I don't think you should reassign the values in the way you suggest. A
better approach would be to make this variable into an unordered factor
variable: e.g., estimate a separate intercept for each type of the
government since the effect of such government types may not be linear.
Kosuke
On Sat, 26 Apr 2003, Phillip Y. Lipscy wrote:
I think I have an intuitive sense of the answer to
this question, but it's come
up in one of the datasets we are trying to replicate so just thought I would
canvass others' opinions.
Suppose a variable such as "divided government" is coded -1 for unified, 0 for
partial, and 1 for divided. After running a regression, the variable is not
significant. Is it legitimate then to reassign values such as 0 for unified, 1
for partial, and 255 for divided to produce significant results?
My intuition tells me that this would be legitimate only if there is some prior
theoretical reason to believe that divided government has 255 times the effect
of partial. Is that right?
Thanks,
Phillip.
-------------------------------------------------
Phillip Y. Lipscy
Perkins Hall Room #129
35 Oxford Street
Cambridge, MA 02138
(617)493-4893 DORM
(617)851-8220 CELL
lipscy(a)fas.harvard.edu
http://www.people.fas.harvard.edu/~lipscy/
First Year Student, Ph.D. Program
Harvard University, FAS, Department of Government
-------------------------------------------------
_______________________________________________
gov2001-l mailing list
gov2001-l(a)fas.harvard.edu
http://www.fas.harvard.edu/mailman/listinfo/gov2001-l