In article <[EMAIL PROTECTED]>,
Jerry Dallal <[EMAIL PROTECTED]> wrote:
> We often have a group of individuals who are judged comparable in
> responsibilities and performance. In such cases, it *may* be
> appropriate to use permutation methods. The rationale would be:
> There is some variation in salary due to situation-specific
> "whatever", so we don't expect everyone to be making exactly the
> same money. One way to attempt to detect discrimination against a
> particular group is to ask whether their salaries are consistent
> with what one would expect assigning salaries at random without
> regard to group membership.
That would test whether one had evidence that salaries were related in
some way to group membership. It would NOT test whether any such
relationship is due to "discrimination". It would do so only under
the assumption that there were no relevant differences between groups
other than salary. That is of course exactly what the report under
discussion is disputing. In disputing this, it is necessary only to
show that the actual people in the groups differ, not to make
inferences to a larger population (in the report, however, it seems
that only selected subgroups were actually looked at, so there is some
inference to the whole group implicitly being made). This is in
contrast to trying to show discrimination, where even after
eliminating any possible legitimate reasons for a difference, one
WOULD need to consider statistical significance, since otherwise, one
could mistake a totally capricious salary scheme for a discriminatory
one.
To add another example to the discussion, suppose that there were a
statistically significant difference in salaries between men and
women. But suppose that it also turned out that all the men were
educated in Britain, whereas all the women were educated in China.
Would anyone take this as evidence of GENDER discrimination? I don't
think so. It makes no difference whether such a relationship between
gender and country of education is or is not statistically significant,
it undermines the conclusion of gender discrimination regardless.
The discussion here seems rather strange to me. Surely the burden of
proof ought to be on those who claim that discrimination exists, since
given the small numbers involved, the defence might be hard put to
definitively refute a claim of discrimination even if it's completely
unfounded. If the large differences in publication citations,
etc. seen aren't statistically significant, it seems quite unlikely
that the differences in salaries are statistically significant. It
seems even less likely that a model that accounted for possible
dependence of salary on performance would find a significant gender
effect. Moreover, data on salaries was apparently not available to
the authors of any of the studies under discussion, so nobody is in a
position to even try to demonstrate that discrimination exists.
----------------------------------------------------------------------------
Radford M. Neal [EMAIL PROTECTED]
Dept. of Statistics and Dept. of Computer Science [EMAIL PROTECTED]
University of Toronto http://www.cs.utoronto.ca/~radford
----------------------------------------------------------------------------
=================================================================
Instructions for joining and leaving this list and remarks about
the problem of INAPPROPRIATE MESSAGES are available at
http://jse.stat.ncsu.edu/
=================================================================
