Date: Tue, 12 Feb 2002 13:18:21 -0500
Reply-To: "Burleson,Joseph A." <email@example.com>
Sender: "SPSSX(r) Discussion" <SPSSX-L@LISTSERV.UGA.EDU>
From: "Burleson,Joseph A." <firstname.lastname@example.org>
Subject: DV's with neg numbers
Content-Type: text/plain; charset="iso-8859-1"
There is no effect of having negative numbers as part of the continuum of
the dependent variable. Means, variances, and SD's are calculated the same,
as are the z-score transformations that precede the formulation of the
standardized regression coefficients. The unstandardized coefficients remain
in terms of the original units of the DV (and the respective IV's).
If you need to transform the DV, however, due to skewness, which can easily
happen with income variables, be careful. A natural log transform, for
example, would require you to make the range of the Net Worth have as its
lowest value, some positive number, but not zero or negative numbers. You
could solve this by adding the positive value of the lowest negative number
COMPUTE newvar = oldvar + 888001.
That would make the lowest value equal 1, and the highest value equal
2888001. Then you could transform if the skewness is positive (tail right):
COMPUTE lnvar = LN(newvar) .
If the skewness is negative (tail to the left), you could then "mirror" the
DV's by subtracting the new variable from the highest value (plus one) of
the resultant new variable:
COMPUTE newvar2 = 2888002 - newvar.
Then log as before:
COMPUTE lnvar = LN(newvar2) .
Keep track of the "inverted" variable, now defined as "negative Net Worth."
Back transform (in two steps) to get back to the orignal means.