Imagine the following study of relationships between
pay and performance. In a given industry, at the
organizational level, firms' data exists as follows:
functional area
---------------
firm payroll a b c
---- ------- - - -
1 1500 1 2 3
2 3000 3 4 5
3 1750 5 3 1
.
.
.
n
Areas "a," "b," and "c" have several variables used to
assess performance within these areas. At the
individual level, persons could have data in one, two,
or possibly all three categories. No individual will
have no data in all categories, but most individuals
will have date in one of three, or two of three
categories. At the organizational level, all
organizations will have data in all categories.
Most of the organizational level studies on this
industry have used multiple regression as main effects
models only. Accordingly, I have seen few, if any,
studies in this area looking at the potential
interaction between these functional areas.
At the individual level, most studies have been
conducted within functional area, once again, with
main effects multiple regression. Some of the studies
have gone so far as to argue that prediction at the
individual level is done better by using
organizational level results, and applying them to
individual level cases.
My questions are:
1. Is it not fundamentally problematic to make
inferences or predictions about individuals from
organizational level data (i.e., ecological fallacy)?
2. Would the organizational level beta weights be
somewhat misstated for individuals because they do not
take into account the distribution, incidence and
prevalence of the attributes at the individual level?
3. Because individual level studies have been done
within functional area, would relative worth (equity)
analyses be more accurate if the variables were
aggregated at the individual level (with zeroes
indicating absence of a trait), and a regression
equation then estimated?
I would think that the resulting beta weights
would be more accurately estimated for individuals,
than using the organizational level equations.
4. Previous studies have not used factor analytic
methods (for data reduction; surrogate variable
selection; or factor score construction), nor have
they used interaction terms in the analyses.
Would factor analysis more accurately reflect the
underlying construct structures and interrelationships
between the variables at the individual level, or the
organizational level?
I would like to be able to examine whether there
is interaction between the constructs. I believe that
factor analysis for data reduction would be a valuable
approach.
If you have suggestions, please email me directly.
Responses can and will be shared with the list, after
receipt of suggestions seems to be complete (unless,
of course, you want your comments kept private).
Regards,
Nick Miceli
__________________________________________________
Do You Yahoo!?
Yahoo! Finance - Get real-time stock quotes
http://finance.yahoo.com
*
* For searches and help try:
* http://www.stata.com/support/faqs/res/findit.html
* http://www.stata.com/support/statalist/faq
* http://www.ats.ucla.edu/stat/stata/
