Accounting problem or no, it looks like a multiple regression issue to
me. As such, I would ask:
1) Have you looked for many possible factors (ind. vars)? I'm
thinking something nearly off the wall might be a big one.
If you miss a major factor in your list, then the analysis will say
there is a lot of scatter, and you will conclude that none of what you
have really do a lot for it. Takes thinking to get these down.
2) social science and business analysis in general cannot insist on
orthogonal factor conditions. Which leads to a lot of soul searching
near the conclusions. And confusion in those conclusions. Clearly, you
are going to have that problem. Getting orthogonal factor conditions
does not mean looking only at individuals who are narrowly defined, but
that you get a mathematically proper spread of variations. Can you
search your data for a few factors that seem to really count, then
select data that fits an orthogonal array? use the 'discarded' data to
test the resultant model for predictive capability.
3) It's going to be interesting to see if you can really pull this
one off. Better put gender into it somewhere, while you're at it, too.
That just popped up in the news lately.
Joe Meyer wrote:
> I am trying to estimate how faculty salaries at my university are allocated
> by instructional level and academic discipline to estimate the actual cost
> of teaching a semester credit hour by instructional level and discipline. I
> developed a regression model with faculty teaching purely in specific
> academic disciplines as my observations. Actual salary is my dependent
> variable. Independent variables include lower-level credit hours taught,
> upper-level credit hours taught, masters-level credit hours taught, doctoral
> credit hours taught, total students taught, and dummy variables for faculty
> rank, tenure status, and academic discipline. My original idea was to
> either:
>
> 1) use dummy variables for instructional levels, or
>
> 2) plug in lower-, upper-, and masters-level credit hour data one at a time
> to get separate estimates for each level and discipline.
>
> The dummy variable idea will not work, because I do not know what amount of
> each salary is spent at each level for my dependent variable.
I believe this is part of what you are trying to find out. So you
shouldn't know it up front! :)
> And, I will
> double count the estimators for tenure status, faculty rank, and discipline
> if I just plug credit hour data for each of the different instructional
> levels into the model while using zeros for other levels of instruction to
> get a prediction for each level. Is there a way to predict the salary cost
> by instructional level and discipline without fitting the model only on
> faculty who are teaching purely in one discipline and purely at one
> instructional level? I can do that, but I am concerned that such faculty
> would not be very representative of the population. I am not too worried
> about fitting the regression model to faculty who teach in a single
> discipline, since most do this anyway, but am afraid that limiting to
> faculty who teach at a particular level will skew the results.
>
> Thanks!
> Joe Meyer
>
>
>
f INAPPROPRIATE MESSAGES are available at
> http://jse.stat.ncsu.edu/
> =================================================================
>
>
>
--
Jay Warner
Principal Scientist
Warner Consulting, Inc.
4444 North Green Bay Road
Racine, WI 53404-1216
USA
Ph: (262) 634-9100
FAX: (262) 681-1133
email: [EMAIL PROTECTED]
web: http://www.a2q.com
The A2Q Method (tm) -- What do you want to improve today?
=================================================================
Instructions for joining and leaving this list and remarks about
the problem of INAPPROPRIATE MESSAGES are available at
http://jse.stat.ncsu.edu/
=================================================================
