (apart from asthma severity) contribute to good or
poor asthma-control. If I sample matched pairs, this should
theoretically maximise the similarities and differences
related to the effectiveness of patients' asthma-control
strategies, whilst minimising the variation due to
underlying illness-severity.
If I select patients at random from both groups, then
presumably underlying severity would come through as a major
category, but this could serve to obscure / confuse other
important issues. Basically, I don't need to re-discover
that increased severity of asthma causes more control
problems!
Sampling matched pairs is my preferred option at the
moment, but I'm not sure if such restrictive sampling is
acceptable for a grounded theory approach. Can anyone see
any reason why I should or shouldn't select my initial
sample in this way, or suggest a better sampling strategy.
Colin Greaves
The Health Psychology Research Group
Washington Singer Laboratories
University of Exeter
Perry Road
Exeter
EX4 4QG
Tel: 01392 264610 Fax 01392 264623
%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
|