Reporting Multiple regression statistics

[Karen Wood]

Hi there. This is a very basis question but one where I would like to get an idea what others are doing.

In reporting results of regressional anlyses for my doctoral thesis I prefer to use standardized regression coefficients (betas) as well as reporting other measures such as change in R-squared, R-squared, adjR-squared, F, part correlations and significance levels.  However, in statistical references (e.g., Tabachnick & Fidell, 1997) and the publication manual of the Psychological association, B and SE B are used. I cannot see the benefit in reporting B (and SE B). Is there a reason why this is reported ?

Karen
:-\

[Frans Marcelissen-3]

Hi,

In my opinion B’s and beta’s contain completely different information. B’s are relevant when the dependant and the independent variables have both some natural scale. For instance when the independent variable is income (in 10.000 of euro’s) and the dependant variable is live expectancy (in years). Then a B of 2 means that people having an income that is 10.000 euro’s higher then you have a live expectancy that is 1 year higher than I have. Whether or not this is a relatively large difference can be seen in the beta.

 

-------------------------------------------

Frans Marcelissen

Email: [hidden email]

---

Van: SPSSX(r) Discussion [mailto:[hidden email]] Namens Karen Wood
Verzonden: maandag 9 maart 2009 0:00
Aan: [hidden email]
Onderwerp: Reporting Multiple regression statistics

 

Hi there. This is a very basis question but one where I would like to get an idea what others are doing.

In reporting results of regressional anlyses for my doctoral thesis I prefer to use standardized regression coefficients (betas) as well as reporting other measures such as change in R-squared, R-squared, adjR-squared, F, part correlations and significance levels.  However, in statistical references (e.g., Tabachnick & Fidell, 1997) and the publication manual of the Psychological association, B and SE B are used. I cannot see the benefit in reporting B (and SE B). Is there a reason why this is reported ?

Karen
:-\

 

[David Hitchin]

Quoting Karen Wood <[hidden email]>:
> In reporting results of regressional anlyses for my doctoral thesis
> I prefer to use standardized regression coefficients (betas) ...
> However,
> in statistical references (e.g., Tabachnick & Fidell, 1997) and the
> publication manual of the Psychological association, B and SE B are
> used. I cannot see the benefit in reporting B (and SE B). Is there a
> reason why this is reported ?

One reason why it is useful to report B is that this expresses results
in the same scales on which the variables were measured, i.e. a B of 2
means that each increase of 1 in an X value is associated with an
expected increase of 2 in the Y value.

Thus, if you can manipulate the X variables this gives you an idea of
what this would do to the value of the Y variable.

David Hitchin

=====================
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

[Trejtowicz, Mariusz]

Only to add some statistical point of view to this explanation, there
are more reasons for using Bs rather then Betas, then its interpretation
in natural scales:
 - standarized coefficients are (and unstandarized coefficients aren't)
biased by unreliability of the dependent variable: the more unreliable
is measurement of the dependent variable, the bigger is its unexplained
by the model variance
 - standarized coefficients are biased (and unstandarized coefficients
aren't) by range restrictions in independent variable: measurement of
the independent variable usually is better in the middle of the scale
(and weaker for unusually high or low values) - this influence
standarized coefficients biasing them toward zero
All this mean, that unstandarized coefficients should be chosen when
comparing results between studies, because they are more reliable then
standarized weights. However, when reporting on one study, my personal
preference is to make interpretations based on both types of
coefficients, because Betas are still very usefull to compare relative
importance of independent variables.

Regards,
Mariusz Trejtowicz


-----Original Message-----
From: SPSSX(r) Discussion [mailto:[hidden email]] On Behalf Of
David Hitchin
Sent: Monday, March 09, 2009 9:53 AM
To: [hidden email]
Subject: Re: Reporting Multiple regression statistics


Quoting Karen Wood <[hidden email]>:
> In reporting results of regressional anlyses for my doctoral thesis
> I prefer to use standardized regression coefficients (betas) ...
> However,
> in statistical references (e.g., Tabachnick & Fidell, 1997) and the
> publication manual of the Psychological association, B and SE B are
> used. I cannot see the benefit in reporting B (and SE B). Is there a
> reason why this is reported ?

One reason why it is useful to report B is that this expresses results
in the same scales on which the variables were measured, i.e. a B of 2
means that each increase of 1 in an X value is associated with an
expected increase of 2 in the Y value.

Thus, if you can manipulate the X variables this gives you an idea of
what this would do to the value of the Y variable.

David Hitchin

=====================
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