Multiple Regression: Judging the importance of independent variables: beta or t?

[Karen Wood]

Hi

 I have been using beta to determine the importance of independent variables in the regression equation (based on references). The "SPSS coach" ( version SPSS 15.0 ) suggests that  t-values be used to judge the importance of the variables. Now this is fine because in most cases  the two generally match up- BUT not always. If the independent  variables are ranked according to beta and then ranked according to  t values the order is not exactly the same.
What is the general consensus here? What are you using?

I look forward to hearing your responses.

Karen

[Mark Webb-5]

Search around topics like 'Key Driver Analysis' and you will see that the Standardized Beta is often used.

On 2010/04/09 04:40 AM, Karen Wood wrote:

Hi

 I have been using beta to determine the importance of independent variables in the regression equation (based on references). The "SPSS coach" ( version SPSS 15.0 ) suggests that  t-values be used to judge the importance of the variables. Now this is fine because in most cases  the two generally match up- BUT not always. If the independent  variables are ranked according to beta and then ranked according to  t values the order is not exactly the same.
What is the general consensus here? What are you using?

I look forward to hearing your responses.

Karen

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

[Kornbrot, Diana]

Never use inferential statistics like t, F or chi-square to assess importance. These statistics, if singificant. increase with N, so large sampleswill always have large t.
You need effect sizes and beta is good as it tells you slope of line in terms of standard deviation estimate
Unstandardized b may be even better as it is in real world metric. E,g if you regress salary on years of education, then b tells you how mnuch salary increases for each year of education. This b should enable a potential student to evluate the ‘imprtance’ of edcuation
SPSS coach should be ashamed of themselves. I am copying to IB./SPSS discussion list, so that they can alter their coach appropriately and apologize. NB the coach may have already been changed in later versions.
The regression coefficients, b, beta do NOT change with N, they are simply more accurately estimated with large N. Whichever you choose, also include confidence levels [there are options in SPSS for these]
Best

Diana


On 09/04/2010 03:40, "Karen Wood" <k.wood@...> wrote:

Hi

I have been using beta to determine the importance of independent variables in the regression equation (based on references). The "SPSS coach" ( version SPSS 15.0 ) suggests that  t-values be used to judge the importance of the variables. Now this is fine because in most cases  the two generally match up- BUT not always. If the independent  variables are ranked according to beta and then ranked according to  t values the order is not exactly the same.
What is the general consensus here? What are you using?

I look forward to hearing your responses.

Karen

---

Professor Diana Kornbrot
  email:  d.e.kornbrot@...    
   web:    http://web.mac.com/kornbrot/iweb/KornbrotHome.html
Work
School of Psychology
University of Hertfordshire
College Lane, Hatfield, Hertfordshire AL10 9AB, UK
    voice:     [phone]
    mobile:   [phone]
    fax          [phone]
Home
19 Elmhurst Avenue
London N2 0LT, UK
   landline: [phone]
   mobile:   [phone]
   fax:         [phone]

---

[Mbaye Fall Diallo]

Dear all,

Can you post some good references or tutorials (online or not) on multiple regression? 

Thanks in advance for your help.

Mbaye, 

---

Date: Fri, 9 Apr 2010 08:00:32 +0100
From: [hidden email]
Subject: Re: Multiple Regression: Judging the importance of independent variables: beta or t?
To: [hidden email]

Never use inferential statistics like t, F or chi-square to assess importance. These statistics, if singificant. increase with N, so large sampleswill always have large t.
You need effect sizes and beta is good as it tells you slope of line in terms of standard deviation estimate
Unstandardized b may be even better as it is in real world metric. E,g if you regress salary on years of education, then b tells you how mnuch salary increases for each year of education. This b should enable a potential student to evluate the ‘imprtance’ of edcuation
SPSS coach should be ashamed of themselves. I am copying to IB./SPSS discussion list, so that they can alter their coach appropriately and apologize. NB the coach may have already been changed in later versions.
The regression coefficients, b, beta do NOT change with N, they are simply more accurately estimated with large N. Whichever you choose, also include confidence levels [there are options in SPSS for these]
Best

Diana


On 09/04/2010 03:40, "Karen Wood" <k.wood@...> wrote:

Hi

I have been using beta to determine the importance of independent variables in the regression equation (based on references). The "SPSS coach" ( version SPSS 15.0 ) suggests that  t-values be used to judge the importance of the variables. Now this is fine because in most cases  the two generally match up- BUT not always. If the independent  variables are ranked according to beta and then ranked according to  t values the order is not exactly the same.
What is the general consensus here? What are you using?

I look forward to hearing your responses.

Karen

---

Professor Diana Kornbrot
  email:  d.e.kornbrot@...    
   web:    http://web.mac.com/kornbrot/iweb/KornbrotHome.html
Work
School of Psychology
University of Hertfordshire
College Lane, Hatfield, Hertfordshire AL10 9AB, UK
    voice:     [phone]
    mobile:   [phone]
    fax          [phone]
Home
19 Elmhurst Avenue
London N2 0LT, UK
   landline: [phone]
   mobile:   [phone]
   fax:         [phone]

---

---

Envie de naviguer sur Internet sans laisser de trace? La solution avec Internet Explorer 8

[statisticsdoc]

Karen,

 

Do not use t-values to judge the importance of independent variables.  T is a statistic used to evaluate the null hypothesis, and to make decisions about statistical inferences regarding the null hypothesis. T is not an indicator of the contribution of a variable to the prediction of the dependent variable.  The standardized beta is an indicator of the unique contribution of a specific variable.  Please note that beta values are context-dependent:  the meaning and value of the beta for a specific predictor depends on the other predictor variables in the equation.  For example, if two or more predictors are highly correlated with one another, their beta weights are likely to be smaller because each makes a smaller unique contribution to the prediction equation. 

 

Best,

 

Steve Brand

 

www.StatisticsDoc.com 

-----Original Message-----
From: SPSSX(r) Discussion [mailto:[hidden email]]On Behalf Of Karen Wood
Sent: Thursday, April 08, 2010 10:40 PM
To: [hidden email]
Subject: Multiple Regression: Judging the importance of independent variables: beta or t?

Hi

 I have been using beta to determine the importance of independent variables in the regression equation (based on references). The "SPSS coach" ( version SPSS 15.0 ) suggests that  t-values be used to judge the importance of the variables. Now this is fine because in most cases  the two generally match up- BUT not always. If the independent  variables are ranked according to beta and then ranked according to  t values the order is not exactly the same.
What is the general consensus here? What are you using?

I look forward to hearing your responses.

Karen

[vlad simion]

Hi Karen,

look for this reference: "History and Use of Relative Importance
Indices in Organizational Research - Organizational Research Methods,
Vol. 7, No. 3, 238-257 (2004)"
and look here for some syntax in spss on relative importance analysis
http://www1.psych.purdue.edu/~jlebreto/relative.htm

hth,
Vlad


On Fri, Apr 9, 2010 at 5:40 AM, Karen Wood <[hidden email]> wrote:

>
>
> Hi
>
>  I have been using beta to determine the importance of independent variables
> in the regression equation (based on references). The "SPSS coach" ( version
> SPSS 15.0 ) suggests that  t-values be used to judge the importance of the
> variables. Now this is fine because in most cases  the two generally match
> up- BUT not always. If the independent  variables are ranked according to
> beta and then ranked according to  t values the order is not exactly the
> same.
> What is the general consensus here? What are you using?
>
> I look forward to hearing your responses.
>
> Karen
>
>
>

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

[Bruce Weaver]

Let me add a cautionary note about standardized regression coefficients.  The following is something I posted in response to a similar question in the MedStats group in August 2009.

--- start of post to MedStats ---

In his book "Applied Regression Analysis and Generalized Linear Models" (2008, Sage), John Fox is very cautious about the use of standardized regression coefficients.  He gives this interesting example.  When two variables are measured on the same scale (e.g., years of education, and years of employment), then relative impact of the two can be compared directly.  But suppose those two variables differ substantially in the amount of spread.  In that case, comparison of the standardized regression coefficients would likely
yield a very different story than comparison of the raw regression coefficients.  Fox then says:

"If expressing coefficients relative to a measure of spread potentially distorts their comparison when two explanatory variables are commensurable [i.e., measured on the same scale], then why should the procedure magically allow to compare coefficients [for variables] that are measured in different units?" (p. 95)

Good question!

--- end of post to MedStats ---

kornbrot wrote

Never use inferential statistics like t, F or chi-square to assess
importance. These statistics, if singificant. increase with N, so large
sampleswill always have large t.
You need effect sizes and beta is good as it tells you slope of line in
terms of standard deviation estimate
Unstandardized b may be even better as it is in real world metric. E,g if
you regress salary on years of education, then b tells you how mnuch salary
increases for each year of education. This b should enable a potential
student to evluate the Œimprtance¹ of edcuation
SPSS coach should be ashamed of themselves. I am copying to IB./SPSS
discussion list, so that they can alter their coach appropriately and
apologize. NB the coach may have already been changed in later versions.
The regression coefficients, b, beta do NOT change with N, they are simply
more accurately estimated with large N. Whichever you choose, also include
confidence levels [there are options in SPSS for these]
Best

Diana


On 09/04/2010 03:40, "Karen Wood" <[e-mail]> wrote:

>
>
> Hi
>
>  I have been using beta to determine the importance of independent variables
> in the regression equation (based on references). The "SPSS coach" ( version
> SPSS 15.0 ) suggests that  t-values be used to judge the importance of the
> variables. Now this is fine because in most cases  the two generally match up-
> BUT not always. If the independent  variables are ranked according to beta and
> then ranked according to  t values the order is not exactly the same.
> What is the general consensus here? What are you using?
>
> I look forward to hearing your responses.
>
> Karen
>
>
>



Professor Diana Kornbrot
   email:  [e-mail]
   web:    http://web.mac.com/kornbrot/iweb/KornbrotHome.html
Work
School of Psychology
University of Hertfordshire
College Lane, Hatfield, Hertfordshire AL10 9AB, UK
    voice:     [phone]
    mobile:   [phone]
    fax          [phone]
Home
19 Elmhurst Avenue
London N2 0LT, UK
   landline: [phone]
   mobile:   [phone]
   fax:         [phone]

[Anthony Babinec]

To add to what has already been said, William Kruskal
has a couple articles on the topic of computing
relative importance by averaging over orderings of
variables in regression models. These appeared in
The American Statistician in 1987 and 1989. In situations
in which you encounter multicollinearity, such as in
assessing drivers of customer satisfaction, looking
at standardized betas or significance levels is problematic.
Joseph Retzer has some papers on this.

Tony Babinec
[hidden email]

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

[statisticsdoc]

Anthony raises an important point here.  Averaging Over Orderings was once very computationally demanding and labor intensive, but the Averaging Over Orderings approach can now be implemented by using the relaimpo package in R.
Best,
Steve Brand
www.StatisticsDoc.com

-----Original Message-----
From: Anthony Babinec <[hidden email]>
Date:         Fri, 9 Apr 2010 09:57:48
To: <[hidden email]>
Subject:      Re: Multiple Regression: Judging the importance of independent
              variables: beta or t?

To add to what has already been said, William Kruskal
has a couple articles on the topic of computing
relative importance by averaging over orderings of
variables in regression models. These appeared in
The American Statistician in 1987 and 1989. In situations
in which you encounter multicollinearity, such as in
assessing drivers of customer satisfaction, looking
at standardized betas or significance levels is problematic.
Joseph Retzer has some papers on this.

Tony Babinec
[hidden email]

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

[Bruce Weaver]

Have you seen David Garson's (online) notes?  You should find them easily via Google.

Mbaye Fall Diallo wrote

Dear all,
Can you post some good references or tutorials (online or not) on multiple regression?
Thanks in advance for your help.
Mbaye,


Date: Fri, 9 Apr 2010 08:00:32 +0100
From: [e-mail]
Subject: Re: Multiple Regression: Judging the importance of independent              variables: beta or t?
To: [e-mail]





Re: Multiple Regression: Judging the importance of independent variables: beta or t?


Never use inferential statistics like t, F or chi-square to assess importance. These statistics, if singificant. increase with N, so large sampleswill always have large t.

You need effect sizes and beta is good as it tells you slope of line in terms of standard deviation estimate

Unstandardized b may be even better as it is in real world metric. E,g if you regress salary on years of education, then b tells you how mnuch salary increases for each year of education. This b should enable a potential student to evluate the ‘imprtance’ of edcuation

SPSS coach should be ashamed of themselves. I am copying to IB./SPSS discussion list, so that they can alter their coach appropriately and apologize. NB the coach may have already been changed in later versions.

The regression coefficients, b, beta do NOT change with N, they are simply more accurately estimated with large N. Whichever you choose, also include confidence levels [there are options in SPSS for these]

Best



Diana





On 09/04/2010 03:40, "Karen Wood" <[e-mail]> wrote:







Hi



 I have been using beta to determine the importance of independent variables in the regression equation (based on references). The "SPSS coach" ( version SPSS 15.0 ) suggests that  t-values be used to judge the importance of the variables. Now this is fine because in most cases  the two generally match up- BUT not always. If the independent  variables are ranked according to beta and then ranked according to  t values the order is not exactly the same.

What is the general consensus here? What are you using?



I look forward to hearing your responses.



Karen











Professor Diana Kornbrot

   email:  [e-mail]    

   web:    http://web.mac.com/kornbrot/iweb/KornbrotHome.html 

Work

School of Psychology

University of Hertfordshire

College Lane, Hatfield, Hertfordshire AL10 9AB, UK

    voice:     [phone]

    mobile:   [phone]

    fax          [phone]

Home

19 Elmhurst Avenue

London N2 0LT, UK

   landline: [phone]

   mobile:   [phone]

   fax:         [phone]









                                         
_________________________________________________________________
Découvrez comment SURFER DISCRETEMENT sur un site de rencontres !
http://clk.atdmt.com/FRM/go/206608211/direct/01/