cronbach alpha for binary responses

classic Classic list List threaded Threaded
20 messages Options
Reply | Threaded
Open this post in threaded view
|

cronbach alpha for binary responses

E. Bernardo
May I know if Cronbach alpha can be used as measure of reliability for a correct/wrong response (coded 0 or 1) items?  I want to use it for each subscale (with 8 items each) of my instrument with three scales.  Thank you for your comments.
 
Eins


How many ways can you cook Adobo?
Find out all of them on Yahoo! Answers.
Reply | Threaded
Open this post in threaded view
|

Re: cronbach alpha for binary responses

Olivier van der Vet
Hello Eins,
 
You can use The Kuder-Richardson (KR20) coefficient. Via SPSS --> Is the same as Cronbach's alpha when items are dichotomous. See source.
 
Olivier
----- Original Message -----
Sent: Tuesday, December 29, 2009 8:14 AM
Subject: cronbach alpha for binary responses

May I know if Cronbach alpha can be used as measure of reliability for a correct/wrong response (coded 0 or 1) items?  I want to use it for each subscale (with 8 items each) of my instrument with three scales.  Thank you for your comments.
 
Eins


How many ways can you cook Adobo?
Find out all of them on Yahoo! Answers.


I am using the Free version of SPAMfighter.
We are a community of 6 million users fighting spam.
SPAMfighter has removed 20029 of my spam emails to date.
The Professional version does not have this message.
Reply | Threaded
Open this post in threaded view
|

Re: cronbach alpha for binary responses

Steve Simon, P.Mean Consulting
In reply to this post by E. Bernardo
Eins Bernardo wrote:

> May I know if Cronbach alpha can be used as measure of reliability for a
> correct/wrong response (coded 0 or 1) items?  I want to use it for each
> subscale (with 8 items each) of my instrument with three scales.  Thank
> you for your comments.

PASW Statistics/SPSS will allow you to calculate Cronbach's alpha for
this type of data, but the more important question is whether you can
get those results published. As far as I know, Cronbach's alpha does not
make any distributional assumptions. There may be problems with
artificial bounds being placed on Cronbach's alpha by the discrete
nature of your data. This might make it impossible for this measure to
get very close to +1, depending on the degree of skewness of your binary
variables.

I suspect that your peer reviewers would give you a hard time about this
measure, although there isn't much in the way of alternatives available
here. I wouldn't fuss about it, but I am rarely asked to be a
peer-reviewer.

It costs nothing other than a few electrons to make the calculation. So
put it in, mention the binary nature of your data as a limitation and
see what the peer-reviewers say.

I hope this helps.
--
Steve Simon, Standard Disclaimer
"The first three steps in a descriptive
data analysis, with examples in PASW/SPSS"
Thursday, January 21, 2010, 11am-noon, CST.
Details at www.pmean.com/webinars

=====================
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
Reply | Threaded
Open this post in threaded view
|

2 cases that propensity scores can't be created.

Ergul, Emel A.
Hi there:

I created estimated probabilities (propensity scores)for a group and for some
reason have two cases that scores could not be created. I checked few things but
still do not understand why.

Anyone with possible answer? What is that I'm missing?

Many thanks!

emel


The information in this e-mail is intended only for the person to whom it is
addressed. If you believe this e-mail was sent to you in error and the e-mail
contains patient information, please contact the Partners Compliance HelpLine at
http://www.partners.org/complianceline . If the e-mail was sent to you in error
but does not contain patient information, please contact the sender and properly
dispose of the e-mail.

=====================
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
Reply | Threaded
Open this post in threaded view
|

Re: 2 cases that propensity scores can't be created.

ViAnn Beadle
Perhaps it's not you who are missing, but one or more of the values for the
predictors in the model ;-)

-----Original Message-----
From: SPSSX(r) Discussion [mailto:[hidden email]] On Behalf Of
Ergul, Emel A.
Sent: Tuesday, December 29, 2009 11:00 AM
To: [hidden email]
Subject: 2 cases that propensity scores can't be created.

Hi there:

I created estimated probabilities (propensity scores)for a group and for
some reason have two cases that scores could not be created. I checked few
things but still do not understand why.

Anyone with possible answer? What is that I'm missing?

Many thanks!

emel


The information in this e-mail is intended only for the person to whom it is
addressed. If you believe this e-mail was sent to you in error and the
e-mail contains patient information, please contact the Partners Compliance
HelpLine at http://www.partners.org/complianceline . If the e-mail was sent
to you in error but does not contain patient information, please contact the
sender and properly dispose of the e-mail.

=====================
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
Reply | Threaded
Open this post in threaded view
|

Re: cronbach alpha for binary responses

Baker, Harley
In reply to this post by Steve Simon, P.Mean Consulting
Eins,

Actually, with dichotomous data, coefficient alpha is mathematically
equivalent to KR-20, which calculates the internal consistency of a scale
composed of dichotomous items. As a reviewer for a couple of journals, my
recommendation is to use coefficient alpha.

Harley

Dr. Harley Baker
Professor and Chair, Psychology Program
Sage Hall 2061
California State University Channel Islands
One University Drive
Camarillo, CA 93012

805.437.8997 (p)
805.437.8951 (f)

[hidden email]



> From: "Steve Simon, P.Mean Consulting" <[hidden email]>
> Reply-To: "Steve Simon, P.Mean Consulting" <[hidden email]>
> Date: Tue, 29 Dec 2009 11:07:15 -0600
> To: <[hidden email]>
> Subject: Re: cronbach alpha for binary responses
>
> Eins Bernardo wrote:
>
>> May I know if Cronbach alpha can be used as measure of reliability for a
>> correct/wrong response (coded 0 or 1) items?  I want to use it for each
>> subscale (with 8 items each) of my instrument with three scales.  Thank
>> you for your comments.
>
> PASW Statistics/SPSS will allow you to calculate Cronbach's alpha for
> this type of data, but the more important question is whether you can
> get those results published. As far as I know, Cronbach's alpha does not
> make any distributional assumptions. There may be problems with
> artificial bounds being placed on Cronbach's alpha by the discrete
> nature of your data. This might make it impossible for this measure to
> get very close to +1, depending on the degree of skewness of your binary
> variables.
>
> I suspect that your peer reviewers would give you a hard time about this
> measure, although there isn't much in the way of alternatives available
> here. I wouldn't fuss about it, but I am rarely asked to be a
> peer-reviewer.
>
> It costs nothing other than a few electrons to make the calculation. So
> put it in, mention the binary nature of your data as a limitation and
> see what the peer-reviewers say.
>
> I hope this helps.
> --
> Steve Simon, Standard Disclaimer
> "The first three steps in a descriptive
> data analysis, with examples in PASW/SPSS"
> Thursday, January 21, 2010, 11am-noon, CST.
> Details at www.pmean.com/webinars
>
> =====================
> 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
Reply | Threaded
Open this post in threaded view
|

Re: cronbach alpha for binary responses

Ryan
In reply to this post by E. Bernardo
Eins,

Instead of restating what was said succintly already by a poster in another group addressing a very similar question, I'll provide you with the link below.

--

http://groups.google.com/group/sci.stat.consult/browse_thread/thread/61eff485b422a0d2/8e6838b85c1a9b7f?hl=en&lnk=gst&q=cronbach%27s+dichotomous+#8e6838b85c1a9b7f

--

If for some reason the link above does not work, I've reproduced the response in quotes below:

"It depends on what you expect alpha to tell you. If you're looking
for confirmation that all the items are "measuring the same thing"
then forget alpha -- you want to factor the items, and you should
probably do it using tetrachoric correlations. A high alpha does
not imply that there is only one factor, and a low alpha does not
imply that there are many factors.

The proper use of alpha is as a quick and dirty estimate of a
lower bound for the reliability of the total score. (If you're
not computing a total score then you don't care about alpha.)
The interpretation is asymmetric: a high alpha means a high
reliability, but a low alpha does not mean a low reliability.
Alpha is a function of only two things: the number of items,
and the average correlation among them. An easy way to remember
this is  alpha/(1 - alpha) = k * rbar/(1 - rbar),
where k = the number of items, and
rbar = (average item-item covariance)/(average item variance),
which is a kind of weighted average item-item correlation. "

--

Ryan

eins wrote
May I know if Cronbach alpha can be used as measure of reliability for a correct/wrong response (coded 0 or 1) items?  I want to use it for each subscale (with 8 items each) of my instrument with three scales.  Thank you for your comments.
 
Eins


      Design your own exclusive Pingbox today! It's easy to create your personal chat space on your blogs. http://ph.messenger.yahoo.com/pingbox
Reply | Threaded
Open this post in threaded view
|

Re: cronbach alpha for binary responses

SR Millis-3
Or, alternatively, perform a Rasch analysis of the scale to assess unidimensionality of the scale items.

Scott Millis

--- On Wed, 12/30/09, rblack <[hidden email]> wrote:

> From: rblack <[hidden email]>

> Subject: Re: cronbach alpha for binary responses

> "It depends on what you expect alpha to tell you. If you're
> looking
> for confirmation that all the items are "measuring the same
> thing"
> then forget alpha -- you want to factor the items, and you
> should
> probably do it using tetrachoric correlations.

=====================
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
Reply | Threaded
Open this post in threaded view
|

Regression or correlation

Humphrey Paulie
In reply to this post by Steve Simon, P.Mean Consulting

Dear folks,

I am studying the relationship between EQ and job success (JB). I also want to know which of the subscales of EQ is a better predictor of JB.

I think I should use multiple regression.

My colleague, however, says there is no need to take the trouble and complexities of regression. We can simply correlate each subscale of EQ with JB separately and compare the correlations. He believes that the subscale that has the highest correlation with JB will also turn out to be its best predictor in regression analysis. And that regression is an unnecessary and useless statistical development. There is nothing regression does that correlation cannot do!!! I cant think of anything to justify the superiority of regression over correlation here.

Can you please help me justify it?

Cheers

Humphrey

 

Reply | Threaded
Open this post in threaded view
|

Re: Regression or correlation

Hector Maletta

Your friend is plain wrong. The different subscales are correlated among them, and each probably predicts only part of the variance in job success. Using the various subscales will cover more variance, may predict better, and may also reveal the different weight of the subscales in explaining JB. The subscale with the highest bivariate correlation may ultimately have a lower regression coefficient.

Your friend could be right if only one subscale has a high correlation and all the others have not, but that is not likely to be the case.

Finally, remember than prediction is a probabilistic affair. You cannot predict INDIVIDUAL job success: what correlation or regression may tell you is the EXPECTED job success of a GROUP of people sharing the same value of predictors. Likewise, smoking is a predictor of lung cancer, but you cannot predict lung cancer for specific individuals: some smokers live to their 90s, and some non-smokers get lung cancer anyway. You can only predict that the RELATIVE FREQUENCY of lung cancer would be much higher among smokers than non-smokers (assuming all are representative of the general population: the prediction may break down if a majority of non-smokers, and a minority of smokers, happen to be coal miners or some such, which makes them more likely to breath in toxic substances causing lung cancer). In the latter case (coal mining as a secondary predictor) including occupation in the regression equation would help separate the two as independent causes of lung cancer.

 

Hector

 


From: SPSSX(r) Discussion [mailto:[hidden email]] On Behalf Of Humphrey Paulie
Sent: 01 January 2010 05:59
To: [hidden email]
Subject: Regression or correlation

 

Dear folks,

I am studying the relationship between EQ and job success (JB). I also want to know which of the subscales of EQ is a better predictor of JB.

I think I should use multiple regression.

My colleague, however, says there is no need to take the trouble and complexities of regression. We can simply correlate each subscale of EQ with JB separately and compare the correlations. He believes that the subscale that has the highest correlation with JB will also turn out to be its best predictor in regression analysis. And that regression is an unnecessary and useless statistical development. There is nothing regression does that correlation cannot do!!! I cant think of anything to justify the superiority of regression over correlation here.

Can you please help me justify it?

Cheers

Humphrey

 

 

 

Reply | Threaded
Open this post in threaded view
|

Re: Regression or correlation

Ryan
In reply to this post by Humphrey Paulie
Humphrey,

Since another poster has addressed why multiple regression (MR) is preferred over bivariate correlations, I will skip to how one can go about answering one of your research questions. Suppose you wanted to test if the coefficient of subscale 1 is significantly different than the coefficient of subscale 2. Assuming subscales 1 and 2 are on the same scale, one could run an MR analysis including the following predictors: (1) sum of subscales 1 and 2, (2) difference of subscales 1 and 2, and (3) subscale 3. If the coefficient of the difference variable is statistically significant, then you can conclude that the coefficients of the original subscale variables, 1 and 2, are significantly different. This approach was discussed in detail in another forum--let me know if you're interested and I can send the link.

There are several assumptions to running MR, including but not limited to normally distributed residuals, homoscedasticity and linearity. If you decide to run an MR analysis, I recommend that you review these assumptions among others, and of course, check for outliers and influential points. Much can be said on the necessary diagnostics to running MR. Write back if you have specific questions on assumptions etc.

Ryan


Humphrey-6 wrote
Dear folks,
I am studying the relationship between EQ and job success (JB). I also want to know which of the subscales of EQ is a better predictor of JB.
I think I should use multiple regression.
My colleague, however, says there is no need to take the trouble and complexities of regression. We can simply correlate each subscale of EQ with JB separately and compare the correlations. He believes that the subscale that has the highest correlation with JB will also turn out to be its best predictor in regression analysis. And that regression is an unnecessary and useless statistical development. There is nothing regression does that correlation cannot do!!! I cant think of anything to justify the superiority of regression over correlation here.
Can you please help me justify it?
Cheers
Humphrey



 

Reply | Threaded
Open this post in threaded view
|

Re: Regression or correlation

Albert-Jan Roskam
In reply to this post by Hector Maletta
But aren't multiple correlation and multivariate regression basically the same thing?
I would prefer multiple regression because one has an idea of the individual amounts of explained variance (contrary to the R2), and also because the CI95% is easily obtained.

Cheers!!
Albert-Jan

~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
In the face of ambiguity, refuse the temptation to guess.
~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~

--- On Fri, 1/1/10, Hector Maletta <[hidden email]> wrote:

From: Hector Maletta <[hidden email]>
Subject: Re: [SPSSX-L] Regression or correlation
To: [hidden email]
Date: Friday, January 1, 2010, 2:51 PM

Your friend is plain wrong. The different subscales are correlated among them, and each probably predicts only part of the variance in job success. Using the various subscales will cover more variance, may predict better, and may also reveal the different weight of the subscales in explaining JB. The subscale with the highest bivariate correlation may ultimately have a lower regression coefficient.

Your friend could be right if only one subscale has a high correlation and all the others have not, but that is not likely to be the case.

Finally, remember than prediction is a probabilistic affair. You cannot predict INDIVIDUAL job success: what correlation or regression may tell you is the EXPECTED job success of a GROUP of people sharing the same value of predictors. Likewise, smoking is a predictor of lung cancer, but you cannot predict lung cancer for specific individuals: some smokers live to their 90s, and some non-smokers get lung cancer anyway. You can only predict that the RELATIVE FREQUENCY of lung cancer would be much higher among smokers than non-smokers (assuming all are representative of the general population: the prediction may break down if a majority of non-smokers, and a minority of smokers, happen to be coal miners or some such, which makes them more likely to breath in toxic substances causing lung cancer). In the latter case (coal mining as a secondary predictor) including occupation in the regression equation would help separate the two as independent causes of lung cancer.

 

Hector

 


From: SPSSX(r) Discussion [mailto:[hidden email]] On Behalf Of Humphrey Paulie
Sent: 01 January 2010 05:59
To: [hidden email]
Subject: Regression or correlation

 

Dear folks,

I am studying the relationship between EQ and job success (JB). I also want to know which of the subscales of EQ is a better predictor of JB.

I think I should use multiple regression.

My colleague, however, says there is no need to take the trouble and complexities of regression. We can simply correlate each subscale of EQ with JB separately and compare the correlations. He believes that the subscale that has the highest correlation with JB will also turn out to be its best predictor in regression analysis. And that regression is an unnecessary and useless statistical development. There is nothing regression does that correlation cannot do!!! I cant think of anything to justify the superiority of regression over correlation here.

Can you please help me justify it?

Cheers

Humphrey

 

 

 


Reply | Threaded
Open this post in threaded view
|

Re: Regression or correlation

Hector Maletta

Multiple correlation is an implication of multiple regression. There is a squared multiple correlation coefficient (R2) that arises as a result of multiple regression. It can also be obtained by combining simple correlation coefficients, obtaining partial correlation coefficients of several orders, and finally get to R. But the question referred to SIMPLE (BIVARIATE) correlation of each predictor with the outcome.

 

Hector

 


From: SPSSX(r) Discussion [mailto:[hidden email]] On Behalf Of Albert-Jan Roskam
Sent: 01 January 2010 13:59
To: [hidden email]
Subject: Re: Regression or correlation

 

But aren't multiple correlation and multivariate regression basically the same thing?
I would prefer multiple regression because one has an idea of the individual amounts of explained variance (contrary to the R2), and also because the CI95% is easily obtained.

Cheers!!
Albert-Jan

~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
In the face of ambiguity, refuse the temptation to guess.
~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~

--- On Fri, 1/1/10, Hector Maletta <[hidden email]> wrote:


From: Hector Maletta <[hidden email]>
Subject: Re: [SPSSX-L] Regression or correlation
To: [hidden email]
Date: Friday, January 1, 2010, 2:51 PM

Your friend is plain wrong. The different subscales are correlated among them, and each probably predicts only part of the variance in job success. Using the various subscales will cover more variance, may predict better, and may also reveal the different weight of the subscales in explaining JB. The subscale with the highest bivariate correlation may ultimately have a lower regression coefficient.

Your friend could be right if only one subscale has a high correlation and all the others have not, but that is not likely to be the case.

Finally, remember than prediction is a probabilistic affair. You cannot predict INDIVIDUAL job success: what correlation or regression may tell you is the EXPECTED job success of a GROUP of people sharing the same value of predictors. Likewise, smoking is a predictor of lung cancer, but you cannot predict lung cancer for specific individuals: some smokers live to their 90s, and some non-smokers get lung cancer anyway. You can only predict that the RELATIVE FREQUENCY of lung cancer would be much higher among smokers than non-smokers (assuming all are representative of the general population: the prediction may break down if a majority of non-smokers, and a minority of smokers, happen to be coal miners or some such, which makes them more likely to breath in toxic substances causing lung cancer). In the latter case (coal mining as a secondary predictor) including occupation in the regression equation would help separate the two as independent causes of lung cancer.

 

Hector

 


From: SPSSX(r) Discussion [mailto:[hidden email]] On Behalf Of Humphrey Paulie
Sent: 01 January 2010 05:59
To: [hidden email]
Subject: Regression or correlation

 

Dear folks,

I am studying the relationship between EQ and job success (JB). I also want to know which of the subscales of EQ is a better predictor of JB.

I think I should use multiple regression.

My colleague, however, says there is no need to take the trouble and complexities of regression. We can simply correlate each subscale of EQ with JB separately and compare the correlations. He believes that the subscale that has the highest correlation with JB will also turn out to be its best predictor in regression analysis. And that regression is an unnecessary and useless statistical development. There is nothing regression does that correlation cannot do!!! I cant think of anything to justify the superiority of regression over correlation here.

Can you please help me justify it?

Cheers

Humphrey

 

 

 

 

Reply | Threaded
Open this post in threaded view
|

Re: Regression or correlation

Humphrey Paulie
Well, he doesn't mean multiple correlation. He simply means a bivariate Pearson corrlations between each subscale of EG and JP and checking which correlation is higher.
We don't understand what advantage regression  has over this approach.
Cheers
Anthony

--- On Fri, 1/1/10, Hector Maletta <[hidden email]> wrote:

From: Hector Maletta <[hidden email]>
Subject: Re: Regression or correlation
To: [hidden email]
Date: Friday, January 1, 2010, 11:09 AM

Multiple correlation is an implication of multiple regression. There is a squared multiple correlation coefficient (R2) that arises as a result of multiple regression. It can also be obtained by combining simple correlation coefficients, obtaining partial correlation coefficients of several orders, and finally get to R. But the question referred to SIMPLE (BIVARIATE) correlation of each predictor with the outcome.

 

Hector

 


From: SPSSX(r) Discussion [mailto:[hidden email]] On Behalf Of Albert-Jan Roskam
Sent: 01 January 2010 13:59
To: [hidden email]
Subject: Re: Regression or correlation

 

But aren't multiple correlation and multivariate regression basically the same thing?
I would prefer multiple regression because one has an idea of the individual amounts of explained variance (contrary to the R2), and also because the CI95% is easily obtained.

Cheers!!
Albert-Jan

~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
In the face of ambiguity, refuse the temptation to guess.
~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~

--- On Fri, 1/1/10, Hector Maletta <[hidden email]> wrote:


From: Hector Maletta <[hidden email]>
Subject: Re: [SPSSX-L] Regression or correlation
To: [hidden email]
Date: Friday, January 1, 2010, 2:51 PM

Your friend is plain wrong. The different subscales are correlated among them, and each probably predicts only part of the variance in job success. Using the various subscales will cover more variance, may predict better, and may also reveal the different weight of the subscales in explaining JB. The subscale with the highest bivariate correlation may ultimately have a lower regression coefficient.

Your friend could be right if only one subscale has a high correlation and all the others have not, but that is not likely to be the case.

Finally, remember than prediction is a probabilistic affair. You cannot predict INDIVIDUAL job success: what correlation or regression may tell you is the EXPECTED job success of a GROUP of people sharing the same value of predictors. Likewise, smoking is a predictor of lung cancer, but you cannot predict lung cancer for specific individuals: some smokers live to their 90s, and some non-smokers get lung cancer anyway. You can only predict that the RELATIVE FREQUENCY of lung cancer would be much higher among smokers than non-smokers (assuming all are representative of the general population: the prediction may break down if a majority of non-smokers, and a minority of smokers, happen to be coal miners or some such, which makes them more likely to breath in toxic substances causing lung cancer). In the latter case (coal mining as a secondary predictor) including occupation in the regression equation would help separate the two as independent causes of lung cancer..

 

Hector

 


From: SPSSX(r) Discussion [mailto:[hidden email]] On Behalf Of Humphrey Paulie
Sent: 01 January 2010 05:59
To: [hidden email]..EDU
Subject: Regression or correlation

 

Dear folks,

I am studying the relationship between EQ and job success (JB). I also want to know which of the subscales of EQ is a better predictor of JB.

I think I should use multiple regression.

My colleague, however, says there is no need to take the trouble and complexities of regression. We can simply correlate each subscale of EQ with JB separately and compare the correlations. He believes that the subscale that has the highest correlation with JB will also turn out to be its best predictor in regression analysis. And that regression is an unnecessary and useless statistical development. There is nothing regression does that correlation cannot do!!! I cant think of anything to justify the superiority of regression over correlation here.

Can you please help me justify it?

Cheers

Humphrey

 

 

 

 


Reply | Threaded
Open this post in threaded view
|

Re: Regression or correlation

Humphrey Paulie
In reply to this post by Hector Maletta
Well, he doesn't mean multiple correlation. He simply means a bivariate Pearson corrlations between each subscale of EG and JP and checking which correlation is higher.
We don't understand what advantage regression  has over this approach.
Cheers
Humphrey

--- On Fri, 1/1/10, Hector Maletta <[hidden email]> wrote:

From: Hector Maletta <[hidden email]>
Subject: Re: Regression or correlation
To: [hidden email]
Date: Friday, January 1, 2010, 11:09 AM

Multiple correlation is an implication of multiple regression. There is a squared multiple correlation coefficient (R2) that arises as a result of multiple regression. It can also be obtained by combining simple correlation coefficients, obtaining partial correlation coefficients of several orders, and finally get to R. But the question referred to SIMPLE (BIVARIATE) correlation of each predictor with the outcome.

 

Hector

 


From: SPSSX(r) Discussion [mailto:[hidden email]] On Behalf Of Albert-Jan Roskam
Sent: 01 January 2010 13:59
To: [hidden email]
Subject: Re: Regression or correlation

 

But aren't multiple correlation and multivariate regression basically the same thing?
I would prefer multiple regression because one has an idea of the individual amounts of explained variance (contrary to the R2), and also because the CI95% is easily obtained.

Cheers!!
Albert-Jan

~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
In the face of ambiguity, refuse the temptation to guess.
~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~

--- On Fri, 1/1/10, Hector Maletta <[hidden email]> wrote:


From: Hector Maletta <[hidden email]>
Subject: Re: [SPSSX-L] Regression or correlation
To: [hidden email]
Date: Friday, January 1, 2010, 2:51 PM

Your friend is plain wrong. The different subscales are correlated among them, and each probably predicts only part of the variance in job success. Using the various subscales will cover more variance, may predict better, and may also reveal the different weight of the subscales in explaining JB. The subscale with the highest bivariate correlation may ultimately have a lower regression coefficient.

Your friend could be right if only one subscale has a high correlation and all the others have not, but that is not likely to be the case.

Finally, remember than prediction is a probabilistic affair. You cannot predict INDIVIDUAL job success: what correlation or regression may tell you is the EXPECTED job success of a GROUP of people sharing the same value of predictors. Likewise, smoking is a predictor of lung cancer, but you cannot predict lung cancer for specific individuals: some smokers live to their 90s, and some non-smokers get lung cancer anyway. You can only predict that the RELATIVE FREQUENCY of lung cancer would be much higher among smokers than non-smokers (assuming all are representative of the general population: the prediction may break down if a majority of non-smokers, and a minority of smokers, happen to be coal miners or some such, which makes them more likely to breath in toxic substances causing lung cancer). In the latter case (coal mining as a secondary predictor) including occupation in the regression equation would help separate the two as independent causes of lung cancer..

 

Hector

 


From: SPSSX(r) Discussion [mailto:[hidden email]] On Behalf Of Humphrey Paulie
Sent: 01 January 2010 05:59
To: [hidden email]..EDU
Subject: Regression or correlation

 

Dear folks,

I am studying the relationship between EQ and job success (JB). I also want to know which of the subscales of EQ is a better predictor of JB.

I think I should use multiple regression.

My colleague, however, says there is no need to take the trouble and complexities of regression. We can simply correlate each subscale of EQ with JB separately and compare the correlations. He believes that the subscale that has the highest correlation with JB will also turn out to be its best predictor in regression analysis. And that regression is an unnecessary and useless statistical development. There is nothing regression does that correlation cannot do!!! I cant think of anything to justify the superiority of regression over correlation here.

Can you please help me justify it?

Cheers

Humphrey

 

 

 

 


Reply | Threaded
Open this post in threaded view
|

Re: Regression or correlation

Hector Maletta

I explained the rationale for regression in your case, in a previous post in this thread. Your friend is plain wrong if he thinks he can do with simple bivariate correlations instead of regression.

Hector


From: SPSSX(r) Discussion [mailto:[hidden email]] On Behalf Of Humphrey Paulie
Sent: 02 January 2010 03:39
To: [hidden email]
Subject: Re: Regression or correlation

 

Well, he doesn't mean multiple correlation. He simply means a bivariate Pearson corrlations between each subscale of EG and JP and checking which correlation is higher.
We don't understand what advantage regression  has over this approach.
Cheers
Humphrey

--- On Fri, 1/1/10, Hector Maletta <[hidden email]> wrote:


From: Hector Maletta <[hidden email]>
Subject: Re: Regression or correlation
To: [hidden email]
Date: Friday, January 1, 2010, 11:09 AM

Multiple correlation is an implication of multiple regression. There is a squared multiple correlation coefficient (R2) that arises as a result of multiple regression. It can also be obtained by combining simple correlation coefficients, obtaining partial correlation coefficients of several orders, and finally get to R. But the question referred to SIMPLE (BIVARIATE) correlation of each predictor with the outcome.

 

Hector

 


From: SPSSX(r) Discussion [mailto:[hidden email]] On Behalf Of Albert-Jan Roskam
Sent: 01 January 2010 13:59
To: [hidden email]
Subject: Re: Regression or correlation

 

But aren't multiple correlation and multivariate regression basically the same thing?
I would prefer multiple regression because one has an idea of the individual amounts of explained variance (contrary to the R2), and also because the CI95% is easily obtained.

Cheers!!
Albert-Jan

~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
In the face of ambiguity, refuse the temptation to guess.
~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~

--- On Fri, 1/1/10, Hector Maletta <[hidden email]> wrote:


From: Hector Maletta <[hidden email]>
Subject: Re: [SPSSX-L] Regression or correlation
To: [hidden email]
Date: Friday, January 1, 2010, 2:51 PM

Your friend is plain wrong. The different subscales are correlated among them, and each probably predicts only part of the variance in job success. Using the various subscales will cover more variance, may predict better, and may also reveal the different weight of the subscales in explaining JB. The subscale with the highest bivariate correlation may ultimately have a lower regression coefficient.

Your friend could be right if only one subscale has a high correlation and all the others have not, but that is not likely to be the case.

Finally, remember than prediction is a probabilistic affair. You cannot predict INDIVIDUAL job success: what correlation or regression may tell you is the EXPECTED job success of a GROUP of people sharing the same value of predictors. Likewise, smoking is a predictor of lung cancer, but you cannot predict lung cancer for specific individuals: some smokers live to their 90s, and some non-smokers get lung cancer anyway. You can only predict that the RELATIVE FREQUENCY of lung cancer would be much higher among smokers than non-smokers (assuming all are representative of the general population: the prediction may break down if a majority of non-smokers, and a minority of smokers, happen to be coal miners or some such, which makes them more likely to breath in toxic substances causing lung cancer). In the latter case (coal mining as a secondary predictor) including occupation in the regression equation would help separate the two as independent causes of lung cancer..

 

Hector

 


From: SPSSX(r) Discussion [mailto:[hidden email]] On Behalf Of Humphrey Paulie
Sent: 01 January 2010 05:59
To: [hidden email]..EDU
Subject: Regression or correlation

 

Dear folks,

I am studying the relationship between EQ and job success (JB). I also want to know which of the subscales of EQ is a better predictor of JB.

I think I should use multiple regression.

My colleague, however, says there is no need to take the trouble and complexities of regression. We can simply correlate each subscale of EQ with JB separately and compare the correlations. He believes that the subscale that has the highest correlation with JB will also turn out to be its best predictor in regression analysis. And that regression is an unnecessary and useless statistical development. There is nothing regression does that correlation cannot do!!! I cant think of anything to justify the superiority of regression over correlation here.

Can you please help me justify it?

Cheers

Humphrey

 

 

 

 

 

Reply | Threaded
Open this post in threaded view
|

Re: Regression or correlation

Henrik Lolle
In reply to this post by Humphrey Paulie
Dear Humphrey,

I think that your text is a bit confusing. For instance, when you write
"that regression is an unnecessary and useless statistical development.
There is nothing regression does that correlation cannot do!!!", this
will be understood as a general statement about the comparison of the
two methods. But then you also write that "I cant think of anything to
justify the superiority of regression over correlation here". The word
"here" indicates that you instead only write about this example of
yours, about finding the single best predictor in a series. It also
indicates that you perhaps recognize the value of linear regression on
other occasions. And what about your study overall, conserning the
relationship between EQ and job success? Would you say that you could
answer all of your questions in this study by bivariate correlation
cofficients, and that linear regression would give you no extra
knowledge?

Best,
Henrik

Quoting Humphrey Paulie <[hidden email]>:

> Dear folks,
> I am studying the relationship between EQ and job success (JB). I
> also want to know which of the subscales of EQ is a better predictor
> of JB.
> I think I should use multiple regression.
> My colleague, however, says there is no need to take the trouble and
> complexities of regression. We can simply correlate each subscale of
> EQ with JB separately and compare the correlations. He believes that
> the subscale that has the highest correlation with JB will also turn
> out to be its best predictor in regression analysis. And that
> regression is an unnecessary and useless statistical development.
> There is nothing regression does that correlation cannot do!!! I cant
> think of anything to justify the superiority of regression over
> correlation here.
> Can you please help me justify it?
> Cheers
> Humphrey
>
>
>
>
>
>
>



************************************************************
Henrik Lolle
Department of Economics, Politics and Public Administration
Aalborg University
Fibigerstraede 1
9200 Aalborg
Phone: (+45) 99 40 81 84
************************************************************

=====================
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
Reply | Threaded
Open this post in threaded view
|

Re: cronbach alpha for binary responses

Muhammad Ghias
In reply to this post by E. Bernardo
any body who can tell me that why the value of chronbach alfa is becom low.if we have low value could we carry on our statistical analysis like logistic regression etc....or not....

Ghias
Reply | Threaded
Open this post in threaded view
|

Re: cronbach alpha for binary responses

kwame woei
Hi Muhammed,

You have probably added a new variable to the scale.

Remove the variable which causes a lower Cronbach's Alpha. The value of the Cronbach's Alpha should be higher than .60 before you use the constructed scale in statistical analyses

Regards,
Kwame

Op 18 mei 2012 om 16:20 heeft "Muhammad Ghias" <[hidden email]> het volgende geschreven:

> any body who can tell me that why the value of chronbach alfa is becom low.if
> we have low value could we carry on our statistical analysis like logistic
> regression etc....or not....
>
> Ghias
>
> --
> View this message in context: http://spssx-discussion.1045642.n5.nabble.com/cronbach-alpha-for-binary-responses-tp1086058p5711747.html
> Sent from the SPSSX Discussionw mailing list archive at Nabble.com.
>
> =====================
> 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
Reply | Threaded
Open this post in threaded view
|

Re: cronbach alpha for binary responses

Art Kendall
In reply to this post by Muhammad Ghias
Please describe what your data is like and what questions you are using the data to answer.
Art Kendall
Social Research Consultants

On 5/18/2012 1:44 AM, Muhammad Ghias wrote:
any body who can tell me that why the value of chronbach alfa is becom low.if
we have low value could we carry on our statistical analysis like logistic
regression etc....or not....

Ghias

--
View this message in context: http://spssx-discussion.1045642.n5.nabble.com/cronbach-alpha-for-binary-responses-tp1086058p5711747.html
Sent from the SPSSX Discussion mailing list archive at Nabble.com.

=====================
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
Art Kendall
Social Research Consultants