Date: Fri, 15 Jun 2001 02:19:16 -0300
Reply-To: Hector Maletta <hmaletta@OVERNET.COM.AR>
Sender: "SPSSX(r) Discussion" <SPSSX-L@LISTSERV.UGA.EDU>
From: Hector Maletta <hmaletta@OVERNET.COM.AR>
Subject: Re: tolerance II - forcing excluded variables
Content-Type: text/plain; charset=us-ascii
You can do it by syntax, using the CRITERIA subcommand of the REGRESSION
command. One of the possible specifications within criteria is the
keyword TOLERANCE(value) in which you can specify a value different from
the default which is 0.0001. You can lower it, but I do not know whether
0 is acceptable as a value. You may also receive a warning if you
specify a very low value, but the results would be obtained anyway.
The syntax is for instance as follows:
REGRESSION [other subcommands]
/CRITERIA=TOLERANCE(0.000000001)
/[other subcommands].
Hector Maletta
Universidad del Salvador
Buenos Aires, Argentina
Mark Clark wrote:
>
> Hi again folks-
> I'm going to repost this question on the chance that it went under the
> radar screen of anyone who might have some helpful advice. I am trying to
> determine if there is a way to "force" excluded variables in a hierarchical
> regression that are booted due to "tolerance = .000 limits reached" (the
> theoretical approach demands inclusion of all variables). Is there a way to
> set the tolerance limit manually or otherwise force the variables to remain
> in the equation? Or should I just report the coefficient of those excluded
> variables as .00 (and does anyone have a supportive cite for this)?
>
> I did receive one very nice reply, which I answered off list. It dealt with
> the cause of the exclusion (multicollinearity), which unfortunately does
> not remedy this particular situation. I am using polynomial regression
> (that is, main effects, squared effects, and interactions) and therefore
> expect multicollinearity. However, each term must be included to satisfy
> the theoretical position.
>
> I would appreciate any help that the list can provide. Thanks in advance!
> -Mark Clark
| 
