Transformation of an independent variable                        (STB-10: sg14)
-----------------------------------------

              ^tidwell^ depvar indepvar [^if^ exp] [^in^ range]

where depvar is the dependent variable in your regression and indepvar is the
independent variable that you are considering transforming.

Description
-----------

^tidwell^ temporarily adds a variable to your data set that is a function of 
the independent variable.  This new variable (called ^resp0^) is added to your 
regression.  If the t-test of ^resp0^ is statistically significant, then you
should transform.  The t-test and its p-value are shown although the regres-
sion itself is not.  A crude guess of the transform is (coef. of ^resp0^/coef.
of variable from model without including ^resp0^) + 1. This value is also
provided, as is a version of this value rounded to the nearest .5.  If the
suggested transform is outside the range of what you consider reasonable
(e.g., -2 to +2), then ignore it (absurd results are possible--BE CAREFUL).

Note that a general alternative for transformation of the predictor(s) is
polynomial regression.  If you want to check more than one predictor, use 
this procedure sequentially.

Author
------

Richard Goldstein, Qualitas, Brighton, MA,   EMAIL goldst@@harvarda.bitnet