[Air-l] response on ginis

[Philip Howard]

Max Forte is absolutely right that using gini coefficients this way would
require that the researcher make sensible judgments about attribution and
content value.  To really know your subjects and their relationship a
researcher probably would need deep experience with their interactions.
However, the graph could be composed even if the researcher had no
qualitative experience with the group, using number of people on one axis
and number of words, number of lines, number of sentences, number of posts
(as Hendricksen does).  These things easy to count no need for qualitative
judgments.  Probably more interesting would be to graph number of people on
one axis and ‘good ideas generated’, ‘smart repartee’ or ‘meaningful
engagement’ on the other axis.  These things would require deep knowledge of
the group and a qualitative researcher’s instinct for how to define, measure
and attribute a good contribution.  A qualitative researcher’s insight would
also be needed to figure out any causal connection there might be between
being a member of a social elite, writing a lot of text, and generating good
ideas.  The coding would probably have to be done by hand.

I interpreted the original post to be a query about how to measure any
deviance from a perfectly egalitarian group, where all the respondents
participate equally, generating the same amount of text, ideas, or hot air.
Mathematically we can know what that perfect distribution is, so my gini
idea is just a way of calculating deviation from that ideal.  I absolutely
would not argue that ginis can be meaningful alone, for the reasons Forte
offers.  I like the qualitative approach that then provides some
quantitative context to show how qualitative findings are transportable.
And anyway multi method triangulation is always best.  I originally
suggested that the number would be most useful in a comparative context.
Lakhani agrees, and even if we did ginis for AIR it would be interesting to
compare ginis on different topics or variation over time.  He also points
out that there would be noise inany database of content, but there must be a
way of doing somekind of confidence interval, error term on the datapoints,
or significance tests on the gini itself.  My math skills end here, so if
anyone can develop a significance test for a quantitative measure of
deviation from a theoretically perfect distribution of content/ideas in a
group of people please post it!  But please don’t do it to the AIR list too
inside baseball I think.
p.
Philip N. Howard
Assistant Professor
Department of Communication
University of Washington
http://faculty.washington.edu/pnhoward/