Database Reference
In-Depth Information
box, and X1 through X15 to the “Independents(s)” box (even though you would have
to scroll through them to see them all.) Now we deviate from simply clicking “OK,”
which would get us the output we obtained earlier in
Figure 10.12
.
Now, here's the key to performing stepwise regression versus regular mul-
tiple regression: We need to change what is labeled in the Linear Regression dialog
box as “Method” (see arrow in
Figure 10.13
). When we pull down that menu, we
obtain the circled options that are displayed in
Figure 10.14
:
FIGURE 10.14
The different linear regression
methods
available; SPSS.
SIDEBAR: WHAT ARE THOSE OTHER METHODS, ANYWAY?
As you see in
Figure 10.14
, there are other choices, beside
Stepwise
. We do not believe that there is
any merit in choosing
Forward
; that is equivalent to
Stepwise
, except that once a variable is in the
equation, it cannot be deleted. We believe that the deletion provision of Stepwise is something you
should normally want. The choice,
Remove
, is basically
Stepwise
in reverse. You start with an equa-
tion with ALL the eligible variables and keep removing the least signiicant variable at each stage,
until all the variables left are signiicant and then the process stops; however, it may occur that a vari-
able that is removed ends up back in the equation, perhaps becoming signiicant after other variables
have left the equation—the counterpart of a variable in
Stepwise
that enters the equation, but later gets
deleted. General wisdom suggests that
Remove
is equally good as
Stepwise
, but
Stepwise
is more well
known and is far more popular. The two methods often yield the same inal equation (and it is this last
step result that matters!!), or an equation that is quite similar, and, in cases where the inal equation
is not the same, the predictions resulting from the two equations (i.e., from the two methods) would
usually be nearly the same.
Backward
is
Remove
without the option of bringing back a variable that
becomes signiicant after leaving the equation—analogous to the relationship of
Forward
to
Stepwise
.
Search WWH ::
Custom Search