Hi Fred,
Yes, you should have a look at:
Dunlap, W. P., Burke, M. J., & Greer, T. (1995). The effect of skew on the magnitude of product-moment correlations. Journal of General Psychology, 122, 365-377.
Norris, A.E., & Aroian, K.J. (2004). To transform or not transform skewed data for psychometric analysis: that is the question! Nursing Research, 53(1), 67-71.
Lai, C.D., Rayner, J.C.W., & Hutchinson, T.P. (1999). Robustness of the sample correlation: the bivariate lognormal case. Journal of Applied Mathematics & Decision Sciences, 3(1), 7-19.
http://www.emis.de/journals/HOA/JAMDS/3/17.pdf
Hutchinson, T.P. (1997). A comment on correlation in skewed distributions. Journal of General Psychology, 124(2), 211-215.
Your correlations may also be attenuated to some extent just because they are point-biserial, and your ERP dichotomy is probably not a 50/50 split among the firms involved.
Cam
> Date: Mon, 10 Aug 2009 18:08:44 +0200
> Subject: st: Weird Results
> From: [email protected]
> To: [email protected]
>
> Dear Listers,
>
> I'm encountering an odd issue, which you might be able to explain/resolve.
>
> I'm running xtlogit. My dependent variable is the log of odds of the
> firm experiencing increase in its turnover. The main variable of
> interest on the right hand side is a dummy, indicating if the firm
> uses ERP (a specific type of IT system). I have a set of control
> variables, including number of employees to control for size and
> economy-of-scale effects. When I add the number of employees, I get
> insignificant result for that but significant result for ERP.
> Alternatively, for a better interpretations, when I add
> log(employees), this variables turns to significant but the ERP
> variable becomes insignificant. To further look for the problem, I
> realize that corr(erp, employees) = 0.11 while corr(erp, log(emp)) =
> 0.34. This means that ERP has a much stronger correlation with
> log(emp) then with emp, as log is a non-linear transformation.
> Therefore, in the specification with log(emp) part of the effect of
> ERP is absorbed by this variable and thus ERP turns insignificant. For
> your further info, my sample size is in a order of 20000 observations.
>
> Now, I'm puzzled with the real econometric explanation of this issue,
> ways to resolve it, suggestions to improve it or advice on what
> specification shall be rigorously chosen.
>
> Thank you so much for your support.
>
>
> All the bests,
> Fred
> *
> * For searches and help try:
> * http://www.stata.com/help.cgi?search
> * http://www.stata.com/support/statalist/faq
> * http://www.ats.ucla.edu/stat/stata/
_________________________________________________________________
Send and receive email from all of your webmail accounts.
http://go.microsoft.com/?linkid=9671356
*
* For searches and help try:
* http://www.stata.com/help.cgi?search
* http://www.stata.com/support/statalist/faq
* http://www.ats.ucla.edu/stat/stata/
