How do I get Beta in GENLIN?
Locked
 
|
GENLIN gives the unstandardized coefficient of B. Does anyone know whether it is possible to get the
Beta? How? I'm using this syntax (for a negative binomial regression model): GENLIN A WITH B C /MODEL B C INTERCEPT=YES DISTRIBUTION=NEGBIN(MLE) LINK=LOG. Thank you in advance for your reply! ===================== 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 |
|
|
Are you asking how to get the GENLIN procedure to exponentiate the unstandardized coefficients for you? If yes, then the answer is to add an OUTPUT statement.
Ryan
On Thu, Oct 14, 2010 at 7:15 AM, Sytske <[hidden email]> wrote: GENLIN gives the unstandardized coefficient of B. Does anyone know whether it is possible to get the |
|
|
Of course, you could standardize "B" and "C" before entering them into the model, if that's what you're after. Please clarify exactly what you're trying to obtain.
Ryan
On Thu, Oct 14, 2010 at 12:00 PM, R B <[hidden email]> wrote:
|
Re: How do I get Beta in GENLIN?
Administrator
|
Ryan, I think the OP is following the convention that B = raw (unstandardized) coefficient, and Beta = standardized coefficient. So your suggestion below (standardizing the variables before entering them in the model) should do the trick.
Bruce
--
Bruce Weaver bweaver@lakeheadu.ca http://sites.google.com/a/lakeheadu.ca/bweaver/ "When all else fails, RTFM." PLEASE NOTE THE FOLLOWING: 1. My Hotmail account is not monitored regularly. To send me an e-mail, please use the address shown above. 2. The SPSSX Discussion forum on Nabble is no longer linked to the SPSSX-L listserv administered by UGA (https://listserv.uga.edu/). |
|
|
Hi Bruce,
As I know you know, in linear regression, if interested in obtaining the "standardized coefficient," one could certainly standardize both the response and predictor variable before entering them into the model. In such a scenario, the unstandardized regression coefficient actually becomes the standardized regression coefficient. The REGRESSION procedure spits out standardized regression coefficients, so there really is no need to do standarize the variables before entering them into the model, except for purposes of learning.
Linear regression assumes the link between the continuous response and predictor variable is an identity link. However, with a count regression model such as the negative binomial model, the response is assumed to be a count variable, and is typically linked to the predictor using a log link. Although I haven't tested what I'm about to state, my guess is that the GENLIN procedure does not permit values with decimals in the response when specifying a negative binomial distribution. As a result, I do not believe it is possible to standardize the response before entering it into the model via the GENLIN procedure. This is why I did not suggest standardizing the variable, "A" in my previous post. One can certainly standardize the predictor(s), as I indicated in the previous post. In this scenario, I would argue that the coefficient is partially standardized.
Anyway, I just wanted to be clear that my suggestion in the previous post to standardize the predictors will not result in a fully standardized coefficient.
Ryan
On Thu, Oct 14, 2010 at 2:33 PM, Bruce Weaver <[hidden email]> wrote: Ryan, I think the OP is following the convention that B = raw |
Re: How do I get Beta in GENLIN?
|
Administrator
|
Yes, you're right, Ryan. I was forgetting that the Y-variable also has to be standardized for regression to yield the standardized regression coefficients. I don't know what happens when you standardize the Y variable when the model uses something other than an identity link--I've never tried it.
Bruce
--
Bruce Weaver bweaver@lakeheadu.ca http://sites.google.com/a/lakeheadu.ca/bweaver/ "When all else fails, RTFM." PLEASE NOTE THE FOLLOWING: 1. My Hotmail account is not monitored regularly. To send me an e-mail, please use the address shown above. 2. The SPSSX Discussion forum on Nabble is no longer linked to the SPSSX-L listserv administered by UGA (https://listserv.uga.edu/). |
|
To see what Ryan and Bruce are saying, open a new instance of SPSS.
Copy the syntax below into a syntax window. Run it.
GET FILE='C:\Program Files\SPSSInc\PASWStatistics18\Samples\English\world95.sav'. DATASET NAME DataSet1 WINDOW=FRONT. descriptive variables = babymort calories fertilty/save. DATASET ACTIVATE DataSet1. REGRESSION /MISSING LISTWISE /STATISTICS COEFF /DEPENDENT babymort /METHOD=ENTER calories fertilty. REGRESSION /MISSING LISTWISE /STATISTICS COEFF /DEPENDENT babymort /METHOD=ENTER Zcalories Zfertilty. REGRESSION /MISSING LISTWISE /STATISTICS COEFF /DEPENDENT Zbabymort /METHOD=ENTER Zcalories Zfertilty. Art Kendall Social Research Consultants On 10/15/2010 9:10 AM, Bruce Weaver wrote: ===================== 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 REFCARDYes, you're right, Ryan. I was forgetting that the Y-variable also has to be standardized for regression to yield the standardized regression coefficients. I don't know what happens when you standardize the Y variable when the model uses something other than an identity link--I've never tried it. Bruce R B wrote:Hi Bruce, As I know you know, in linear regression, if interested in obtaining the "standardized coefficient," one could certainly standardize both the response and predictor variable before entering them into the model. In such a scenario, the unstandardized regression coefficient actually becomes the standardized regression coefficient. The REGRESSION procedure spits out standardized regression coefficients, so there really is no need to do standarize the variables before entering them into the model, except for purposes of learning. Linear regression assumes the link between the continuous response and predictor variable is an identity link. However, with a count regression model such as the negative binomial model, the response is assumed to be a count variable, and is typically linked to the predictor using a log link. Although I haven't tested what I'm about to state, my guess is that the GENLIN procedure does not permit values with decimals in the response when specifying a negative binomial distribution. As a result, I do not believe it is possible to standardize the response before entering it into the model via the GENLIN procedure. This is why I did not suggest standardizing the variable, "A" in my previous post. One can certainly standardize the predictor(s), as I indicated in the previous post. In this scenario, I would argue that the coefficient is partially standardized. Anyway, I just wanted to be clear that my suggestion in the previous post to standardize the predictors will not result in a fully standardized coefficient. Ryan On Thu, Oct 14, 2010 at 2:33 PM, Bruce Weaver [hidden email]wrote:Ryan, I think the OP is following the convention that B = raw (unstandardized) coefficient, and Beta = standardized coefficient. So your suggestion below (standardizing the variables before entering them in the model) should do the trick. Bruce R B wrote:Of course, you could standardize "B" and "C" before entering them intothemodel, if that's what you're after. Please clarify exactly what you're trying to obtain. Ryan On Thu, Oct 14, 2010 at 12:00 PM, R B [hidden email]wrote:Are you asking how to get the GENLIN procedure to exponentiate the unstandardized coefficients for you? If yes, then the answer is to addanOUTPUT statement. Ryan On Thu, Oct 14, 2010 at 7:15 AM, Sytske [hidden email] wrote:GENLIN gives the unstandardized coefficient of B. Does anyone know whether it is possible to get the Beta? How? I'm using this syntax (for a negative binomial regression model): GENLIN A WITH B C /MODEL B C INTERCEPT=YES DISTRIBUTION=NEGBIN(MLE) LINK=LOG. Thank you in advance for your reply! ===================== To manage your subscription to SPSSX-L, send a message to [hidden email] (not to SPSSX-L), with no body text exceptthecommand. To leave the list, send the command SIGNOFF SPSSX-L For a list of commands to manage subscriptions, send the command INFO REFCARD----- -- Bruce Weaver [hidden email] http://sites.google.com/a/lakeheadu.ca/bweaver/ "When all else fails, RTFM." NOTE: My Hotmail account is not monitored regularly. To send me an e-mail, please use the address shown above. -- View this message in context: http://spssx-discussion.1045642.n5.nabble.com/How-do-I-get-Beta-in-GENLIN-tp3211903p3212595.html Sent from the SPSSX Discussion mailing list archive at Nabble.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----- -- Bruce Weaver [hidden email] http://sites.google.com/a/lakeheadu.ca/bweaver/ "When all else fails, RTFM." NOTE: My Hotmail account is not monitored regularly. To send me an e-mail, please use the address shown above. -- View this message in context: http://spssx-discussion.1045642.n5.nabble.com/How-do-I-get-Beta-in-GENLIN-tp3211903p3213615.html Sent from the SPSSX Discussion mailing list archive at Nabble.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
Art Kendall
Social Research Consultants |
|
|
In reply to this post by Bruce Weaver
Hi all-
I've created a little program below to concretize the point I was making previously. If you run the code below you will notice that the GENLIN procedure produces an error message when the dependent variable takes on values other than non-negative integers if a Poisson distribution is specified. Same issue applies to a negative binomial distribution.
HTH,
Ryan
--
*Generate Data.
set seed 98765432. new file. inp pro. loop ID= 1 to 1000. comp x1 = rv.normal(2,1). comp b0 = -0.5.
comp b1 = 1.0. comp lambda = b0 + b1*x1. comp mu = exp(lambda). comp y = rv.poisson(mu). end case. end loop. end file. end inp pro. exe. Delete variables b0 b1 lambda mu. *Fit model using y.
GENLIN y WITH x1 /MODEL x1 INTERCEPT=YES DISTRIBUTION=POISSON LINK=LOG /PRINT CPS DESCRIPTIVES MODELINFO FIT SUMMARY SOLUTION . *Standardize y.
DESCRIPTIVES VARIABLES=y
/SAVE. *Fit model using standardized y.
GENLIN Zy WITH x1 /MODEL x1 INTERCEPT=YES DISTRIBUTION=POISSON LINK=LOG /PRINT CPS DESCRIPTIVES MODELINFO FIT SUMMARY SOLUTION . On Fri, Oct 15, 2010 at 9:10 AM, Bruce Weaver <[hidden email]> wrote: Yes, you're right, Ryan. I was forgetting that the Y-variable also has to be |
|
|
Hi all,
Thank you very much for your thoughts and sorry for my late reply. I am referring to the standardized regression coefficient beta for the predictor variables in my analysis. As I understand it now, you suggested to standardize the variables before entering them into the regression analysis, but then you wrote that this does not work for GENLIN (with negative binomial regression). Shouldn't it be possible to get the standardized effect size measure BETA without first standardizing the variables before you put them in the analysis? Or to calculate BETA after you have run the analysis? If you run a standard linear regression in SPSS, the output gives Beta automatically. Thanks in advance for your reply, Sytske On 18 Oct 2010, at 01:55, R B wrote:
|
|
|
I have tried your syntax where you standardize the variables before putting them in the regression. I applied this to my negative binomial regression in GEE, but it doesn't work (as you already expected).
So, should I assume that it is not possible to get a standardized regression coefficient from negative binomial regression in GEE? Isn't that weird? I still hope there is a way in which I can get a standardized regression coefficient, so I can actually compare different regression coefficients? I can hardly believe that would not be possible. Sytske |
Re: How do I get Beta in GENLIN?
|
Administrator
|
Here's a note I wrote for myself after reading John Fox's comment on standardized regression coefficients. You may find it interesting. --- start of note --- In his book "Applied Regression Analysis and Generalized Linear Models" (2008, Sage), John Fox is very cautious about the use of standardized regression coefficients. He gives this interesting example. When two variables are measured on the same scale (e.g.,years of education, and years of employment), then relative impact of the two can be compared directly. But suppose those two variables differ substantially in the amount of spread. In that case, comparison of the standardized regression coefficients would likely yield a very different story than comparison of the raw regression coefficients. Fox then says: "If expressing coefficients relative to a measure of spread potentially distorts their comparison when two explanatory variables are commensurable [i.e., measured on the same scale], then why should the procedure magically allow to compare coefficients [for variables] that are measured in different units?" (p. 95) Good question! --- end of note --- HTH.
--
Bruce Weaver bweaver@lakeheadu.ca http://sites.google.com/a/lakeheadu.ca/bweaver/ "When all else fails, RTFM." PLEASE NOTE THE FOLLOWING: 1. My Hotmail account is not monitored regularly. To send me an e-mail, please use the address shown above. 2. The SPSSX Discussion forum on Nabble is no longer linked to the SPSSX-L listserv administered by UGA (https://listserv.uga.edu/). |
|
|
In reply to this post by Sytske
Syske,
The Negative Binomial (NB) regression model is a count model that is considered an extension of the Poisson regression model. The NB model allows the conditional variance to be larger than the conditional mean, while the Poisson model assumes the conditional mean and variance are equal. I make this point because the formula I'm about to provide is intended to be used within the context of a Poisson regression model. It may very well work with the NB model, though I cannot say for sure. Here's one way to obtain the standardized regression coefficient in in a Poisson regression model: (unstandardized beta) * (sd of predictor variable) ---------------------------------------------------------------------- (sd of linear prediction) Click on the link below for further details: http://www.ats.ucla.edu/stat/mplus/faq/stdcoefs.htm Ryan On Thu, Jan 20, 2011 at 7:02 AM, Sytske <[hidden email]> wrote: > I have tried your syntax where you standardize the variables before putting > them in the regression. I applied this to my negative binomial regression in > GEE, but it doesn't work (as you already expected). > > So, should I assume that it is not possible to get a standardized regression > coefficient from negative binomial regression in GEE? Isn't that weird? > > I still hope there is a way in which I can get a standardized regression > coefficient, so I can actually compare different regression coefficients? I > can hardly believe that would not be possible. > > Sytske > > -- > View this message in context: http://spssx-discussion.1045642.n5.nabble.com/How-do-I-get-Beta-in-GENLIN-tp3211903p3349471.html > Sent from the SPSSX Discussion mailing list archive at Nabble.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 > ===================== 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 |
|
|
Dear Ryan,
Thanks for the formula. I am not sure I can use it though, since I am using negative binomial regression analysis in GEE (GENLIN). I have a sample consisting of siblings and GEE takes the shared family variance into account in the analysis. So, I would think I cannot 'just' use the standard deviation, can I? Sytske On 20 Jan 2011, at 21:44, R B wrote:
Sytske Besemer Ph.D. Candidate Institute of Criminology University of Cambridge Sidgwick Avenue Cambridge CB3 9DA United Kingdom |
|
|
Sytske,
I, too, would be cautious about applying the formula I provided previously to such a model. Unfortunately, I don't have an immediate solution for you. You could probably obtain standardized estimates with other software (e.g. SAS), if it's that important to you. I have to echo Bruce's concerns about standardized estimates, especially if they're being used to determine "relative importance." Interpretation of standardized estimates in such a way can be quite misleading. Ryan On Fri, Jan 21, 2011 at 11:23 AM, Sytske Besemer <[hidden email]> wrote: > Dear Ryan, > Thanks for the formula. I am not sure I can use it though, since I am using > negative binomial regression analysis in GEE (GENLIN). I have a sample > consisting of siblings and GEE takes the shared family variance into account > in the analysis. So, I would think I cannot 'just' use the standard > deviation, can I? > Sytske > On 20 Jan 2011, at 21:44, R B wrote: > > Syske, > > The Negative Binomial (NB) regression model is a count model that is > considered an extension of the Poisson regression model. The NB model > allows the conditional variance to be larger than the conditional > mean, while the Poisson model assumes the conditional mean and > variance are equal. > > I make this point because the formula I'm about to provide is intended > to be used within the context of a Poisson regression model. It may > very well work with the NB model, though I cannot say for sure. Here's > one way to obtain the standardized regression coefficient in in a > Poisson regression model: > > (unstandardized beta) * (sd of predictor variable) > ---------------------------------------------------------------------- > (sd of linear prediction) > > > Click on the link below for further details: > > http://www.ats.ucla.edu/stat/mplus/faq/stdcoefs.htm > > Ryan > > On Thu, Jan 20, 2011 at 7:02 AM, Sytske <[hidden email]> wrote: > > I have tried your syntax where you standardize the variables before putting > > them in the regression. I applied this to my negative binomial regression in > > GEE, but it doesn't work (as you already expected). > > So, should I assume that it is not possible to get a standardized regression > > coefficient from negative binomial regression in GEE? Isn't that weird? > > I still hope there is a way in which I can get a standardized regression > > coefficient, so I can actually compare different regression coefficients? I > > can hardly believe that would not be possible. > > Sytske > > -- > > View this message in context: > http://spssx-discussion.1045642.n5.nabble.com/How-do-I-get-Beta-in-GENLIN-tp3211903p3349471.html > > Sent from the SPSSX Discussion mailing list archive at Nabble.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 > > > ===================== > 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 > > Sytske Besemer > Ph.D. Candidate > Institute of Criminology > University of Cambridge > Sidgwick Avenue > Cambridge > CB3 9DA > United Kingdom > ( +44 (0)1223 335360 > * [hidden email] > > > > > > ===================== 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: How do I get Beta in GENLIN?
|
Administrator
|
In reply to this post by Bruce Weaver
While I was looking at John Fox's book (Applied Regression Analysis and Generalized Linear Models, 2008, Sage) regarding the quote I posted elsewhere in this thread, I came across this comment, that pertains to the discussion Ryan and I were having back in October.
"The usual practice standardizes the response variable as well, but this is an inessential element of the computation of standardized coefficients, because the relative size of the slope coefficients does not change when Y is rescaled." (p. 95)
--
Bruce Weaver bweaver@lakeheadu.ca http://sites.google.com/a/lakeheadu.ca/bweaver/ "When all else fails, RTFM." PLEASE NOTE THE FOLLOWING: 1. My Hotmail account is not monitored regularly. To send me an e-mail, please use the address shown above. 2. The SPSSX Discussion forum on Nabble is no longer linked to the SPSSX-L listserv administered by UGA (https://listserv.uga.edu/). |
|
|
The information about not needing to standardize the response should
prove useful to the OP. I've decided to create a simulation program below to show how to (1) generate data from a Poisson regression model (incorporating correlated responses), (2) fit a GEE model entering all of the predictors in their original, unstandardized form, and then (3) fit a GEE model after standardizing two of the predictors. Ryan -- *Generate data. set seed 4836210. new file. inp pro. comp personID = -99. comp person_random_effect = -99. comp subtest = -99. comp b0 = -1.10. comp b1 = -1.25. comp b2 = 1.20. comp b3 = 0.50. comp b4 = 0.75. comp person_variance = 0.30. leave personID to person_variance. loop personID = 1 to 1000. comp person_random_effect = sqrt(person_variance)*rv.normal(0,1). loop subtest = 1 to 3. comp x1 = rv.normal(1,4). comp x2 = rv.normal(1,2). comp lambda = b0 + b1*(subtest=1) + b2*(subtest=2) + b3*x1 + b4*x2 + person_random_effect. comp mu = exp(lambda). comp y = rv.poisson(mu). end case. end loop. end loop. end file. end inp pro. exe. delete variables person_random_effect b0 b1 b2 b3 b4 person_variance lambda mu. *Fit model. GENLIN y WITH x1 x2 BY subtest (ORDER=ASCENDING) /MODEL subtest x1 x2 INTERCEPT=YES DISTRIBUTION=POISSON LINK=LOG /REPEATED SUBJECT=personID WITHINSUBJECT=subtest SORT=YES CORRTYPE=EXCHANGEABLE ADJUSTCORR=YES /PRINT CPS DESCRIPTIVES MODELINFO FIT SUMMARY SOLUTION (EXPONENTIATED). *Standardize two of the predictors. DESCRIPTIVES VARIABLES=x1 x2 /SAVE. *Fit model. GENLIN y WITH Zx1 Zx2 BY subtest (ORDER=ASCENDING) /MODEL subtest Zx1 Zx2 INTERCEPT=YES DISTRIBUTION=POISSON LINK=LOG /REPEATED SUBJECT=personID WITHINSUBJECT=subtest SORT=YES CORRTYPE=EXCHANGEABLE ADJUSTCORR=YES /PRINT CPS DESCRIPTIVES MODELINFO FIT SUMMARY SOLUTION (EXPONENTIATED). On Fri, Jan 21, 2011 at 5:27 PM, Bruce Weaver <[hidden email]> wrote: > While I was looking at John Fox's book (Applied Regression Analysis and > Generalized Linear Models, 2008, Sage) regarding the quote I posted > elsewhere in this thread, I came across this comment, that pertains to the > discussion Ryan and I were having back in October. > > "The usual practice standardizes the response variable as well, but this is > an inessential element of the computation of standardized coefficients, > because the relative size of the slope coefficients does not change when Y > is rescaled." (p. 95) > > > > > Bruce Weaver wrote: >> >> Yes, you're right, Ryan. I was forgetting that the Y-variable also has to >> be standardized for regression to yield the standardized regression >> coefficients. I don't know what happens when you standardize the Y >> variable when the model uses something other than an identity link--I've >> never tried it. >> >> Bruce >> >> >> R B wrote: >>> >>> Hi Bruce, >>> >>> As I know you know, in linear regression, if interested in obtaining the >>> "standardized coefficient," one could certainly standardize both the >>> response and predictor variable before entering them into the model. In >>> such >>> a scenario, the unstandardized regression coefficient actually becomes >>> the >>> standardized regression coefficient. The REGRESSION procedure spits out >>> standardized regression coefficients, so there really is no need to do >>> standarize the variables before entering them into the model, except for >>> purposes of learning. >>> >>> Linear regression assumes the link between the continuous response and >>> predictor variable is an identity link. However, with a count regression >>> model such as the negative binomial model, the response is assumed to be >>> a >>> count variable, and is typically linked to the predictor using a log >>> link. >>> Although I haven't tested what I'm about to state, my guess is that the >>> GENLIN procedure does not permit values with decimals in the response >>> when >>> specifying a negative binomial distribution. As a result, I do not >>> believe >>> it is possible to standardize the response before entering it into the >>> model >>> via the GENLIN procedure. This is why I did not suggest standardizing the >>> variable, "A" in my previous post. One can certainly standardize the >>> predictor(s), as I indicated in the previous post. In this scenario, I >>> would >>> argue that the coefficient is partially standardized. >>> >>> Anyway, I just wanted to be clear that my suggestion in the previous post >>> to >>> standardize the predictors will not result in a fully standardized >>> coefficient. >>> >>> Ryan >>> >>> On Thu, Oct 14, 2010 at 2:33 PM, Bruce Weaver >>> <[hidden email]>wrote: >>> >>>> Ryan, I think the OP is following the convention that B = raw >>>> (unstandardized) coefficient, and Beta = standardized coefficient. So >>>> your >>>> suggestion below (standardizing the variables before entering them in >>>> the >>>> model) should do the trick. >>>> >>>> Bruce >>>> >>>> >>>> R B wrote: >>>> > >>>> > Of course, you could standardize "B" and "C" before entering them into >>>> the >>>> > model, if that's what you're after. Please clarify exactly what you're >>>> > trying to obtain. >>>> > >>>> > Ryan >>>> > >>>> > On Thu, Oct 14, 2010 at 12:00 PM, R B <[hidden email]> >>>> wrote: >>>> > >>>> >> Are you asking how to get the GENLIN procedure to exponentiate the >>>> >> unstandardized coefficients for you? If yes, then the answer is to >>>> add >>>> an >>>> >> OUTPUT statement. >>>> >> >>>> >> Ryan >>>> >> On Thu, Oct 14, 2010 at 7:15 AM, Sytske <[hidden email]> wrote: >>>> >> >>>> >>> GENLIN gives the unstandardized coefficient of B. Does anyone know >>>> >>> whether >>>> >>> it is possible to get the >>>> >>> Beta? How? >>>> >>> I'm using this syntax (for a negative binomial regression model): >>>> >>> >>>> >>> GENLIN >>>> >>> A WITH B C >>>> >>> /MODEL B C >>>> >>> INTERCEPT=YES >>>> >>> DISTRIBUTION=NEGBIN(MLE) >>>> >>> LINK=LOG. >>>> >>> >>>> >>> Thank you in advance for your reply! >>>> >>> >>>> >>> ===================== >>>> >>> 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 >>>> >>> >>>> >> >>>> >> >>>> > >>>> > >>>> >>>> >>>> ----- >>>> -- >>>> Bruce Weaver >>>> [hidden email] >>>> http://sites.google.com/a/lakeheadu.ca/bweaver/ >>>> >>>> "When all else fails, RTFM." >>>> >>>> NOTE: My Hotmail account is not monitored regularly. >>>> To send me an e-mail, please use the address shown above. >>>> >>>> -- >>>> View this message in context: >>>> http://spssx-discussion.1045642.n5.nabble.com/How-do-I-get-Beta-in-GENLIN-tp3211903p3212595.html >>>> Sent from the SPSSX Discussion mailing list archive at Nabble.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 >>>> >>> >>> >> >> > > > ----- > -- > Bruce Weaver > [hidden email] > http://sites.google.com/a/lakeheadu.ca/bweaver/ > > "When all else fails, RTFM." > > NOTE: My Hotmail account is not monitored regularly. > To send me an e-mail, please use the address shown above. > > -- > View this message in context: http://spssx-discussion.1045642.n5.nabble.com/How-do-I-get-Beta-in-GENLIN-tp3211903p3352332.html > Sent from the SPSSX Discussion mailing list archive at Nabble.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 > ===================== 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 |
|
|
In reply to this post by Bruce Weaver
Dear all,
I have a research design problem. I'd be thankful if you helped. I read 20 words for 30 subjects and asked them to tell me the first word that comes to their mind. I categorized the words generated by the participants under 3 linguistic categories: Category A, Category B and Category C. 29% of the 30 words produced for the word "train" were in Cat. A. 19% of the words produced for the word "book" were in Cat. A. 25% of the words produced for the word "computer" were in Cat. A. The mean of these percentages over all the 20 words was calculated. The mean of Cat. A was 27%. In Phase 2, I gave pictures of the same words as prompts to the same subjects (instead of words) and asked them to tell me the first word that comes to their mind. I categorized the words in the same way and computed the means of the three categories in the same way. Results show that the mean of Cat. A when words are used as prompts is 27% and its mean is 13% when pictures are used as prompts. This suggests that words generate more Cat. A concepts than pictures. How can I test the statistical significance of this difference? Can I use paired t-test for this? Cheers Anthony ===================== 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 |
|
|
It sounds like you have 30 cases with 2 repeated measure design
factors. One factor is concept with 20 levels. The other is mode
presentation with 2 levels, text and picture. You have a dependent
variable, category, with 3 values. The DV is nominal or ordinal.
Please let the list know whether this is correct and if you have an ordinal or nominal DV. It appears that you are throwing a lot information by summing (collapsing) before checking whether you have an interaction of word by mode. If I am correct about your design, I would familiarize myself with the results by looking at the crosstab of mode by category by word ignoring any statistics except for counts and percentages. Then to test the relation of mode, word, and mode by word, I would check whether Categorical Regression, Correspondence Analysis, multinomial regression, and/or General Estimating Equations can handle doubly repeated designs. Be aware that these methods of analysis grew out of different dialects of statistics. In GEE parlance, "repeated" corresponds to "correlated" in the sense of correlated inclusion, not in the sense of the relation of variables. I don't have SPSS on this computer and I am not where my books are. In multinomial regression parlance, IIRC "repeated" corresponds to "contingent". It is something like contingent. Perhaps another list participant can check this. Perhaps other list participants know off hand whether with a) a strictly nominal DV or b) with an ordinal DV, if any/all of these can handle doubly repeated designs 1) categorical regression 2) correspondence analysis 3) GEE 4) multinomial regression Art Kendall Social Research Consultants On 1/23/2011 3:16 AM, Anthony James wrote: ===================== 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 REFCARDDear all, I have a research design problem. I'd be thankful if you helped. I read 20 words for 30 subjects and asked them to tell me the first word that comes to their mind. I categorized the words generated by the participants under 3 linguistic categories: Category A, Category B and Category C. 29% of the 30 words produced for the word "train" were in Cat. A. 19% of the words produced for the word "book" were in Cat. A. 25% of the words produced for the word "computer" were in Cat. A. The mean of these percentages over all the 20 words was calculated. The mean of Cat. A was 27%. In Phase 2, I gave pictures of the same words as prompts to the same subjects (instead of words) and asked them to tell me the first word that comes to their mind. I categorized the words in the same way and computed the means of the three categories in the same way. Results show that the mean of Cat. A when words are used as prompts is 27% and its mean is 13% when pictures are used as prompts. This suggests that words generate more Cat. A concepts than pictures. How can I test the statistical significance of this difference? Can I use paired t-test for this? Cheers Anthony ===================== 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
Art Kendall
Social Research Consultants |
|
|
In reply to this post by Ryan
Thanks for your thoughts Ryan and Bruce. I've tried your program Ryan and it works for me: using standardized predictors, but the original outcome variable in GENLIN works.
Thanks very much! Sytske |
«
Return to SPSSX Discussion
|
1 view|%1 views
| Free forum by Nabble | Edit this page |