Model selection
Locked
 
|
Hi all,
I'm reading a paper that has used log likelihood ratio tests and Bayesian Information Criterion for model selection. I'm not sure whether of either of these were done in SPSS, so am looking for a little advice on the procedure. That is, can anyone direct me to some good resources or texts or provide me with any information about these statistical techniques? Thanks |
|
|
Hi K S Scot
KSS> I'm reading a paper that has used log likelihood ratio tests and Bayesian KSS> Information Criterion for model selection. I'm not sure whether of either KSS> of these were done in SPSS, so am looking for a little advice on the KSS> procedure. That is, can anyone direct me to some good resources or texts or KSS> provide me with any information about these statistical techniques? You don't mention the type of model (linear, logistic, Cox, Poisson, nominal, ordinal.. regression). LR tests are part of the standard output of SPSS for any model but linear regression. BIC can be obtained for linear regression models (subcommand /SELECTION, it must be added thru syntax). Right now I can't think of a book that covers those topics, I'll check the University library and see if I find someone. -- Regards, Dr. Marta García-Granero,PhD mailto:[hidden email] Statistician --- "It is unwise to use a statistical procedure whose use one does not understand. SPSS syntax guide cannot supply this knowledge, and it is certainly no substitute for the basic understanding of statistics and statistical thinking that is essential for the wise choice of methods and the correct interpretation of their results". (Adapted from WinPepi manual - I'm sure Joe Abrahmson will not mind) |
|
|
In reply to this post by Stats Q
Hi Marta, Thanks for your e-mail. The paper says that a Log-linear model was used. From what you are saying then, I guess I've interpreted things incorrectly then i.e., log-likelihood ratio tests will be part of the output from the BIC analysis. Do you know if is the case that the lower the log-likelihood value, the better the model fits the data? And if so, how low does the value have to be to be considered low? And why would the log-likelihood value need to be reported as well as a BIC value? Thanks. I appreciate the advice.
Fed up with spam in your inbox? Find out how to deal with junk e-mail here! |
|
|
Folks,
This is a test...not a response to K S Scot. I seem to be unable to post to the SPSS-X listserv. My query is returned as rejected by the server as "spam." Any ideas? Please e-mail me at [hidden email] Thanks...and sorry, K S Scot Stats Q <[hidden email]> wrote: Hi Marta, Thanks for your e-mail. The paper says that a Log-linear model was used. From what you are saying then, I guess I've interpreted things incorrectly then i.e., log-likelihood ratio tests will be part of the output from the BIC analysis. Do you know if is the case that the lower the log-likelihood value, the better the model fits the data? And if so, how low does the value have to be to be considered low? And why would the log-likelihood value need to be reported as well as a BIC value? Thanks. I appreciate the advice. K S Scot --------------------------------- From: Marta García-Granero <[hidden email]> Reply-To: Marta García-Granero <[hidden email]> To: [hidden email] Subject: Re: Model selection Date: Wed, 11 Oct 2006 11:42:39 +0200 >Hi K S Scot > >KSS> I'm reading a paper that has used log likelihood ratio tests and Bayesian >KSS> Information Criterion for model selection. I'm not sure whether of either >KSS> of these were done in SPSS, so am looking for a little advice on the >KSS> procedure. That is, can anyone direct me to some good resources or texts or >KSS> provide me with any information about these statistical techniques? > >You don't mention the type of model (linear, logistic, Cox, >Poisson, nominal, ordinal.. regression). > >LR tests are part of the standard output of SPSS for any model but >linear regression. BIC can be obtained for linear regression models >(subcommand /SELECTION, it must be added thru syntax). > >Right now I can't think of a book that covers those topics, I'll check >the University library and see if I find someone. > > >-- >Regards, >Dr. Marta García-Granero,PhD mailto:[hidden email] >Statistician > >--- >"It is unwise to use a statistical procedure whose use one does >not understand. SPSS syntax guide cannot supply this knowledge, and it >is certainly no substitute for the basic understanding of statistics >and statistical thinking that is essential for the wise choice of >methods and the correct interpretation of their results". > >(Adapted from WinPepi manual - I'm sure Joe Abrahmson will not mind) --------------------------------- Fed up with spam in your inbox? Find out how to deal with junk e-mail here! |
|
|
King,
This message was duly distributed to the list. Perhaps it was also reported to you as rejected. Server behavior is a little contradictory at times. Hector -----Mensaje original----- De: SPSSX(r) Discussion [mailto:[hidden email]] En nombre de King Douglas Enviado el: 11 October 2006 12:47 Para: [hidden email] Asunto: Re: Model selection Folks, This is a test...not a response to K S Scot. I seem to be unable to post to the SPSS-X listserv. My query is returned as rejected by the server as "spam." Any ideas? Please e-mail me at [hidden email] Thanks...and sorry, K S Scot Stats Q <[hidden email]> wrote: Hi Marta, Thanks for your e-mail. The paper says that a Log-linear model was used. From what you are saying then, I guess I've interpreted things incorrectly then i.e., log-likelihood ratio tests will be part of the output from the BIC analysis. Do you know if is the case that the lower the log-likelihood value, the better the model fits the data? And if so, how low does the value have to be to be considered low? And why would the log-likelihood value need to be reported as well as a BIC value? Thanks. I appreciate the advice. K S Scot --------------------------------- From: Marta García-Granero <[hidden email]> Reply-To: Marta García-Granero <[hidden email]> To: [hidden email] Subject: Re: Model selection Date: Wed, 11 Oct 2006 11:42:39 +0200 >Hi K S Scot > >KSS> I'm reading a paper that has used log likelihood ratio tests and Bayesian >KSS> Information Criterion for model selection. I'm not sure whether of either >KSS> of these were done in SPSS, so am looking for a little advice on the >KSS> procedure. That is, can anyone direct me to some good resources or texts or >KSS> provide me with any information about these statistical techniques? > >You don't mention the type of model (linear, logistic, Cox, >Poisson, nominal, ordinal.. regression). > >LR tests are part of the standard output of SPSS for any model but >linear regression. BIC can be obtained for linear regression models >(subcommand /SELECTION, it must be added thru syntax). > >Right now I can't think of a book that covers those topics, I'll check >the University library and see if I find someone. > > >-- >Regards, >Dr. Marta García-Granero,PhD mailto:[hidden email] >Statistician > >--- >"It is unwise to use a statistical procedure whose use one does >not understand. SPSS syntax guide cannot supply this knowledge, and it >is certainly no substitute for the basic understanding of statistics >and statistical thinking that is essential for the wise choice of >methods and the correct interpretation of their results". > >(Adapted from WinPepi manual - I'm sure Joe Abrahmson will not mind) --------------------------------- Fed up with spam in your inbox? Find out how to deal with junk e-mail here! |
"Spam" to SPSSX-L (was, re: Model selection)
|
In reply to this post by King Douglas
At 11:46 AM 10/11/2006, King Douglas wrote:
>I seem to be unable to post to the SPSS-X listserv. My query is >returned as rejected by the server as "spam." Any ideas? I recently had a posting rejected as spam. I asked, and at 05:15 PM 10/5/2006, UGA Listserv Admin (thanks, Jean Snow) wrote: >Damn... Sorry about this. We had to seriously lower spam thresholds >and add tests last month because Listserv and our mail gateways were >constantly being blocked by SpamCop, and sadly, many of our >subscriber's sites use it to determine what's spam. People were >getting autodeleted. It was really bad. And at 09:02 PM 10/6/2006, UGA Listserv Admin wrote: >Give Nathan a little time, and the UGA rules will be tweaked to let >your, and similar, posts through. And do let us know if this happens >in the future. So it's a known problem. I've no idea how widespread; have other people had legitimate posts rejected? Jean, should people forward you copies of the messages for rejected legitimate postings? -Cheers and tangles, Richard |
Re: "Spam" to SPSSX-L (was, re: Model selection)
|
|
I did not have any of my messages rejected by the UGA list server, but SOME,
only some messages to the list (e.g. two messages sent today by our colleague Sebastian Daza) got marked as spam by my own ISP, heavens know why. Messages, you know, arrive from the server, not from the list member, so I do not understand why some of them are caught and others are not. These server deities governing our lives and communications are as capricious as Greek gods in Mount Olympus (or Teutonic gods in Valhalla). Would they be appeased by sacrificing a lamb, or they would require something more substantial? Smashing a laptop perhaps? Hector -----Mensaje original----- De: SPSSX(r) Discussion [mailto:[hidden email]] En nombre de Richard Ristow Enviado el: 11 October 2006 16:32 Para: [hidden email] Asunto: "Spam" to SPSSX-L (was, re: Model selection) At 11:46 AM 10/11/2006, King Douglas wrote: >I seem to be unable to post to the SPSS-X listserv. My query is >returned as rejected by the server as "spam." Any ideas? I recently had a posting rejected as spam. I asked, and at 05:15 PM 10/5/2006, UGA Listserv Admin (thanks, Jean Snow) wrote: >Damn... Sorry about this. We had to seriously lower spam thresholds >and add tests last month because Listserv and our mail gateways were >constantly being blocked by SpamCop, and sadly, many of our >subscriber's sites use it to determine what's spam. People were >getting autodeleted. It was really bad. And at 09:02 PM 10/6/2006, UGA Listserv Admin wrote: >Give Nathan a little time, and the UGA rules will be tweaked to let >your, and similar, posts through. And do let us know if this happens >in the future. So it's a known problem. I've no idea how widespread; have other people had legitimate posts rejected? Jean, should people forward you copies of the messages for rejected legitimate postings? -Cheers and tangles, Richard |
«
Return to SPSSX Discussion
|
1 view|%1 views
| Free forum by Nabble | Edit this page |