Tetrachoric correlation then PAF?

[E. Bernardo]

1. Given the number of variables, how to create a tetrachoric correlation matrix in SPSS version 16?
  2.  How to do a Principal Axis Factoring with Varimax rotation method in SPSS 16 using a correlation matrix as input data?

  Thanks.
  Eins


---------------------------------
  Get your preferred Email name!
Now you can @ymail.com and @rocketmail.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

[Maguin, Eugene]

Eins,

I can't comment on how to compute a tetrachoric matrix in spss. I don't know
the formula and would have to look it up. Possibly the computation could be
done using the matrix-end matrix command set (look this up in the syntax
reference). I'd bet there are several people on the list that know exactly
how to do it. I don't.

As far as reading in a matrix goes, that's kind of easy. Look at the Matrix
Data command, also in the syntax reference. Note that, as far as I know, the
matrix has to be written so that it is as many columns wide as the number of
variables. That might seem obvious but the point is that some programs ouput
(or can output) a matrix in different arrangements on the page (and by this,
I do not mean lower triangular vs rectangular). The very fact that you are
asking about reading a matrix in means that you are computing that matrix in
another program.

I think the PAF part is kind of trivial. Again, look at the syntax ref. The
key part is getting the matrix file in and selecting the extraction method
(choose the PAF option). A varimax rotation is the default.

Repost to the list if you have more questions.

Gene Maguin

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

[Maguin, Eugene]

All,

I'm working with someone whose DV is a proportion. Specifically, a count of
tasks completed a worker divided by total tasks undertaken in a unit time.
There are about 10 workers per unit, three units per condition, and two
conditions. Disregarding the DV type issue, I'm regarding this as a nested
design, units within condition. I really never work with proportions and
don't have hardly any experience. I'm thinking that one problem with
proportions is that the standard deviation of a set of proportions depends
on the mean proportion because the standard deviation of a proportion is
simply sqrt(p*q). To fix up this problem, one solution has been to transform
the raw proportions. So I'd like to hear advice on two lines of questions.

1) Is there a newer and more preferred way to analyze proportions within a
GLM framework than transformations?

2) What are the recommended types of transformations to use with
proportions?

Any excellent refs are appreciated.

Thanks, Gene Maguin

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

[Peck, Jon]

Tetrachoric correlation is just polychoric correlation with dichotomous variables, I think.  There is an SPSS Statistics 17 extension command, SPSSINC HETCOR, that computes polychoric correlations.  You can download it from SPSS Developer Central (www.spss.com/devcentral).  Besides Version 17, it requires R 2.7.0.  Full requirements are in the readme file.

HTH,
Jon Peck

-----Original Message-----
From: SPSSX(r) Discussion [mailto:[hidden email]] On Behalf Of Gene Maguin
Sent: Monday, October 13, 2008 9:04 AM
To: [hidden email]
Subject: Re: [SPSSX-L] Tetrachoric correlation then PAF?

Eins,

I can't comment on how to compute a tetrachoric matrix in spss. I don't know
the formula and would have to look it up. Possibly the computation could be
done using the matrix-end matrix command set (look this up in the syntax
reference). I'd bet there are several people on the list that know exactly
how to do it. I don't.

As far as reading in a matrix goes, that's kind of easy. Look at the Matrix
Data command, also in the syntax reference. Note that, as far as I know, the
matrix has to be written so that it is as many columns wide as the number of
variables. That might seem obvious but the point is that some programs ouput
(or can output) a matrix in different arrangements on the page (and by this,
I do not mean lower triangular vs rectangular). The very fact that you are
asking about reading a matrix in means that you are computing that matrix in
another program.

I think the PAF part is kind of trivial. Again, look at the syntax ref. The
key part is getting the matrix file in and selecting the extraction method
(choose the PAF option). A varimax rotation is the default.

Repost to the list if you have more questions.

Gene Maguin

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

[Justin Black]

There is also a macro, r_tetra, available for computing polychoric
(tetrachoric) correlations.
http://www2.jura.uni-hamburg.de/instkrim/kriminologie/Mitarbeiter/Enzmann/Software/Enzmann_Software.html

--Justin

On Mon, Oct 13, 2008 at 1:58 PM, Peck, Jon <[hidden email]> wrote:

> Tetrachoric correlation is just polychoric correlation with dichotomous
> variables, I think.  There is an SPSS Statistics 17 extension command,
> SPSSINC HETCOR, that computes polychoric correlations.  You can download it
> from SPSS Developer Central (www.spss.com/devcentral).  Besides Version
> 17, it requires R 2.7.0.  Full requirements are in the readme file.
>
> HTH,
> Jon Peck
>
> -----Original Message-----
> From: SPSSX(r) Discussion [mailto:[hidden email]] On Behalf Of
> Gene Maguin
> Sent: Monday, October 13, 2008 9:04 AM
> To: [hidden email]
> Subject: Re: [SPSSX-L] Tetrachoric correlation then PAF?
>
> Eins,
>
> I can't comment on how to compute a tetrachoric matrix in spss. I don't
> know
> the formula and would have to look it up. Possibly the computation could be
> done using the matrix-end matrix command set (look this up in the syntax
> reference). I'd bet there are several people on the list that know exactly
> how to do it. I don't.
>
> As far as reading in a matrix goes, that's kind of easy. Look at the Matrix
> Data command, also in the syntax reference. Note that, as far as I know,
> the
> matrix has to be written so that it is as many columns wide as the number
> of
> variables. That might seem obvious but the point is that some programs
> ouput
> (or can output) a matrix in different arrangements on the page (and by
> this,
> I do not mean lower triangular vs rectangular). The very fact that you are
> asking about reading a matrix in means that you are computing that matrix
> in
> another program.
>
> I think the PAF part is kind of trivial. Again, look at the syntax ref. The
> key part is getting the matrix file in and selecting the extraction method
> (choose the PAF option). A varimax rotation is the default.
>
> Repost to the list if you have more questions.
>
> Gene Maguin
>
> =====================
> 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
>

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

[Reutter, Alex]

Assuming that you have both the count of tasks completed and the total tasks undertaken available, have a look at using Generalized Linear Models (GENLIN) with a binomial distribution and the response specified as the number of events occurring in a set of trials.

Alex


-----Original Message-----
From: SPSSX(r) Discussion [mailto:[hidden email]] On Behalf Of Gene Maguin
Sent: Monday, October 13, 2008 11:58 AM
To: [hidden email]
Subject: Working proportions and GLM or equivalent

All,

I'm working with someone whose DV is a proportion. Specifically, a count of
tasks completed a worker divided by total tasks undertaken in a unit time.
There are about 10 workers per unit, three units per condition, and two
conditions. Disregarding the DV type issue, I'm regarding this as a nested
design, units within condition. I really never work with proportions and
don't have hardly any experience. I'm thinking that one problem with
proportions is that the standard deviation of a set of proportions depends
on the mean proportion because the standard deviation of a proportion is
simply sqrt(p*q). To fix up this problem, one solution has been to transform
the raw proportions. So I'd like to hear advice on two lines of questions.

1) Is there a newer and more preferred way to analyze proportions within a
GLM framework than transformations?

2) What are the recommended types of transformations to use with
proportions?

Any excellent refs are appreciated.

Thanks, Gene Maguin

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

[Swank, Paul R]

Given the nested design, I think you are looking at a general non-linear
mixed model. This can be done in HLM, MLWin, SAS, and a few others, but
not as far as I know in SPSS.

Paul R. Swank, Ph.D
Professor and Director of Research
Children's Learning Institute
University of Texas Health Science Center
Houston, TX 77038


-----Original Message-----
From: SPSSX(r) Discussion [mailto:[hidden email]] On Behalf Of
Reutter, Alex
Sent: Monday, October 13, 2008 3:23 PM
To: [hidden email]
Subject: Re: Working proportions and GLM or equivalent

Assuming that you have both the count of tasks completed and the total
tasks undertaken available, have a look at using Generalized Linear
Models (GENLIN) with a binomial distribution and the response specified
as the number of events occurring in a set of trials.

Alex


-----Original Message-----
From: SPSSX(r) Discussion [mailto:[hidden email]] On Behalf Of
Gene Maguin
Sent: Monday, October 13, 2008 11:58 AM
To: [hidden email]
Subject: Working proportions and GLM or equivalent

All,

I'm working with someone whose DV is a proportion. Specifically, a count
of
tasks completed a worker divided by total tasks undertaken in a unit
time.
There are about 10 workers per unit, three units per condition, and two
conditions. Disregarding the DV type issue, I'm regarding this as a
nested
design, units within condition. I really never work with proportions and
don't have hardly any experience. I'm thinking that one problem with
proportions is that the standard deviation of a set of proportions
depends
on the mean proportion because the standard deviation of a proportion is
simply sqrt(p*q). To fix up this problem, one solution has been to
transform
the raw proportions. So I'd like to hear advice on two lines of
questions.

1) Is there a newer and more preferred way to analyze proportions within
a
GLM framework than transformations?

2) What are the recommended types of transformations to use with
proportions?

Any excellent refs are appreciated.

Thanks, Gene Maguin

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

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

[Maguin, Eugene]

Pablo, thank you for your reply. I have Raudenbush's HLM book 1st edition
and I'll check that. Or, are you thinking specifically of the second
edition. However, we won't have the HLM program. Mplus, yes, but not HLM.


Paul, thanks also. To be a bit lazy, how would this analysis be classfied in
Mplus? Right now, it seems to be trials within worker within unit within
condition. But, I don't think that is right.


Alex, I'd like to follow up your reply since you've replied in the context
of spss. First, I'd like to make sure I understand the data setup and
command setup.

Would the (minimal) command setup be

GENLIN TasksDone of TotalTasks by unit condition/model unit(condition)
   distribution=binomial link=logit.


Ok, data setup. Would it look like this?

Id   condition  unit  tasksdone  totaltasks
101      1        1       15        30


Thanks, Gene Maguin


>>I'm working with someone whose DV is a proportion. Specifically, a count
of
tasks completed a worker divided by total tasks undertaken in a unit time.
There are about 10 workers per unit, three units per condition, and two
conditions. Disregarding the DV type issue, I'm regarding this as a nested
design, units within condition. I really never work with proportions and
don't have hardly any experience. I'm thinking that one problem with
proportions is that the standard deviation of a set of proportions depends
on the mean proportion because the standard deviation of a proportion is
simply sqrt(p*q). To fix up this problem, one solution has been to transform
the raw proportions. So I'd like to hear advice on two lines of questions.

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

[Swank, Paul R]

I thought I said HLM, MLWin, or SAS. Mplus will handle one level of
nesting but I'm not sure about the proportional data. I was thinking
about nested logistic regression with a number of events/number of
opportunities. I am pretty sure HLM 2 will do this, but not HLM 1. I
have all those programs but if I were going to do it, I would use SAS
proc glimmix or nlmixed.

Paul R. Swank, Ph.D
Professor and Director of Research
Children's Learning Institute
University of Texas Health Science Center
Houston, TX 77038


-----Original Message-----
From: SPSSX(r) Discussion [mailto:[hidden email]] On Behalf Of
Gene Maguin
Sent: Monday, October 13, 2008 4:54 PM
To: [hidden email]
Subject: Re: Working proportions and GLM or equivalent

Pablo, thank you for your reply. I have Raudenbush's HLM book 1st
edition
and I'll check that. Or, are you thinking specifically of the second
edition. However, we won't have the HLM program. Mplus, yes, but not
HLM.


Paul, thanks also. To be a bit lazy, how would this analysis be
classfied in
Mplus? Right now, it seems to be trials within worker within unit within
condition. But, I don't think that is right.


Alex, I'd like to follow up your reply since you've replied in the
context
of spss. First, I'd like to make sure I understand the data setup and
command setup.

Would the (minimal) command setup be

GENLIN TasksDone of TotalTasks by unit condition/model unit(condition)
   distribution=binomial link=logit.


Ok, data setup. Would it look like this?

Id   condition  unit  tasksdone  totaltasks
101      1        1       15        30


Thanks, Gene Maguin


>>I'm working with someone whose DV is a proportion. Specifically, a
count
of
tasks completed a worker divided by total tasks undertaken in a unit
time.
There are about 10 workers per unit, three units per condition, and two
conditions. Disregarding the DV type issue, I'm regarding this as a
nested
design, units within condition. I really never work with proportions and
don't have hardly any experience. I'm thinking that one problem with
proportions is that the standard deviation of a set of proportions
depends
on the mean proportion because the standard deviation of a proportion is
simply sqrt(p*q). To fix up this problem, one solution has been to
transform
the raw proportions. So I'd like to hear advice on two lines of
questions.

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

[E. Bernardo]

I am quite intrigue with SPSS that requires R software.  Does it mean that I would have to install R and SPSS in the same computer, then a syntax may be written to the spss syntax editor?

  Eins

"Peck, Jon" <[hidden email]> wrote:
  Tetrachoric correlation is just polychoric correlation with dichotomous variables, I think. There is an SPSS Statistics 17 extension command, SPSSINC HETCOR, that computes polychoric correlations. You can download it from SPSS Developer Central (www.spss.com/devcentral). Besides Version 17, it requires R 2.7.0. Full requirements are in the readme file.

HTH,
Jon Peck

-----Original Message-----
From: SPSSX(r) Discussion [mailto:[hidden email]] On Behalf Of Gene Maguin
Sent: Monday, October 13, 2008 9:04 AM
To: [hidden email]
Subject: Re: [SPSSX-L] Tetrachoric correlation then PAF?

Eins,

I can't comment on how to compute a tetrachoric matrix in spss. I don't know
the formula and would have to look it up. Possibly the computation could be
done using the matrix-end matrix command set (look this up in the syntax
reference). I'd bet there are several people on the list that know exactly
how to do it. I don't.

As far as reading in a matrix goes, that's kind of easy. Look at the Matrix
Data command, also in the syntax reference. Note that, as far as I know, the
matrix has to be written so that it is as many columns wide as the number of
variables. That might seem obvious but the point is that some programs ouput
(or can output) a matrix in different arrangements on the page (and by this,
I do not mean lower triangular vs rectangular). The very fact that you are
asking about reading a matrix in means that you are computing that matrix in
another program.

I think the PAF part is kind of trivial. Again, look at the syntax ref. The
key part is getting the matrix file in and selecting the extraction method
(choose the PAF option). A varimax rotation is the default.

Repost to the list if you have more questions.

Gene Maguin

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



---------------------------------
  New Email addresses available on Yahoo!
Get the Email name you&#39;ve always wanted on the new @ymail and @rocketmail.
Hurry before someone else does!

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

[Peck, Jon]

You can install the optional R Plug-In from SPSS Developer Central, which requires the appropriate version of R as well (2.5.0 for SPSS 16 and 2.7.0 for version 17).  Then you can write R programs in the SPSS Viewer between BEGIN PROGRAM R. and END PROGRAM.  These can fetch the active SPSS dataset or portions thereof, run R code, and have the results appear as text, pivot tables and/or charts (version 17) in the SPSS Viewer.

 

Using the Version 17 Custom Dialog Builder, you can create SPSS dialog boxes for your R programs or extension commands (or for regular SPSS syntax, too).

 

Of course, the standard SPSS commands do not use R, but you now have the option of extending SPSS with R packages.

 

Check out the documentation that comes with the R plug-in and the downloadable Data Management book, which is linked on Developer Central (www.spss.com/devcentral).  You can also try out the many extension commands in Version 17 for running R packages.

 

HTH,

Jon Peck

 

________________________________

From: Eins Bernardo [mailto:[hidden email]]
Sent: Monday, October 13, 2008 11:48 PM
To: Peck, Jon; [hidden email]
Subject: Re: Tetrachoric correlation then PAF?

 

I am quite intrigue with SPSS that requires R software.  Does it mean that I would have to install R and SPSS in the same computer, then a syntax may be written to the spss syntax editor?

 

Eins

"Peck, Jon" <[hidden email]> wrote:

        Tetrachoric correlation is just polychoric correlation with dichotomous variables, I think. There is an SPSS Statistics 17 extension command, SPSSINC HETCOR, that computes polychoric correlations. You can download it from SPSS Developer Central (www.spss.com/devcentral). Besides Version 17, it requires R 2.7.0. Full requirements are in the readme file.
       
        HTH,
        Jon Peck
       
        -----Original Message-----
        From: SPSSX(r) Discussion [mailto:[hidden email]] On Behalf Of Gene Maguin
        Sent: Monday, October 13, 2008 9:04 AM
        To: [hidden email]
        Subject: Re: [SPSSX-L] Tetrachoric correlation then PAF?
       
        Eins,
       
        I can't comment on how to compute a tetrachoric matrix in spss. I don't know
        the formula and would have to look it up. Possibly the computation could be
        done using the matrix-end matrix command set (look this up in the syntax
        reference). I'd bet there are several people on the list that know exactly
        how to do it. I don't.
       
        As far as reading in a matrix goes, that's kind of easy. Look at the Matrix
        Data command, also in the syntax reference. Note that, as far as I know, the
        matrix has to be written so that it is as many columns wide as the number of
        variables. That might seem obvious but the point is that some programs ouput
        (or can output) a matrix in different arrangements on the page (and by this,
        I do not mean lower triangular vs rectangular). The very fact that you are
        asking about reading a matrix in means that you are computing that matrix in
        another program.
       
        I think the PAF part is kind of trivial. Again, look at the syntax ref. The
        key part is getting the matrix file in and selecting the extraction method
        (choose the PAF option). A varimax rotation is the default.
       
        Repost to the list if you have more questions.
       
        Gene Maguin
       
        =====================
        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

 

 

________________________________

New Email addresses available on Yahoo! <http://sg.rd.yahoo.com/ph/mail/domainchoice/mail/signature/*http:/mail.promotions.yahoo.com/newdomains/ph/>
Get the Email name you've always wanted on the new @ymail and @rocketmail.
Hurry before someone else does!

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

[Reutter, Alex]

That looks right.  Note that this assumes that the probability a given worker completes a given task is independent of the probability that she completes the other tasks.  If you want to treat the tasks as repeated measurements, then have a look at Generalized Estimating Equations (also GENLIN).  Your data setup would then be like:

Id   condition  unit  task  taskcomplete
101      1        1    1        0
...
101      1        1    30       1

And you'd add a REPEATED subcommand with id as your SUBJECT variable and task as your WITHINSUBJECT variable.

Note that GENLIN doesn't handle random effects.

Alex


-----Original Message-----
From: SPSSX(r) Discussion [mailto:[hidden email]] On Behalf Of Gene Maguin
Sent: Monday, October 13, 2008 4:54 PM
To: [hidden email]
Subject: Re: Working proportions and GLM or equivalent

Alex, I'd like to follow up your reply since you've replied in the context
of spss. First, I'd like to make sure I understand the data setup and
command setup.

Would the (minimal) command setup be

GENLIN TasksDone of TotalTasks by unit condition/model unit(condition)
   distribution=binomial link=logit.


Ok, data setup. Would it look like this?

Id   condition  unit  tasksdone  totaltasks
101      1        1       15        30


Thanks, Gene Maguin


>>I'm working with someone whose DV is a proportion. Specifically, a count
of
tasks completed a worker divided by total tasks undertaken in a unit time.
There are about 10 workers per unit, three units per condition, and two
conditions. Disregarding the DV type issue, I'm regarding this as a nested
design, units within condition. I really never work with proportions and
don't have hardly any experience. I'm thinking that one problem with
proportions is that the standard deviation of a set of proportions depends
on the mean proportion because the standard deviation of a proportion is
simply sqrt(p*q). To fix up this problem, one solution has been to transform
the raw proportions. So I'd like to hear advice on two lines of questions.

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

[Maguin, Eugene]

I'm running the following genlin procedure.

GENLIN TasksDone of TotalTasks by unit condition/model unit(condition)
   distribution=binomial link=logit.

Condition has two levels and unit has six levels. N per level of unit ranges
between 4 and 9.

I got this warning.

Warnings
The maximum number of step-halvings was reached but the log-likelihood value
cannot be further improved. Output for the last iteration is displayed.
The GENLIN procedure continues despite the above warning(s). Subsequent
results shown are based on the last iteration. Validity of the model fit is
uncertain.

I am completely willing to accept the criticism that my sample is too small.
However, within that limitation, is there an estimation parameter that I can
vary to try to get a valid solution?

Thanks, Gene Maguin

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

[Reutter, Alex]

What does your iteration history look like?  Use /PRINT HISTORY(1).  In particular, what are the last few values of the parameter estimates, and which ones appear to not be converging?

Alex


-----Original Message-----
From: SPSSX(r) Discussion [mailto:[hidden email]] On Behalf Of Gene Maguin
Sent: Tuesday, October 14, 2008 2:09 PM
To: [hidden email]
Subject: Re: Working proportions and GLM or equivalent

I'm running the following genlin procedure.

GENLIN TasksDone of TotalTasks by unit condition/model unit(condition)
   distribution=binomial link=logit.

Condition has two levels and unit has six levels. N per level of unit ranges
between 4 and 9.

I got this warning.

Warnings
The maximum number of step-halvings was reached but the log-likelihood value
cannot be further improved. Output for the last iteration is displayed.
The GENLIN procedure continues despite the above warning(s). Subsequent
results shown are based on the last iteration. Validity of the model fit is
uncertain.

I am completely willing to accept the criticism that my sample is too small.
However, within that limitation, is there an estimation parameter that I can
vary to try to get a valid solution?

Thanks, Gene Maguin

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