Negative Binomial Regression with PASW (or is it IBM-SPSS?) 18

[Marta Garcia-Granero]

Hi everybody:

I'm frustrated and tired at the same time, therefore, I'm asking for
help. I have Hilbe's "Negative Binomial Regression" besides me, and I'm
trying to get negative binomial regression models with PASW/SPSS 18.
Here are my questions:

1) How can I test for overdispersion in a Poisson Regression Model?
Hilbe's book it's absolutely focused on Stata (including some software
he actually wrote to easy the task). Hilbe mentions a Z test and a
Lagrange multiplier test, and shows some Stata commands that let me
absolutely non the wiser.

2) What role plays the scale (usually fixed at 1) factor in PASW 18
negative binomial regression output?. I have tried to use the "estimate
it" option, but the output still shows its value fixed at 1. I don't see
anything like that in Hilbe's book

Thanks
(with a bit of migraine)
Marta GG

--
For miscellaneous SPSS related statistical stuff, visit:
http://gjyp.nl/marta/

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

[Marija Norusis]

I think this is the way it still works:

 

1) For the negative binomial distribution,  to test that the value of the dispersion parameter is 0, specify a custom model with negative binomial as the distribution function, log for the link, and the value of 0 for the parameter. On the statistics tab, check the Lagrange multiplier test box.

 

2) For the binomial, Poisson, and negative binomial distributions, you can fix the scale parameter by dividing either the deviance or Pearson chi-square by its degrees of freedom. For the continuous distributions, you can estimate the scale parameter using maximum likelihood.

 

Marija Norusis

www.norusis.com

 

 

 

 

In a message dated 4/8/2010 11:29:48 A.M. Central Daylight Time, [hidden email] writes:

Hi everybody:

I'm frustrated and tired at the same time, therefore, I'm asking for
help. I have Hilbe's "Negative Binomial Regression" besides me, and I'm
trying to get negative binomial regression models with PASW/SPSS 18.
Here are my questions:

1) How can I test for overdispersion in a Poisson Regression Model?
Hilbe's book it's absolutely focused on Stata (including some software
he actually wrote to easy the task). Hilbe mentions a Z test and a
Lagrange multiplier test, and shows some Stata commands that let me
absolutely non the wiser.

2) What role plays the scale (usually fixed at 1) factor in PASW 18
negative binomial regression output?. I have tried to use the "estimate
it" option, but the output still shows its value fixed at 1. I don't see
anything like that in Hilbe's book

Thanks
(with a bit of migraine)
Marta GG

--
For miscellaneous SPSS related statistical stuff, visit:
http://gjyp.nl/marta/

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

 

[Marta Garcia-Granero]

Thank you very much Marija. It has helped a lot. After my little
tantrum, I was able to get the Z dispersion test, but I was still stuck
trying to compute the Lagrange multiplier test.

Best regards,
Marta GG


[hidden email] escribió:

> I think this is the way it still works:
>
> 1) For the negative binomial distribution,  to test that the value of
> the dispersion parameter is 0, specify a custom model with negative
> binomial as the distribution function, log for the link, and the value
> of 0 for the parameter. On the statistics tab, check the Lagrange
> multiplier test box.
>
> 2) For the binomial, Poisson, and negative binomial distributions, you
> can fix the scale parameter by dividing either the deviance or Pearson
> chi-square by its degrees of freedom. For the continuous
> distributions, you can estimate the scale parameter using maximum
> likelihood.
>
> Marija Norusis
> www.norusis.com <http://www.norusis.com>
>
>
>
>
> In a message dated 4/8/2010 11:29:48 A.M. Central Daylight Time,
> [hidden email] writes:
>
>     Hi everybody:
>
>     I'm frustrated and tired at the same time, therefore, I'm asking for
>     help. I have Hilbe's "Negative Binomial Regression" besides me,
>     and I'm
>     trying to get negative binomial regression models with PASW/SPSS 18.
>     Here are my questions:
>
>     1) How can I test for overdispersion in a Poisson Regression Model?
>     Hilbe's book it's absolutely focused on Stata (including some software
>     he actually wrote to easy the task). Hilbe mentions a Z test and a
>     Lagrange multiplier test, and shows some Stata commands that let me
>     absolutely non the wiser.
>
>     2) What role plays the scale (usually fixed at 1) factor in PASW 18
>     negative binomial regression output?. I have tried to use the
>     "estimate
>     it" option, but the output still shows its value fixed at 1. I
>     don't see
>     anything like that in Hilbe's book
>
>     Thanks
>     (with a bit of migraine)
>     Marta GG
>
>     --
>     For miscellaneous SPSS related statistical stuff, visit:
>     http://gjyp.nl/marta/
>
>     =====================
>     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
>
>

--
For miscellaneous SPSS related statistical stuff, visit:
http://gjyp.nl/marta/

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

[Joseph Hilbe]

The second edition of my Negative Binomial Regression (2011, Cambridge University Press) employs both Stata and R code throughout the book, which is over twice the length and depth of the first edition. SPSS is a fine package, but it is mostly menu run now by users, which does not make it easy to show sometimes complex program code for examples of data management, statistical modeling and tests. Moreover, SPSS has very few count models compared to SAS, Stata, R, Genstat, Limdep, and Matlab. So it is not a good choice to use for showing examples in a general text on applied count models.

With the complete Stata and R code now given for most every test related to count models in the literature, anyone with a good knowledge of SPSS programming should be able to convert the code to SPSS macro code. I also have a PDF ebook on my Selected Works site  
                                     http://works.bepress.com/joseph_hilbe/ 
which can be downloaded called Negative Binomial Regression Extensions, which provides even more code, including some SAS macro code for various count models. I am sure that decent SPSS programmers can convert Stata, R, and SAS programming code to SPSS. In fact, if anyone is interested in writing such code, I will post it to the ebook with full acknowledgement of the author who wrote it. I'd love to show SPSS code for al of the tests in the book.

Best,  Joseph Hilbe