Increased betas in multiple regression
Locked
Increased betas in multiple regression
 
|
I have a collegaue's data set in in which variables X1, X2, X3 and signficant predictors of Y in simple linear regressions. I find that when X1, X2, and X3 are all in the model, the betas are conisderably larger for All predictors then they were in the simple linear models. My colleague is excited aboout the fiact that the independnt and combined effects are so differnt and I am looking for a way to tease apart what is going on in this model. Any thoughst? William N. Dudley, PhD Associate Dean for Research The School of Health and Human Performance Office of Research The University of North Carolina at Greensboro 126 HHP Building, PO Box 26170 Greensboro, NC 27402-6170 VOICE 336.2562475 FAX 336.334.3238 |
Re: Increased betas in multiple regression
|
|
William,
At first read-thru, you seem to be describing a suppression effect. Suppression has been discussed off and on both here and, especially, on SEMNET. If you subscribe, you could look in the archives. Cohen's multiple regression book has several entries for suppression. Gene Maguin >> I have a collegaue's data set in in which variables X1, X2, X3 and signficant predictors of Y in simple linear regressions. I find that when X1, X2, and X3 are all in the model, the betas are conisderably larger for All predictors then they were in the simple linear models. My colleague is excited aboout the fiact that the independnt and combined effects are so differnt and I am looking for a way to tease apart what is going on in this model. Any thoughst? William N. Dudley, PhD Associate Dean for Research The School of Health and Human Performance Office of Research The University of North Carolina at Greensboro 126 HHP Building, PO Box 26170 Greensboro, NC 27402-6170 VOICE 336.2562475 FAX 336.334.3238 ===================== 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 |
Re: Increased betas in multiple regression
|
|
Gene Thanks for the pointer. I had looked at Cohen's text and I had always thought of supression as what happens when a variable comes into the model. Are you suggesting that in the simple linear model we have suppression effects for variables NOT in the model. BTW the correlations among the x's is low. Agian thanks for the advice. I will look closer at Cohen & Cohen Bill William N. Dudley, PhD Associate Dean for Research The School of Health and Human Performance Office of Research The University of North Carolina at Greensboro 126 HHP Building, PO Box 26170 Greensboro, NC 27402-6170 VOICE 336.2562475 FAX 336.334.3238
William, At first read-thru, you seem to be describing a suppression effect. Suppression has been discussed off and on both here and, especially, on SEMNET. If you subscribe, you could look in the archives. Cohen's multiple regression book has several entries for suppression. Gene Maguin >> I have a collegaue's data set in in which variables X1, X2, X3 and signficant predictors of Y in simple linear regressions. I find that when X1, X2, and X3 are all in the model, the betas are conisderably larger for All predictors then they were in the simple linear models. My colleague is excited aboout the fiact that the independnt and combined effects are so differnt and I am looking for a way to tease apart what is going on in this model. Any thoughst? William N. Dudley, PhD Associate Dean for Research The School of Health and Human Performance Office of Research The University of North Carolina at Greensboro 126 HHP Building, PO Box 26170 Greensboro, NC 27402-6170 VOICE 336.2562475 FAX 336.334.3238 ===================== 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 |
Re: Increased betas in multiple regression
|
|
William,
The variables in the suppressor role have to be in the model. I want to plead 'misunderstanding of your posting'. Let's say I'm confused by your posting. Let's say I think the first sentence has some missing words. >> I have a collegaue's data set in in which variables X1, X2, X3 and signficant predictors of Y in simple linear regressions. I find that when X1, X2, and X3 are all in the model, the betas are conisderably larger for All predictors then they were in the simple linear models. Could you be saying that X1, X2, and X3 are each a significant predictor of Y when separate regressions are constructed for Y? Are there other variables in these separate regressions, say z1, z2, z3? And then when a single regression is constructed with X1, X2, and X3 all predicting Y, the betas for X1, X2, and X3 are all larger than in the individual regressions? Gene ===================== 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 |
|
|
Hello listers,
I would like to do a longitudinal logistic regression. I have health data, and I would like to see if a procedure is administered or not over time and at different facilities. I guess that makes it longitudinal and multilevel. Should I be doing generalized estimating equations for this analysis? And if so, can someone recommend a good book for this type of analysis that provides some guidance for GEEs with SPSS? Thanks Matt Matthew Pirritano, Ph.D. Research Analyst IV Medical Services Initiative (MSI) Orange County Health Care Agency (714) 568-5648 ===================== 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 |
Re: Generalized Estimating Equations
|
|
Actually, this sounds like a non-linear mixed or hierarchical model. I don't believe SPSS has that yet but HLM, Mlwin, or SAS will do it.
Dr. Paul R. Swank, Professor and Director of Research Children's Learning Institute University of Texas Health Science Center-Houston -----Original Message----- From: SPSSX(r) Discussion [mailto:[hidden email]] On Behalf Of Pirritano, Matthew Sent: Monday, March 30, 2009 12:37 PM To: [hidden email] Subject: Generalized Estimating Equations Hello listers, I would like to do a longitudinal logistic regression. I have health data, and I would like to see if a procedure is administered or not over time and at different facilities. I guess that makes it longitudinal and multilevel. Should I be doing generalized estimating equations for this analysis? And if so, can someone recommend a good book for this type of analysis that provides some guidance for GEEs with SPSS? Thanks Matt Matthew Pirritano, Ph.D. Research Analyst IV Medical Services Initiative (MSI) Orange County Health Care Agency (714) 568-5648 ===================== 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 |
Re: Generalized Estimating Equations
|
|
Matt,
Could you tell the list a little more about the structure of the dataset? At what level are you measuring whether a procedure was performed; on a patient-by-patient basis, or on a facility-by-facility basis, or something else? It's not clear to me what your "subjects" are, so while it's possible that you might not be able to use GEE, it's also possible that you might. Cheers, Alex -----Original Message----- From: SPSSX(r) Discussion [mailto:[hidden email]] On Behalf Of Swank, Paul R Sent: Monday, March 30, 2009 12:48 PM To: [hidden email] Subject: Re: Generalized Estimating Equations Actually, this sounds like a non-linear mixed or hierarchical model. I don't believe SPSS has that yet but HLM, Mlwin, or SAS will do it. Dr. Paul R. Swank, Professor and Director of Research Children's Learning Institute University of Texas Health Science Center-Houston -----Original Message----- From: SPSSX(r) Discussion [mailto:[hidden email]] On Behalf Of Pirritano, Matthew Sent: Monday, March 30, 2009 12:37 PM To: [hidden email] Subject: Generalized Estimating Equations Hello listers, I would like to do a longitudinal logistic regression. I have health data, and I would like to see if a procedure is administered or not over time and at different facilities. I guess that makes it longitudinal and multilevel. Should I be doing generalized estimating equations for this analysis? And if so, can someone recommend a good book for this type of analysis that provides some guidance for GEEs with SPSS? Thanks Matt Matthew Pirritano, Ph.D. Research Analyst IV Medical Services Initiative (MSI) Orange County Health Care Agency (714) 568-5648 ===================== 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 |
Re: Generalized Estimating Equations
|
|
My data is patient level data with a variable that identifies the
facility. It is detailed claims data. So there is information about all procedures. The subject is the patient. Each line of data is a service. I have read that this type of data is at the intersection of GEEs and mixed models, and that SPSS cannot do this. But being that I only have SPSS is there a work around or a next best fit for my data? Thanks matt Matthew Pirritano, Ph.D. Research Analyst IV Medical Services Initiative (MSI) Orange County Health Care Agency (714) 568-5648 -----Original Message----- From: SPSSX(r) Discussion [mailto:[hidden email]] On Behalf Of Reutter, Alex Sent: Monday, March 30, 2009 11:52 AM To: [hidden email] Subject: Re: Generalized Estimating Equations Matt, Could you tell the list a little more about the structure of the dataset? At what level are you measuring whether a procedure was performed; on a patient-by-patient basis, or on a facility-by-facility basis, or something else? It's not clear to me what your "subjects" are, so while it's possible that you might not be able to use GEE, it's also possible that you might. Cheers, Alex -----Original Message----- From: SPSSX(r) Discussion [mailto:[hidden email]] On Behalf Of Swank, Paul R Sent: Monday, March 30, 2009 12:48 PM To: [hidden email] Subject: Re: Generalized Estimating Equations Actually, this sounds like a non-linear mixed or hierarchical model. I don't believe SPSS has that yet but HLM, Mlwin, or SAS will do it. Dr. Paul R. Swank, Professor and Director of Research Children's Learning Institute University of Texas Health Science Center-Houston -----Original Message----- From: SPSSX(r) Discussion [mailto:[hidden email]] On Behalf Of Pirritano, Matthew Sent: Monday, March 30, 2009 12:37 PM To: [hidden email] Subject: Generalized Estimating Equations Hello listers, I would like to do a longitudinal logistic regression. I have health data, and I would like to see if a procedure is administered or not over time and at different facilities. I guess that makes it longitudinal and multilevel. Should I be doing generalized estimating equations for this analysis? And if so, can someone recommend a good book for this type of analysis that provides some guidance for GEEs with SPSS? Thanks Matt Matthew Pirritano, Ph.D. Research Analyst IV Medical Services Initiative (MSI) Orange County Health Care Agency (714) 568-5648 ===================== 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 |
Re: Generalized Estimating Equations
|
|
Thanks Matt,
The problem you'll run into with fitting a GEE with GENLIN is that you won't be able to model your "facility effect" as a random effect because GENLIN doesn't support random effects. If you can ignore the facility effect or can fit it as a fixed effect, then I think you could fit a GEE with GENLIN specifying your patients as subjects and the services as within-subjects effects (if I'm reading the below properly) -- however, I wouldn't really recommend this without checking against a generalized linear mixed model with facility as a random effect... which GENLIN unfortunately cannot currently fit. So I think you want to look at some of the programs Paul suggested below. There are also a number of potentially promising R packages out there (googling "generalized linear mixed model r" turns up nearly 7 million hits), but I haven't personally used these packages. Good luck, Alex -----Original Message----- From: SPSSX(r) Discussion [mailto:[hidden email]] On Behalf Of Pirritano, Matthew Sent: Monday, March 30, 2009 2:15 PM To: [hidden email] Subject: Re: Generalized Estimating Equations My data is patient level data with a variable that identifies the facility. It is detailed claims data. So there is information about all procedures. The subject is the patient. Each line of data is a service. I have read that this type of data is at the intersection of GEEs and mixed models, and that SPSS cannot do this. But being that I only have SPSS is there a work around or a next best fit for my data? Thanks matt Matthew Pirritano, Ph.D. Research Analyst IV Medical Services Initiative (MSI) Orange County Health Care Agency (714) 568-5648 -----Original Message----- From: SPSSX(r) Discussion [mailto:[hidden email]] On Behalf Of Reutter, Alex Sent: Monday, March 30, 2009 11:52 AM To: [hidden email] Subject: Re: Generalized Estimating Equations Matt, Could you tell the list a little more about the structure of the dataset? At what level are you measuring whether a procedure was performed; on a patient-by-patient basis, or on a facility-by-facility basis, or something else? It's not clear to me what your "subjects" are, so while it's possible that you might not be able to use GEE, it's also possible that you might. Cheers, Alex -----Original Message----- From: SPSSX(r) Discussion [mailto:[hidden email]] On Behalf Of Swank, Paul R Sent: Monday, March 30, 2009 12:48 PM To: [hidden email] Subject: Re: Generalized Estimating Equations Actually, this sounds like a non-linear mixed or hierarchical model. I don't believe SPSS has that yet but HLM, Mlwin, or SAS will do it. Dr. Paul R. Swank, Professor and Director of Research Children's Learning Institute University of Texas Health Science Center-Houston -----Original Message----- From: SPSSX(r) Discussion [mailto:[hidden email]] On Behalf Of Pirritano, Matthew Sent: Monday, March 30, 2009 12:37 PM To: [hidden email] Subject: Generalized Estimating Equations Hello listers, I would like to do a longitudinal logistic regression. I have health data, and I would like to see if a procedure is administered or not over time and at different facilities. I guess that makes it longitudinal and multilevel. Should I be doing generalized estimating equations for this analysis? And if so, can someone recommend a good book for this type of analysis that provides some guidance for GEEs with SPSS? Thanks Matt Matthew Pirritano, Ph.D. Research Analyst IV Medical Services Initiative (MSI) Orange County Health Care Agency (714) 568-5648 ===================== 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 |
Re: Generalized Estimating Equations
|
|
In reply to this post by Swank, Paul R
Thanks paul I had though tthat the genearl lienar model in SPSS was intended to bring SPSS up to speed in this area but it appears that this is not the case Bill William N. Dudley, PhD Associate Dean for Research The School of Health and Human Performance Office of Research The University of North Carolina at Greensboro 126 HHP Building, PO Box 26170 Greensboro, NC 27402-6170 VOICE 336.2562475 FAX 336.334.3238
Actually, this sounds like a non-linear mixed or hierarchical model. I don't believe SPSS has that yet but HLM, Mlwin, or SAS will do it. Dr. Paul R. Swank, Professor and Director of Research Children's Learning Institute University of Texas Health Science Center-Houston -----Original Message----- From: SPSSX(r) Discussion [mailto:[hidden email]] On Behalf Of Pirritano, Matthew Sent: Monday, March 30, 2009 12:37 PM To: [hidden email] Subject: Generalized Estimating Equations Hello listers, I would like to do a longitudinal logistic regression. I have health data, and I would like to see if a procedure is administered or not over time and at different facilities. I guess that makes it longitudinal and multilevel. Should I be doing generalized estimating equations for this analysis? And if so, can someone recommend a good book for this type of analysis that provides some guidance for GEEs with SPSS? Thanks Matt Matthew Pirritano, Ph.D. Research Analyst IV Medical Services Initiative (MSI) Orange County Health Care Agency (714) 568-5648 ===================== 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 |
Re: Generalized Estimating Equations
|
|
Here’s a quick and dirty rundown:
The current gap is in coverage of
generalized linear mixed models. Alex From: SPSSX(r)
Discussion [mailto:[hidden email]] On
Behalf Of William Dudley WNDUDLEY
|
|
|
In reply to this post by William Dudley WNDUDLEY
Listers, I am having a variety of issues since
upgrading to PASW. There are coming up too often to repeatedly announce them on
the list. One general issue that I can’t quite put my finger on is that
sometimes syntax seems to not run without any obvious reason. After trying for
sometime to get it to run it will then unexplainably decide to run. I will give
more details the next time this happens. Maybe PASW is more sensitive to unwanted
spaces and the like that are lurking unseen at the fringes of syntax. But more specifically it appears that the
Output Export command cannot be run from syntax. Not that big a deal but a
little annoying. Thanks Matt Matthew Pirritano, Ph.D. Research Analyst IV Medical Services Initiative (MSI) Orange County Health Care Agency (714) 568-5648 |
|
|
Are you saying that the syntax is not submitted or generates
errors. If the former—you might be setting a breakpoint in the syntax editor
by clicking in the left margin. If the latter, tell us the syntax that failed
and what the error message was. What happens to your OUTPUT EXPORT command? From: SPSSX(r) Discussion
[mailto:[hidden email]] On Behalf Of Pirritano, Matthew Listers, I am having a variety of issues since upgrading to PASW. There are
coming up too often to repeatedly announce them on the list. One general issue
that I can’t quite put my finger on is that sometimes syntax seems to not
run without any obvious reason. After trying for sometime to get it to run it
will then unexplainably decide to run. I will give more details the next time
this happens. Maybe PASW is more sensitive to unwanted spaces and the like that
are lurking unseen at the fringes of syntax. But more specifically it appears that the Output Export command
cannot be run from syntax. Not that big a deal but a little annoying. Thanks Matt Matthew Pirritano, Ph.D. Research Analyst IV Medical Services Initiative (MSI) Orange County Health Care Agency (714) 568-5648 |
|
|
Thanks ViAnn,
No, I know what a breakpoint is and I am not setting one. If I select export from point and click and click on OK instead of Paste it runs just fine. If I paste it and try and run it from syntax I get an export report that is simply an empty box and no document (.rtf in this case) is produced. Thanks Matt Email: [hidden email] From: ViAnn Beadle <[hidden email]> To: [hidden email] Sent: Tuesday, April 14, 2009 5:32:30 PM Subject: Re: probs with PASW Are you saying that the syntax is not submitted or generates errors. If the former—you might be setting a breakpoint in the syntax editor by clicking in the left margin. If the latter, tell us the syntax that failed and what the error message was. What happens to your OUTPUT EXPORT command?
From: SPSSX(r) Discussion
[mailto:[hidden email]] On Behalf Of Pirritano, Matthew
Listers,
I am having a variety of issues since upgrading to PASW. There are coming up too often to repeatedly announce them on the list. One general issue that I can’t quite put my finger on is that sometimes syntax seems to not run without any obvious reason. After trying for sometime to get it to run it will then unexplainably decide to run. I will give more details the next time this happens. Maybe PASW is more sensitive to unwanted spaces and the like that are lurking unseen at the fringes of syntax.
But more specifically it appears that the Output Export command cannot be run from syntax. Not that big a deal but a little annoying.
Thanks Matt
Matthew Pirritano, Ph.D. Research Analyst IV Medical Services Initiative (MSI) Orange County Health Care Agency (714) 568-5648 |
|
|
So the only syntax that doesn’t run is the OUTPUT EXPORT
command? Compare what is run from the dialog by looking at the syntax echoed in
the log and the syntax that you pasted—what differences do you see? Can you
post both to the listserv? From: SPSSX(r) Discussion
[mailto:[hidden email]] On Behalf Of Matthew Pirritano Thanks ViAnn, Matthew Pirritano, Ph.D. From: ViAnn Beadle <[hidden email]> Are you saying that the syntax is not submitted or generates
errors. If the former—you might be setting a breakpoint in the syntax editor by
clicking in the left margin. If the latter, tell us the syntax that failed and
what the error message was. What happens to your OUTPUT EXPORT command? From: SPSSX(r) Discussion
[mailto:[hidden email]] On Behalf Of Pirritano, Matthew Listers, I am having a variety of issues since upgrading to PASW. There are coming
up too often to repeatedly announce them on the list. One general issue that I
can’t quite put my finger on is that sometimes syntax seems to not run without
any obvious reason. After trying for sometime to get it to run it will then
unexplainably decide to run. I will give more details the next time this
happens. Maybe PASW is more sensitive to unwanted spaces and the like that are
lurking unseen at the fringes of syntax. But more specifically it appears that the Output Export command
cannot be run from syntax. Not that big a deal but a little annoying. Thanks Matt Matthew Pirritano, Ph.D. Research Analyst IV Medical Services Initiative (MSI) Orange County Health Care Agency (714) 568-5648 |
|
|
In reply to this post by mpirritano
I have a number of contaminant concentration variables (ca. 25) which are coded as above detectable limit (1) or below detection limit (0). I'm trying to get the percent of subjects in which each of these contaminants were detectable, broken down by 3 grouping variables (community, gender, and age group).
I'm pretty sure there must be a way to do this, maybe in CROSSTABS or FREQUENCIES, but all I can get are individual tables for each of the 25 contaminants, whereas I'd really like them all in one table. Here's the syntax that gives me individual tables for each of the contaminants (a restricted set of 6 PCBs in this example): CROSSTABS /TABLES=depcb138 depcb153 depcb156 depcb170 depcb180 depcb183 BY agegroup BY communit BY sex_ok /FORMAT= AVALUE TABLES /CELLS= COUNT . Any pointers on how to get all 6 (or all 25 variables) in one table with CROSSTABS or FREQUENCIES? thanks, Ian Ian D. Martin, Ph.D. Tsuji Laboratory University of Waterloo Dept. of Environment & Resource Studies |
|
|
If you don’t have the tables module which does this very
easily, try MULT RESPONSE. Think of the contaminant variables as a dichotomous
multiple response group with the value 1 as a counting value. From: SPSSX(r) Discussion
[mailto:[hidden email]] On Behalf Of Ian Martin I have a number of contaminant concentration variables (ca.
25) which are coded as above detectable limit (1) or below detection limit
(0). I'm trying to get the percent of subjects in which each of these
contaminants were detectable, broken down by 3 grouping variables (community,
gender, and age group). I'm pretty sure there must be a way to do this, maybe in
CROSSTABS or FREQUENCIES, but all I can get are individual tables for each of
the 25 contaminants, whereas I'd really like them all in one table.
Here's the syntax that gives me individual tables for each of the contaminants
(a restricted set of 6 PCBs in this example): CROSSTABS /TABLES=depcb138 depcb153 depcb156
depcb170 depcb180 depcb183 BY agegroup BY communit BY sex_ok /FORMAT= AVALUE TABLES /CELLS= COUNT . Any pointers on how to get all 6 (or all 25 variables) in
one table with CROSSTABS or FREQUENCIES? thanks, Ian Ian D. Martin, Ph.D. Tsuji Laboratory University of Waterloo Dept. of Environment & Resource Studies
|
| Free forum by Nabble | Edit this page |