Re: Accurate Beta Coefficients in SPSS Regression Output?

[Peck, Jon]

-----Original Message-----
From: SPSSX(r) Discussion [mailto:[hidden email]] On Behalf Of Dave M
Sent: Wednesday, July 23, 2008 2:45 PM
To: [hidden email]
Subject: [SPSSX-L] Accurate Beta Coefficients in SPSS Regression Output?

I was recently told that SPSS does not provide accurate Beta coefficients in
its linear regression output;
[>>>Peck, Jon] That is utterly ridiculous.

 and, that the only way to derive accurate
Betas is to standardize all of the continuous dependent and independent
variables before entering them into the regression and then to look at the
unstandardized coefficients. I tried this a couple of times and found that,
although the two methods produced similar results, there were some small
differences in the Beta coefficients.
[>>>Peck, Jon]
Please get a new consultant.  Then learn about the nature of floating point arithmetic.

-Jon Peck

My first questions is, why are there differences?  And, my second question
is, which set of coefficients should I report?

Thanks for your help,
David

=====================
To manage your subscription to SPSSX-L, send a message to
[hidden email] (not to SPSSX-L), with no body text except the
command. To leave the list, send the command
SIGNOFF SPSSX-L
For a list of commands to manage subscriptions, send the command
INFO REFCARD

[David Hitchin]

Quoting Dave M <[hidden email]>:

> I was recently told that SPSS does not provide accurate Beta
> coefficients in its linear regression output; etc

Providing that a regression program uses decent numerical methods (and
even the Excel regression routine is not wonderful) will be good enough
for all practical purposes, even though the results may be slightly
different when you standardize the variables.

This would only matter if your input data was accurate to very many
decimal places, yet in real life most of the figures that we use are
good to only two or three decimal places, if that.

=====================
To manage your subscription to SPSSX-L, send a message to
[hidden email] (not to SPSSX-L), with no body text except the
command. To leave the list, send the command
SIGNOFF SPSSX-L
For a list of commands to manage subscriptions, send the command
INFO REFCARD

[Peck, Jon]

Please note: the OP had standardized his variables in a way that was inconsistent with what was happening in the regression.  This is not in any way an estimation precision issue.

-Jon Peck

-----Original Message-----
From: SPSSX(r) Discussion [mailto:[hidden email]] On Behalf Of David Hitchin
Sent: Thursday, July 24, 2008 4:43 AM
To: [hidden email]
Subject: Re: [SPSSX-L] Accurate Beta Coefficients in SPSS Regression Output?

Quoting Dave M <[hidden email]>:

> I was recently told that SPSS does not provide accurate Beta
> coefficients in its linear regression output; etc

Providing that a regression program uses decent numerical methods (and
even the Excel regression routine is not wonderful) will be good enough
for all practical purposes, even though the results may be slightly
different when you standardize the variables.

This would only matter if your input data was accurate to very many
decimal places, yet in real life most of the figures that we use are
good to only two or three decimal places, if that.

=====================
To manage your subscription to SPSSX-L, send a message to
[hidden email] (not to SPSSX-L), with no body text except the
command. To leave the list, send the command
SIGNOFF SPSSX-L
For a list of commands to manage subscriptions, send the command
INFO REFCARD

=====================
To manage your subscription to SPSSX-L, send a message to
[hidden email] (not to SPSSX-L), with no body text except the
command. To leave the list, send the command
SIGNOFF SPSSX-L
For a list of commands to manage subscriptions, send the command
INFO REFCARD