Cubic spline variable formation                                 (STB-10: srd15)
-------------------------------

    ^spline^ # variable [^if^ exp] [^in^ range]

This ado-file designs a cubic spline variable(s) with anywhere from 3-7 knots
at default joins.  The # symbol in the syntax must be replaced by an integer
from 3 to 7 inclusive.  This is followed by a variable name; this should be a
continuous variable.

#-2 new variables are formed and given the default names ^knot1^, etc.  These
should be included in your model; there is little restriction on your model:
it must be a single-equation but it can certainly be nonlinear (logistic, cox,
etc. are allowed).  After estimating your model with your variables, the
linear form of the variable used for the splines and all "knot" variables, you
should perform a joint test of the spline variables (the linear form and all
knot variables simultaneously; I recommend ^testparm var kn*^).  If this test
is significant, the linear form is not sufficient.  Obtain the predicted
values and graph these predicted values against the linear variable--this
graph MAY help you in determining a possible functional form, or
transformation, to apply to this variable.

The variables called knot# remain in your data set; thus, if you don't want
these variables, or you want to use ^spline^ again (either for the same variable
but with a different number of knots or for a different variable), you should
either drop the variables or re-load your data set.

Author
------

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