Latent Gold Tolerance

[Zachary Feinstein]

I do not know if this is the appropriate forum for any discussions about Latent Gold but I figure it is worth a shot given I could not find a forum for Latent Gold and because SPSS now owns it, I think

 

I am currently working with datasets that behave very poorly.  I run LC9 models.  But many times the models do not converge properly.  I therefore adjust the Tolerance levels in the Technical tab, but I am not entirely sure of what each of the three Tolerance parameters represent during the algorithm.

 

I also wonder if there are any papers or sites that go into detail about the adjustment of the Tolerance.  Note that I very frequently adjust the /RCONVERGE criteria in SPSS during Factor Analyses to something like 0.1 to make convergence happen in fewer iterations.  I figure if it is good enough for SPSS Base it must be good enough for Latent Gold.

 

Any and all replies are highly valued and appreciated.

 

Zachary

[hidden email]

 

[Ryan]

Exactly what type of model are you trying to fit?

On Mon, May 24, 2010 at 8:28 PM, Zachary Feinstein <[hidden email]> wrote:

I do not know if this is the appropriate forum for any discussions about Latent Gold but I figure it is worth a shot given I could not find a forum for Latent Gold and because SPSS now owns it, I think

 

I am currently working with datasets that behave very poorly.  I run LC9 models.  But many times the models do not converge properly.  I therefore adjust the Tolerance levels in the Technical tab, but I am not entirely sure of what each of the three Tolerance parameters represent during the algorithm.

 

I also wonder if there are any papers or sites that go into detail about the adjustment of the Tolerance.  Note that I very frequently adjust the /RCONVERGE criteria in SPSS during Factor Analyses to something like 0.1 to make convergence happen in fewer iterations.  I figure if it is good enough for SPSS Base it must be good enough for Latent Gold.

 

Any and all replies are highly valued and appreciated.

 

Zachary

[hidden email]

 

[E. Bernardo]

Dear all,

 

What is the syntax to produce an MSWord/Excel version of the contents in the variable view?

 

To elaborate, my database(SPSS format) contains 20 variables (nominal, ordinal, scale).  I want to create a word/excel version that looks like the following:

 

Name        Label       Values                               Missing

Sex           Sex          1=male                                 9

                                2=female

                                9=Missing

Religion     Religion    1=Roman Catholic                 9

                                2=Muslim

                                3=Others

                                4=No religion

                                9= DK

                                99=No Answer

 

Thank you.

Eins

 

 

[Ruben Geert van den Berg]

Dear Zachary,
 
Did you consult the technical guide? It says (p. 52-53):
 

6.5 Convergence

 

The exact algorithm implemented in Latent GOLD works as follows. The

program starts with EM until either the maximum number of EM iterations

(Iteration Limits EM) or the EM convergence criterion (EM Tolerance) is

reached. Then, the program switches to NR iterations which stop when the

maximum number of NR iterations (Iteration Limits Newton-Raphson) or the

overall converge criterion (Tolerance) is reached. The convergence criterion

that is used is

 

[couldn't paste the formula here, sorry!]

 

which is the sum of the absolute relative changes in the parameters. The

program also stops iterating when the change in the log-posterior is negligible,

i.e., smaller than 10EXP−12.

The program reports the iteration process in the Iteration Detail output

file listing. Thus, it can easily be checked whether the maximum number of

iterations is reached without convergence. In addition, a warning is given if

one of the elements of the gradient is larger than 10−3.

 

It should be noted that sometimes it is more efficient to use only the

EM algorithm, which is accomplished by setting Iteration Limits Newton-

Raphson = 0 in the Technical Tab. This is, for instance, the case in models

with many parameters. With very large models, one may also consider suppressing

the computation of standard errors and Wald statistics.
 
The third tolerance (start values tolerance) is explained on pp. 53-54.
 
However, I'd like to know why the algorithm fails to converge. Perhaps you could to try and contact Prof. Vermunt directly. I bothered him a couple of times with questions and this didn't (seem to) bother him at all and he was very helpful.
 
HTH!
 

Ruben van den Berg

Methodologist

TNS NIPO

E: [hidden email]

P: [phone]

I: www.tns-nipo.com

 

---

Date: Mon, 24 May 2010 17:28:06 -0700
From: [hidden email]
Subject: Latent Gold Tolerance
To: [hidden email]

I do not know if this is the appropriate forum for any discussions about Latent Gold but I figure it is worth a shot given I could not find a forum for Latent Gold and because SPSS now owns it, I think

 

I am currently working with datasets that behave very poorly.  I run LC9 models.  But many times the models do not converge properly.  I therefore adjust the Tolerance levels in the Technical tab, but I am not entirely sure of what each of the three Tolerance parameters represent during the algorithm.

 

I also wonder if there are any papers or sites that go into detail about the adjustment of the Tolerance.  Note that I very frequently adjust the /RCONVERGE criteria in SPSS during Factor Analyses to something like 0.1 to make convergence happen in fewer iterations.  I figure if it is good enough for SPSS Base it must be good enough for Latent Gold.

 

Any and all replies are highly valued and appreciated.

 

Zachary

[hidden email]

 

---

New Windows 7: Find the right PC for you. Learn more.

[Albert-Jan Roskam]

I think you can use DISPLAY DICTIONARY together with an appropriate OMS command. There's also a python module (I believe it's CODEBOOK) for this task, I think.

Cheers!!
Albert-Jan

~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
All right, but apart from the sanitation, the medicine, education, wine, public order, irrigation, roads, a fresh water system, and public health, what have the Romans ever done for us?
~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~

--- On Tue, 5/25/10, Eins Bernardo <[hidden email]> wrote:

From: Eins Bernardo <[hidden email]>
Subject: [SPSSX-L] Exporting the contents in the "Variable View" to MS Word/Excel
To: [hidden email]
Date: Tuesday, May 25, 2010, 5:50 AM

Dear all,

 

What is the syntax to produce an MSWord/Excel version of the contents in the variable view?

 

To elaborate, my database(SPSS format) contains 20 variables (nominal, ordinal, scale).  I want to create a word/excel version that looks like the following:

 

Name        Label       Values                               Missing

Sex           Sex          1=male                                 9

                                2=female

                                9=Missing

Religion     Religion    1=Roman Catholic                 9

                                2=Muslim

                                3=Others

                                4=No religion

                                9= DK

                                99=No Answer

 

Thank you.

Eins