Hello all.

I am a novice SPSS user, and cannot find a solution to a problem.
I am interested in finding a way to test if coefficient are
statistically different from each other. E.g. y = B1x1+ B2x2 + B3x3
+ e. And the hypothesis test I want to perform is if B1-B2=0? In
Stata there is syntax to directly test this within the regression
run. Is the same true in SPSS?

The only mention I could find online is:
"Unfortunately, unlike Stata, SPSS does not provide a convenient way
to test hypotheses like â1 = â2, e.g. the effects of education and job
experience are equal." http://www.nd.edu/~rwilliam/stats1/OLS-SPSS.pdf

I will ultimately have to perform 100s of these tests and am hopefuly
I can write the tests directly into the regression syntax.

Any ideas on where I can look for samples of syntax to do this (and
other creative hypothisis testing of regression results)?

Thanks,

Ben

On Jul 31, 10:53 am, BenH34 <benho... at (no spam) earthlink.net> wrote:





Hello all.

I am a novice SPSS user, and cannot find a solution to a problem.
I am interested in finding a way to test if coefficient are
statistically different from each other.  E.g. y = B1x1+ B2x2 + B3x3
+ e.  And the hypothesis test I want to perform is if B1-B2=0?  In
Stata there is syntax to directly test this within the regression
run.  Is the same true in SPSS?

The only mention I could find online is:
"Unfortunately, unlike Stata, SPSS does not provide a convenient way
to test hypotheses like â1 = â2, e.g. the effects of education and job
experience are equal."  http://www.nd.edu/~rwilliam/stats1/OLS-SPSS.pdf

I will ultimately have to perform 100s of these tests and am hopefuly
I can write the tests directly into the regression syntax.

Any ideas on where I can look for samples of syntax to do this (and
other creative hypothisis testing of regression results)?

Thanks,

Ben

Run the regression with x1 and x2 replaced by their sum and
difference, x1+x2 and x1-x2. Then the test of the coefficient
of the difference is equivalent to a test of the difference
between B1 and B2.- Hide quoted text -

- Show quoted text -

On Jul 31, 7:37 pm, Ray Koopman <koop... at (no spam) sfu.ca> wrote:





On Jul 31, 10:53 am, BenH34 <benho... at (no spam) earthlink.net> wrote:

Hello all.

I am a novice SPSS user, and cannot find a solution to a problem.
I am interested in finding a way to test if coefficient are
statistically different from each other.  E.g. y = B1x1+ B2x2 + B3x3
+ e.  And the hypothesis test I want to perform is if B1-B2=0?  In
Stata there is syntax to directly test this within the regression
run.  Is the same true in SPSS?

The only mention I could find online is:
"Unfortunately, unlike Stata, SPSS does not provide a convenient way
to test hypotheses like â1 = â2, e.g. the effects of education and job
experience are equal."  http://www.nd.edu/~rwilliam/stats1/OLS-SPSS..pdf

I will ultimately have to perform 100s of these tests and am hopefuly
I can write the tests directly into the regression syntax.

Any ideas on where I can look for samples of syntax to do this (and
other creative hypothisis testing of regression results)?

Thanks,

Ben

Run the regression with x1 and x2 replaced by their sum and
difference, x1+x2 and x1-x2. Then the test of the coefficient
of the difference is equivalent to a test of the difference
between B1 and B2.- Hide quoted text -

- Show quoted text -

Hi Ray,

I'm always intrigued by your responses! I hope it's okay, but I have a
couple follow up questions. Just to make sure I understand, if you
wanted to know if any of the three "partial regression coefficients"
were different from each other, would you run three separate linear
regression analyses?...

Y= b0 + b1(x1+x2) + b2(x1-x2)

Y= b0+ b3(x2+x3) + b4(x2-x3)

Y= b0+ b5(x1+x3) + b6(x1-x3)

where

b2 reflects difference between partial weights for x1 and x2
b4 reflects difference between partial weights for x1 and x3
b6 reflects difference between partial weights for x2 and x3

This approach seems problematic to me because you're not taking into
consideration/statistically controlling for the third variable when
testing the differences.

Also, if you have the time, would you mind explaining why you need to
include the x1+x2, x2+x3, and x1+x3?

Sorry for so many questions. I completely understand if you're too
busy to respond.

Thank you,

Ryan- Hide quoted text -

- Show quoted text -

On Jul 31, 10:53 am, BenH34 <benho... at (no spam) earthlink.net> wrote:
Hello all.

I am a novice SPSS user, and cannot find a solution to a problem.
I am interested in finding a way to test if coefficient are
statistically different from each other. E.g. y = B1x1+ B2x2 + B3x3
+ e. And the hypothesis test I want to perform is if B1-B2=0? In
Stata there is syntax to directly test this within the regression
run. Is the same true in SPSS?

The only mention I could find online is:
"Unfortunately, unlike Stata, SPSS does not provide a convenient way
to test hypotheses like â1 = â2, e.g. the effects of education and job
experience are equal." http://www.nd.edu/~rwilliam/stats1/OLS-SPSS.pdf

I will ultimately have to perform 100s of these tests and am hopefuly
I can write the tests directly into the regression syntax.

Any ideas on where I can look for samples of syntax to do this (and
other creative hypothisis testing of regression results)?

Thanks,

Ben

Run the regression with x1 and x2 replaced by their sum and
difference, x1+x2 and x1-x2. Then the test of the coefficient
of the difference is equivalent to a test of the difference
between B1 and B2.

On Jul 31, 7:37 pm, Ray Koopman <koop... at (no spam) sfu.ca> wrote:
On Jul 31, 10:53 am, BenH34 <benho... at (no spam) earthlink.net> wrote:

Hello all.

I am a novice SPSS user, and cannot find a solution to a problem.
I am interested in finding a way to test if coefficient are
statistically different from each other. E.g. y = B1x1+ B2x2 + B3x3
+ e. And the hypothesis test I want to perform is if B1-B2=0? In
Stata there is syntax to directly test this within the regression
run. Is the same true in SPSS?

The only mention I could find online is:
"Unfortunately, unlike Stata, SPSS does not provide a convenient way
to test hypotheses like â1 = â2, e.g. the effects of education and job
experience are equal." http://www.nd.edu/~rwilliam/stats1/OLS-SPSS.pdf

I will ultimately have to perform 100s of these tests and am hopefuly
I can write the tests directly into the regression syntax.

Any ideas on where I can look for samples of syntax to do this (and
other creative hypothisis testing of regression results)?

Thanks,

Ben

Run the regression with x1 and x2 replaced by their sum and
difference, x1+x2 and x1-x2. Then the test of the coefficient
of the difference is equivalent to a test of the difference
between B1 and B2.

Hi Ray,

I'm always intrigued by your responses! I hope it's okay, but I have a
couple follow up questions. Just to make sure I understand, if you
wanted to know if any of the three "partial regression coefficients"
were different from each other, would you run three separate linear
regression analyses?...

Y= b0 + b1(x1+x2) + b2(x1-x2)

Y= b0+ b3(x2+x3) + b4(x2-x3)

Y= b0+ b5(x1+x3) + b6(x1-x3)

where

b2 reflects difference between partial weights for x1 and x2
b4 reflects difference between partial weights for x1 and x3
b6 reflects difference between partial weights for x2 and x3

This approach seems problematic to me because you're not taking into
consideration/statistically controlling for the third variable when
testing the differences.

Also, if you have the time, would you mind explaining why you need to
include the x1+x2, x2+x3, and x1+x3?

Sorry for so many questions. I completely understand if you're too
busy to respond.

Thank you,

Ryan

On Jul 31, 6:47 pm, Ryan <Ryan.Andrew.Bl... at (no spam) gmail.com> wrote:



On Jul 31, 7:37 pm, Ray Koopman <koop... at (no spam) sfu.ca> wrote:
On Jul 31, 10:53 am, BenH34 <benho... at (no spam) earthlink.net> wrote:

Hello all.

I am a novice SPSS user, and cannot find a solution to a problem.
I am interested in finding a way to test if coefficient are
statistically different from each other.  E.g. y = B1x1+ B2x2 + B3x3
+ e.  And the hypothesis test I want to perform is if B1-B2=0?  In
Stata there is syntax to directly test this within the regression
run.  Is the same true in SPSS?

The only mention I could find online is:
"Unfortunately, unlike Stata, SPSS does not provide a convenient way
to test hypotheses like â1 = â2, e.g. the effects of education and job
experience are equal."  http://www.nd.edu/~rwilliam/stats1/OLS-SPSS..pdf

I will ultimately have to perform 100s of these tests and am hopefuly
I can write the tests directly into the regression syntax.

Any ideas on where I can look for samples of syntax to do this (and
other creative hypothisis testing of regression results)?

Thanks,

Ben

Run the regression with x1 and x2 replaced by their sum and
difference, x1+x2 and x1-x2. Then the test of the coefficient
of the difference is equivalent to a test of the difference
between B1 and B2.

Hi Ray,

I'm always intrigued by your responses! I hope it's okay, but I have a
couple follow up questions. Just to make sure I understand, if you
wanted to know if any of the three "partial regression coefficients"
were different from each other, would you run three separate linear
regression analyses?...

Y= b0 + b1(x1+x2) + b2(x1-x2)

Y= b0+ b3(x2+x3) + b4(x2-x3)

Y= b0+ b5(x1+x3) + b6(x1-x3)

where

b2 reflects difference between partial weights for x1 and x2
b4 reflects difference between partial weights for x1 and x3
b6 reflects difference between partial weights for x2 and x3

This approach seems problematic to me because you're not taking into
consideration/statistically controlling for the third variable when
testing the differences.

Also, if you have the time, would you mind explaining why you need to
include the x1+x2, x2+x3, and x1+x3?

Sorry for so many questions. I completely understand if you're too
busy to respond.

Thank you,

Ryan

No, I'd do a different transformation, replacing {x1, x2, x3} with
{x1', x2' x3'} = {x1+x2+x3, 2x1-x2-x3, x2-x3}. Do it in two steps:
first put in x1', then add x2' and x3'. If the second step increases
R^2 significantly then you reject H: b1 = b2 = b3.

It may help to consider the general question: in a p-predictor model,
how can you test H: b1 = ... = bp? One way is a generalization of
the p=3 case above, the other is a generalization of the solution
Bruce posted in the SPSS-only version of this discussion. Of course,
the two solutions give the same answer, so they're really only two
different ways of thinking about the problem.

On Fri, 1 Aug 2008 08:27:02 -0700 (PDT), pcapp... at (no spam) yahoo.com wrote:
On Aug 1, 3:43 am, Ray Koopman <koop... at (no spam) sfu.ca> wrote:
On Jul 31, 6:47 pm, Ryan <Ryan.Andrew.Bl... at (no spam) gmail.com> wrote:

On Jul 31, 7:37 pm, Ray Koopman <koop... at (no spam) sfu.ca> wrote:
On Jul 31, 10:53 am, BenH34 <benho... at (no spam) earthlink.net> wrote:

Hello all.

I am a novice SPSS user, and cannot find a solution to a problem.
I am interested in finding a way to test if coefficient are
statistically different from each other.  E.g. y = B1x1+ B2x2 + B3x3
+ e.  And the hypothesis test I want to perform is if B1-B2=0?  In
Stata there is syntax to directly test this within the regression
run.  Is the same true in SPSS?

The only mention I could find online is:
"Unfortunately, unlike Stata, SPSS does not provide a convenient way
to test hypotheses like â1 = â2, e.g. the effects of education and job
experience are equal."  http://www.nd.edu/~rwilliam/stats1/OLS-SPSS.pdf

I will ultimately have to perform 100s of these tests and am hopefuly
I can write the tests directly into the regression syntax.

Any ideas on where I can look for samples of syntax to do this (and
other creative hypothisis testing of regression results)?

Thanks,

Ben

Run the regression with x1 and x2 replaced by their sum and
difference, x1+x2 and x1-x2. Then the test of the coefficient
of the difference is equivalent to a test of the difference
between B1 and B2.

Hi Ray,

I'm always intrigued by your responses! I hope it's okay, but I have a
couple follow up questions. Just to make sure I understand, if you
wanted to know if any of the three "partial regression coefficients"
were different from each other, would you run three separate linear
regression analyses?...

Y= b0 + b1(x1+x2) + b2(x1-x2)

Y= b0+ b3(x2+x3) + b4(x2-x3)

Y= b0+ b5(x1+x3) + b6(x1-x3)

where

b2 reflects difference between partial weights for x1 and x2
b4 reflects difference between partial weights for x1 and x3
b6 reflects difference between partial weights for x2 and x3

This approach seems problematic to me because you're not taking into
consideration/statistically controlling for the third variable when
testing the differences.

Also, if you have the time, would you mind explaining why you need to
include the x1+x2, x2+x3, and x1+x3?

Sorry for so many questions. I completely understand if you're too
busy to respond.

Thank you,

Ryan

No, I'd do a different transformation, replacing {x1, x2, x3} with
{x1', x2' x3'} = {x1+x2+x3, 2x1-x2-x3, x2-x3}. Do it in two steps:
first put in x1', then add x2' and x3'. If the second step increases
R^2 significantly then you reject H: b1 = b2 = b3.

It may help to consider the general question: in a p-predictor model,
how can you test H: b1 = ... = bp? One way is a generalization of
the p=3 case above, the other is a generalization of the solution
Bruce posted in the SPSS-only version of this discussion. Of course,
the two solutions give the same answer, so they're really only two
different ways of thinking about the problem.

All very elegant ways to do it, but if the goal is to have the
parameter estimates of x1, x2, and x3 as part of the output, you would
have to do the math outside of SPSS to get the true parameter
estimates.  I too am new to SPSS, but know in programs like SAS you
can take the output of the regression as a data set.  Within a data
step (again SAS parlance for a set of syntax that manipulates the data
set), you should be able to construct the necessary t-statistic with
the parameter estimates, their variance and covariances (provided you
can output these as well from the regression into a data set) and ask
the package for the p-value of this statistic.  Is this possible to do
in SPSS?

If you want to test b1=b2=b3, you will need to do an F-test. A series
of t-tests won't do it.
-Dick Startz

On Jul 31, 6:47 pm, Ryan <Ryan.Andrew.Bl... at (no spam) gmail.com> wrote:





On Jul 31, 7:37 pm, Ray Koopman <koop... at (no spam) sfu.ca> wrote:
On Jul 31, 10:53 am, BenH34 <benho... at (no spam) earthlink.net> wrote:

Hello all.

I am a novice SPSS user, and cannot find a solution to a problem.
I am interested in finding a way to test if coefficient are
statistically different from each other.  E.g. y = B1x1+ B2x2 + B3x3
+ e.  And the hypothesis test I want to perform is if B1-B2=0?  In
Stata there is syntax to directly test this within the regression
run.  Is the same true in SPSS?

The only mention I could find online is:
"Unfortunately, unlike Stata, SPSS does not provide a convenient way
to test hypotheses like â1 = â2, e.g. the effects of education and job
experience are equal."  http://www.nd.edu/~rwilliam/stats1/OLS-SPSS..pdf

I will ultimately have to perform 100s of these tests and am hopefuly
I can write the tests directly into the regression syntax.

Any ideas on where I can look for samples of syntax to do this (and
other creative hypothisis testing of regression results)?

Thanks,

Ben

Run the regression with x1 and x2 replaced by their sum and
difference, x1+x2 and x1-x2. Then the test of the coefficient
of the difference is equivalent to a test of the difference
between B1 and B2.

Hi Ray,

I'm always intrigued by your responses! I hope it's okay, but I have a
couple follow up questions. Just to make sure I understand, if you
wanted to know if any of the three "partial regression coefficients"
were different from each other, would you run three separate linear
regression analyses?...

Y= b0 + b1(x1+x2) + b2(x1-x2)

Y= b0+ b3(x2+x3) + b4(x2-x3)

Y= b0+ b5(x1+x3) + b6(x1-x3)

where

b2 reflects difference between partial weights for x1 and x2
b4 reflects difference between partial weights for x1 and x3
b6 reflects difference between partial weights for x2 and x3

This approach seems problematic to me because you're not taking into
consideration/statistically controlling for the third variable when
testing the differences.

Also, if you have the time, would you mind explaining why you need to
include the x1+x2, x2+x3, and x1+x3?

Sorry for so many questions. I completely understand if you're too
busy to respond.

Thank you,

Ryan

No, I'd do a different transformation, replacing {x1, x2, x3} with
{x1', x2' x3'} = {x1+x2+x3, 2x1-x2-x3, x2-x3}. Do it in two steps:
first put in x1', then add x2' and x3'. If the second step increases
R^2 significantly then you reject H: b1 = b2 = b3.

It may help to consider the general question: in a p-predictor model,
how can you test H: b1 = ... = bp? One way is a generalization of
the p=3 case above, the other is a generalization of the solution
Bruce posted in the SPSS-only version of this discussion. Of course,
the two solutions give the same answer, so they're really only two
different ways of thinking about the problem.- Hide quoted text -

- Show quoted text -

On Aug 1, 3:43 am, Ray Koopman <koop... at (no spam) sfu.ca> wrote:
On Jul 31, 6:47 pm, Ryan <Ryan.Andrew.Bl... at (no spam) gmail.com> wrote:



On Jul 31, 7:37 pm, Ray Koopman <koop... at (no spam) sfu.ca> wrote:
On Jul 31, 10:53 am, BenH34 <benho... at (no spam) earthlink.net> wrote:

Hello all.

I am a novice SPSS user, and cannot find a solution to a problem.
I am interested in finding a way to test if coefficient are
statistically different from each other.  E.g. y = B1x1+ B2x2 + B3x3
+ e.  And the hypothesis test I want to perform is if B1-B2=0?  In
Stata there is syntax to directly test this within the regression
run.  Is the same true in SPSS?

The only mention I could find online is:
"Unfortunately, unlike Stata, SPSS does not provide a convenient way
to test hypotheses like â1 = â2, e.g. the effects of education and job
experience are equal."  http://www.nd.edu/~rwilliam/stats1/OLS-SPSS.pdf

I will ultimately have to perform 100s of these tests and am hopefuly
I can write the tests directly into the regression syntax.

Any ideas on where I can look for samples of syntax to do this (and
other creative hypothisis testing of regression results)?

Thanks,

Ben

Run the regression with x1 and x2 replaced by their sum and
difference, x1+x2 and x1-x2. Then the test of the coefficient
of the difference is equivalent to a test of the difference
between B1 and B2.

Hi Ray,

I'm always intrigued by your responses! I hope it's okay, but I have a
couple follow up questions. Just to make sure I understand, if you
wanted to know if any of the three "partial regression coefficients"
were different from each other, would you run three separate linear
regression analyses?...

Y= b0 + b1(x1+x2) + b2(x1-x2)

Y= b0+ b3(x2+x3) + b4(x2-x3)

Y= b0+ b5(x1+x3) + b6(x1-x3)

where

b2 reflects difference between partial weights for x1 and x2
b4 reflects difference between partial weights for x1 and x3
b6 reflects difference between partial weights for x2 and x3

This approach seems problematic to me because you're not taking into
consideration/statistically controlling for the third variable when
testing the differences.

Also, if you have the time, would you mind explaining why you need to
include the x1+x2, x2+x3, and x1+x3?

Sorry for so many questions. I completely understand if you're too
busy to respond.

Thank you,

Ryan

No, I'd do a different transformation, replacing {x1, x2, x3} with
{x1', x2' x3'} = {x1+x2+x3, 2x1-x2-x3, x2-x3}. Do it in two steps:
first put in x1', then add x2' and x3'. If the second step increases
R^2 significantly then you reject H: b1 = b2 = b3.

It may help to consider the general question: in a p-predictor model,
how can you test H: b1 = ... = bp? One way is a generalization of
the p=3 case above, the other is a generalization of the solution
Bruce posted in the SPSS-only version of this discussion. Of course,
the two solutions give the same answer, so they're really only two
different ways of thinking about the problem.

All very elegant ways to do it, but if the goal is to have the
parameter estimates of x1, x2, and x3 as part of the output, you would
have to do the math outside of SPSS to get the true parameter
estimates. I too am new to SPSS, but know in programs like SAS you
can take the output of the regression as a data set. Within a data
step (again SAS parlance for a set of syntax that manipulates the data
set), you should be able to construct the necessary t-statistic with
the parameter estimates, their variance and covariances (provided you
can output these as well from the regression into a data set) and ask
the package for the p-value of this statistic. Is this possible to do
in SPSS?

On Aug 1, 11:46 am, Richard Startz <richardstar... at (no spam) comcast.net> wrote:
On Fri, 1 Aug 2008 08:27:02 -0700 (PDT), pcapp... at (no spam) yahoo.com wrote:
On Aug 1, 3:43 am, Ray Koopman <koop... at (no spam) sfu.ca> wrote:
On Jul 31, 6:47 pm, Ryan <Ryan.Andrew.Bl... at (no spam) gmail.com> wrote:

On Jul 31, 7:37 pm, Ray Koopman <koop... at (no spam) sfu.ca> wrote:
On Jul 31, 10:53 am, BenH34 <benho... at (no spam) earthlink.net> wrote:

Hello all.

I am a novice SPSS user, and cannot find a solution to a problem.
I am interested in finding a way to test if coefficient are
statistically different from each other.  E.g. y = B1x1+ B2x2 + B3x3
+ e.  And the hypothesis test I want to perform is if B1-B2=0?  In
Stata there is syntax to directly test this within the regression
run.  Is the same true in SPSS?

The only mention I could find online is:
"Unfortunately, unlike Stata, SPSS does not provide a convenient way
to test hypotheses like â1 = â2, e.g. the effects of education and job
experience are equal."  http://www.nd.edu/~rwilliam/stats1/OLS-SPSS.pdf

I will ultimately have to perform 100s of these tests and am hopefuly
I can write the tests directly into the regression syntax.

Any ideas on where I can look for samples of syntax to do this (and
other creative hypothisis testing of regression results)?

Thanks,

Ben

Run the regression with x1 and x2 replaced by their sum and
difference, x1+x2 and x1-x2. Then the test of the coefficient
of the difference is equivalent to a test of the difference
between B1 and B2.

Hi Ray,

I'm always intrigued by your responses! I hope it's okay, but I have a
couple follow up questions. Just to make sure I understand, if you
wanted to know if any of the three "partial regression coefficients"
were different from each other, would you run three separate linear
regression analyses?...

Y= b0 + b1(x1+x2) + b2(x1-x2)

Y= b0+ b3(x2+x3) + b4(x2-x3)

Y= b0+ b5(x1+x3) + b6(x1-x3)

where

b2 reflects difference between partial weights for x1 and x2
b4 reflects difference between partial weights for x1 and x3
b6 reflects difference between partial weights for x2 and x3

This approach seems problematic to me because you're not taking into
consideration/statistically controlling for the third variable when
testing the differences.

Also, if you have the time, would you mind explaining why you need to
include the x1+x2, x2+x3, and x1+x3?

Sorry for so many questions. I completely understand if you're too
busy to respond.

Thank you,

Ryan

No, I'd do a different transformation, replacing {x1, x2, x3} with
{x1', x2' x3'} = {x1+x2+x3, 2x1-x2-x3, x2-x3}. Do it in two steps:
first put in x1', then add x2' and x3'. If the second step increases
R^2 significantly then you reject H: b1 = b2 = b3.

It may help to consider the general question: in a p-predictor model,
how can you test H: b1 = ... = bp? One way is a generalization of
the p=3 case above, the other is a generalization of the solution
Bruce posted in the SPSS-only version of this discussion. Of course,
the two solutions give the same answer, so they're really only two
different ways of thinking about the problem.

All very elegant ways to do it, but if the goal is to have the
parameter estimates of x1, x2, and x3 as part of the output, you would
have to do the math outside of SPSS to get the true parameter
estimates.  I too am new to SPSS, but know in programs like SAS you
can take the output of the regression as a data set.  Within a data
step (again SAS parlance for a set of syntax that manipulates the data
set), you should be able to construct the necessary t-statistic with
the parameter estimates, their variance and covariances (provided you
can output these as well from the regression into a data set) and ask
the package for the p-value of this statistic.  Is this possible to do
in SPSS?

If you want to test b1=b2=b3, you will need to do an F-test. A series
of t-tests won't do it.
-Dick Startz

Right. I should have been more specific... I was thinking in terms of
the original posting which seemed to be interested in testing the
parameter estimates pairwise (i.e., b1=b2, b2=b3, etc.), not testing
them all for equality. The pairwise test is a t-test, while the three
way test would be an F-test. Sorry for the confusion. My question
still stands though.... can one construct the test statistic from the
model output in a separate data step?

On Aug 1, 3:43 am, Ray Koopman <koop... at (no spam) sfu.ca> wrote:
On Jul 31, 6:47 pm, Ryan <Ryan.Andrew.Bl... at (no spam) gmail.com> wrote:



On Jul 31, 7:37 pm, Ray Koopman <koop... at (no spam) sfu.ca> wrote:
On Jul 31, 10:53 am, BenH34 <benho... at (no spam) earthlink.net> wrote:
Hello all.
I am a novice SPSS user, and cannot find a solution to a problem.
I am interested in finding a way to test if coefficient are
statistically different from each other. E.g. y = B1x1+ B2x2 + B3x3
+ e. And the hypothesis test I want to perform is if B1-B2=0? In
Stata there is syntax to directly test this within the regression
run. Is the same true in SPSS?
The only mention I could find online is:
"Unfortunately, unlike Stata, SPSS does not provide a convenient way
to test hypotheses like â1 = â2, e.g. the effects of education and job
experience are equal." http://www.nd.edu/~rwilliam/stats1/OLS-SPSS.pdf
I will ultimately have to perform 100s of these tests and am hopefuly
I can write the tests directly into the regression syntax.
Any ideas on where I can look for samples of syntax to do this (and
other creative hypothisis testing of regression results)?
Thanks,
Ben
Run the regression with x1 and x2 replaced by their sum and
difference, x1+x2 and x1-x2. Then the test of the coefficient
of the difference is equivalent to a test of the difference
between B1 and B2.
Hi Ray,
I'm always intrigued by your responses! I hope it's okay, but I have a
couple follow up questions. Just to make sure I understand, if you
wanted to know if any of the three "partial regression coefficients"
were different from each other, would you run three separate linear
regression analyses?...
Y= b0 + b1(x1+x2) + b2(x1-x2)
Y= b0+ b3(x2+x3) + b4(x2-x3)
Y= b0+ b5(x1+x3) + b6(x1-x3)
where
b2 reflects difference between partial weights for x1 and x2
b4 reflects difference between partial weights for x1 and x3
b6 reflects difference between partial weights for x2 and x3
This approach seems problematic to me because you're not taking into
consideration/statistically controlling for the third variable when
testing the differences.
Also, if you have the time, would you mind explaining why you need to
include the x1+x2, x2+x3, and x1+x3?
Sorry for so many questions. I completely understand if you're too
busy to respond.
Thank you,
Ryan
No, I'd do a different transformation, replacing {x1, x2, x3} with
{x1', x2' x3'} = {x1+x2+x3, 2x1-x2-x3, x2-x3}. Do it in two steps:
first put in x1', then add x2' and x3'. If the second step increases
R^2 significantly then you reject H: b1 = b2 = b3.

It may help to consider the general question: in a p-predictor model,
how can you test H: b1 = ... = bp? One way is a generalization of
the p=3 case above, the other is a generalization of the solution
Bruce posted in the SPSS-only version of this discussion. Of course,
the two solutions give the same answer, so they're really only two
different ways of thinking about the problem.

All very elegant ways to do it, but if the goal is to have the
parameter estimates of x1, x2, and x3 as part of the output, you would
have to do the math outside of SPSS to get the true parameter
estimates. I too am new to SPSS, but know in programs like SAS you
can take the output of the regression as a data set. Within a data
step (again SAS parlance for a set of syntax that manipulates the data
set), you should be able to construct the necessary t-statistic with
the parameter estimates, their variance and covariances (provided you
can output these as well from the regression into a data set) and ask
the package for the p-value of this statistic. Is this possible to do
in SPSS?