A nice debate about "autocoding" because it raises all the
issues of what computers can and can't do. Perhaps it's because the term
"autocoding" snuck in, but aren't we risking rejecting the use of
mechanical processes as a first step to interpretive ones? I don't want
to do that, indeed see search-coding as offering new qualitative tools.
John wrote:
>I understand that there are circumstances when "auto-coding"
can be useful. But I don't think you can ever escape the
problems inherent in auto-coding.
Mmm. Circumstances where it is useful are obvious and in some of
them, with a good tool, such coding will have fewer problems than human
error during tedious work. I think it helps to distinguish descriptive
coding from interpretive, and to distinguish different ways of
autocoding. Qualitative researchers spend a lot of life doing
descriptive coding. If I can autocode all the answers to each question
in a structured interview and all the demographic details of
participants, I will, thanks! Doing this mechanically is likely to avoid
mistakes as well as save time. Incidentally this doesn't require text
search: you can do both by text search in N4, but we did them
differently in NVivo: autocode by section (the system just picks up the
subheaders you imported or created in the rich text file), import values
of attributes in a table.
Then interpretive coding - at the other extreme there are
circumstances where text search for "autocoding" is useful and
absolutely never safe -(John's case that text search for "pain", gets it
in different contexts, and it's the experience, not the word, you are
researching.) Is there a use here for mechanical search? I'd say yes,
so long as the researcher sees it as the first step and the program is
able to search broadly and exclude false finds that are mechanically
excludable. Basic requirements for broadening it are the ability to
build searches with multiple instructions for alternatives and
approximation searches. Basic requirements for excluding false finds are
ability to inspect finds one at a time, specify exclusions and patterns
as well as pointing the search to particular data. (N4 has always done
all of the above, in NVivo you can scope the search much more finely,
including or excluding data according to documents, coding or
attributes.) But of course you will still get false finds and miss
wanted ones whenever (i.e. usually!) the text string is not a unique
indicator of the theme, topic etc sought.
So how to make it safe? Well, it can't be. Which isn't news - or
a reason to deny researchers these tools. I've yet to meet any idiot
who relied on text search autocoding to code at interpretive categories
without inspecting finds and reflecting on and supplementing the coding
retrieved.
But we do need software to give us useful retrievals - as
fine-tuned as possible - in a way that gives the researcher access to
the retrieved text in as qualitative a context as possible. That doesn't
just mean autocoding context. Of course you should not have to "select
arbitrary numbers of lines before and after the occurrence of a word" -
but we do want an automatic widening of context as a first step, (number
of characters or sections are options in NVivo).
And more importantly, we do need the retrieval to be live in a
way that allows you to spread context to see it, jump back to the
document to rethink it and code as context whatever characters are
relevant. This is why both N4 and NVivo present the finds in a Node
Browser. The researcher can use "autocoding" then qualitatively, as just
a first step, a way of scooping up a lot of possibly relevant data for
rethinking, reviewing in context and coding-on to new categories as
appropriate.
There are uses for this sort of first-stage scoop coding that
are quite new to qualitative computing. I've found with N4 that this
sort of scooping-up of data gives a nice way of re-seeing themes - you
get surprising juxtapositions of material, or passages you'd missed, and
you can move between the finds and the context re-viewing, removing
material that shouldn't be there and expanding context appropriately.
And then by coding-on you can store these new ideas and pursue them.
Wouldn't call that "autocoding", though...
cheers
Lyn
Lyn Richards,
Research Professor of Qualitative Methodology, University of Western
Sydney,
Director, Research Services, Qualitative Solutions and Research.
(email) [log in to unmask]
(Ph) +61 3 9459 1699 (Fax) +61 3 9459 0435
(snail) Box 171, La Trobe University PO, Vic 3083, Australia.
http://www.qsr.com.au
%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
|