dennis roberts seemed to utter in
news:[EMAIL PROTECTED]:
> At 11:28 PM 4/3/2003, jim clark wrote:
>
>
>>As another related argument for the validity of statistical
>>tests, consider a difference of 1$. That is a real difference
>>for the population, but does it represent discrimination?
>
> if you want to discuss salary discrimination ... then one needs to
> consider the guidelines for organizations like EEOC and others ... to
> look at what they do
>
> i do not recall that the original post was specific about salary
> discrimination based on sex ... what i recall was an interest in
> whether there were differences in salaries between males and females
> ... the interest in that could have been for any one of a batch of
> different reasons
Let me add a little detail... The intended use of the
analysis is this (right or wrong): given all of the
employees in a particular job title at a particular
company location, perform the t-test to determine the
number of standard deviations between the means of
males and females. The assumption is that, since these
are "similarly situated employees", there should not
be a "statistically significant" difference in pay between
males and females.
The EEOC deems a difference of 2 or more standard
deviations "statistically significant." If the t-stat
is less than 2, chances are nothing further will be done.
If the t-stat *is* 2 or more, then a regression analysis
should be done to determine what factor is causing the
difference. If the regression analysis shows that education,
experience, seniority, etc. is/are the cause, then
the difference in pay can probably be legally defended.
If the regression analysis shows that the main factor
(excuse me if I am not using the correct statistical
terms) is sex, then an adjustment to pays is in order
to avoid a legal suit. In fact, these analyses are
often done *after* a legal suit has be brought against
the company by an employee.
> the inferential problem here is NOT one of does discrimination exist or
> not ... it is one of considering the employees you have ... AS A SAMPLE
> from all possible employees in the future who might have these same
> sorts of jobs ... and asking is there likely to be a difference in the
> salaries in THAT LARGER population ... based on your "assumed to be"
> sample data
>
> but, i will bet you a dime to a donut that the REAL interest in the
> case being originally posed is ... what evidence do we have that there
> is a CURRENT DIFFERENCE in salaries ... sufficiently large that we need
> to do something about it with THESE employees ... or else they are
> going to take us to court and sue the dickens out of us
You are correct.
> this sort of issue has to do with current payroll employees and whether
> you are compensating them ... males and females doing the same sorts of
> jobs ... fairly
Exactly.
> this is not an inferential problem in the sense of the way we teach
> them in intro stat ... it just is not
So... is the use of the t-test in this situation valid?
The federal government seems to think so since they
use this test themselves.
-- TRW
_______________________________________
My e-mail: t r w 7
@ i x . n e t c o m . c o m
_______________________________________
.
.
=================================================================
Instructions for joining and leaving this list, remarks about the
problem of INAPPROPRIATE MESSAGES, and archives are available at:
. http://jse.stat.ncsu.edu/ .
=================================================================
