

Hi all,
I am running a Univariate GLM. My single dependable variable is continuous
and my independent variables are categorical. I have 4 independent
variables. I want to compare models of which combination of independent
variable best explain the response variable. Could anyone tell me how
could I get the AIC or BIC values of the models in the output in SPSS.
Thank you
Nabaneeta
=====================
To manage your subscription to SPSSXL, send a message to
[hidden email] (not to SPSSXL), with no body text except the
command. To leave the list, send the command
SIGNOFF SPSSXL
For a list of commands to manage subscriptions, send the command
INFO REFCARD

Administrator

Nabaneeta Saha wrote
Hi all,
I am running a Univariate GLM. My single dependable variable is continuous
and my independent variables are categorical. I have 4 independent
variables. I want to compare models of which combination of independent
variable best explain the response variable. Could anyone tell me how
could I get the AIC or BIC values of the models in the output in SPSS.
Thank you
Nabaneeta
You'll have to "roll your own", so to speak. I.e., compute it yourself. For AIC, you need N, k (the number of parameters fit by the model, including the intercept), and RSS (the residual sum of squares). For BIC, things are a bit more complicated; so unless you really want it for some reason, I'd stick with AIC (or the second order corrected version for smaller samples).
http://en.wikipedia.org/wiki/Akaike_information_criterion http://en.wikipedia.org/wiki/Bayesian_information_criterionThis reminds me of something I've been meaning to suggest to the good folks at SPSS: I think that AIC (and corrected AIC) would be very nice additions to the CURVEFIT procedure.

Administrator

Bruce Weaver wrote
Nabaneeta Saha wrote
Hi all,
I am running a Univariate GLM. My single dependable variable is continuous
and my independent variables are categorical. I have 4 independent
variables. I want to compare models of which combination of independent
variable best explain the response variable. Could anyone tell me how
could I get the AIC or BIC values of the models in the output in SPSS.
Thank you
Nabaneeta
You'll have to "roll your own", so to speak. I.e., compute it yourself. For AIC, you need N, k (the number of parameters fit by the model, including the intercept), and RSS (the residual sum of squares). For BIC, things are a bit more complicated; so unless you really want it for some reason, I'd stick with AIC (or the second order corrected version for smaller samples).
http://en.wikipedia.org/wiki/Akaike_information_criterion http://en.wikipedia.org/wiki/Bayesian_information_criterionThis reminds me of something I've been meaning to suggest to the good folks at SPSS: I think that AIC (and corrected AIC) would be very nice additions to the CURVEFIT procedure.
On the way home, I had another idea. If you use the MIXED procedure to run your ANOVA, it will spit out AIC, BIC, and a few other measures of fit. Here's an example:
http://www.angelfire.com/wv/bwhomedir/spss/mixed001.txtHTH.
p.s.  AIC and related measures would STILL be nice additions to CURVEFIT!


Nabaneeta,
Within the ordinary least squares regression framework,
AIC = n*log(SSE/n)+2(k+1)
and
BIC = n*log(SSE/n) + (k+1) * log(n)
where
n = sample size
SSE= sum of squared errors
k = number of predictors
Ryan
On Tue, Oct 12, 2010 at 4:37 PM, Nabaneeta Saha <[hidden email]> wrote:
Hi all,
I am running a Univariate GLM. My single dependable variable is continuous and my independent variables are categorical. I have 4 independent
variables. I want to compare models of which combination of independent variable best explain the response variable. Could anyone tell me how could I get the AIC or BIC values of the models in the output in SPSS.
Thank you
Nabaneeta
===================== To manage your subscription to SPSSXL, send a message to [hidden email] (not to SPSSXL), with no body text except the
command. To leave the list, send the command SIGNOFF SPSSXL For a list of commands to manage subscriptions, send the command INFO REFCARD

Administrator

Bruce Weaver wrote
Bruce Weaver wrote
Nabaneeta Saha wrote
Hi all,
I am running a Univariate GLM. My single dependable variable is continuous
and my independent variables are categorical. I have 4 independent
variables. I want to compare models of which combination of independent
variable best explain the response variable. Could anyone tell me how
could I get the AIC or BIC values of the models in the output in SPSS.
Thank you
Nabaneeta
You'll have to "roll your own", so to speak. I.e., compute it yourself. For AIC, you need N, k (the number of parameters fit by the model, including the intercept), and RSS (the residual sum of squares). For BIC, things are a bit more complicated; so unless you really want it for some reason, I'd stick with AIC (or the second order corrected version for smaller samples).
http://en.wikipedia.org/wiki/Akaike_information_criterion http://en.wikipedia.org/wiki/Bayesian_information_criterionThis reminds me of something I've been meaning to suggest to the good folks at SPSS: I think that AIC (and corrected AIC) would be very nice additions to the CURVEFIT procedure.
On the way home, I had another idea. If you use the MIXED procedure to run your ANOVA, it will spit out AIC, BIC, and a few other measures of fit. Here's an example:
http://www.angelfire.com/wv/bwhomedir/spss/mixed001.txtHTH.
p.s.  AIC and related measures would STILL be nice additions to CURVEFIT!
I didn't read the description of your ANOVA model carefully enough, and had in mind that it was a oneway ANOVA. But now I see you have 4 explanatory variables. You can still use MIXED. Here's an example of a twofactor fully factorial model run via MIXED.
http://www.angelfire.com/wv/bwhomedir/spss/mixed002.txtFor a 4factor model, with explanatory variables AD, it would be something like the following, depending on whether you want the full factorial model or not:
MIXED
Y BY A B C D
/FIXED = A B C D
A*B A*C A*D B*C B*D C*D
A*B*C A*B*D A*C*D B*C*D
A*B*C*D  SSTYPE(3)
/EMMEANS = TABLES(A)
/EMMEANS = TABLES(B)
/EMMEANS = TABLES(C)
etc...
.


It is worth pointing out that the MIXED procedure uses maximum likelihood estimation instead of ordinary least squares estimation. That is, the values of the parameters are estimated by maximizing the likelihood function, not by minimizing the squared difference between the observed and predicted values. The AIC and BIC estimated via the MIXED procedure are partly based on the [log] likelihood function.
Ryan
On Tue, Oct 12, 2010 at 6:31 PM, Bruce Weaver <[hidden email]> wrote:
Bruce Weaver wrote: > > > Nabaneeta Saha wrote: >> >> Hi all, >> >> I am running a Univariate GLM. My single dependable variable is >> continuous >> and my independent variables are categorical. I have 4 independent
>> variables. I want to compare models of which combination of independent >> variable best explain the response variable. Could anyone tell me how >> could I get the AIC or BIC values of the models in the output in SPSS.
>> >> Thank you >> >> Nabaneeta >> >> > > You'll have to "roll your own", so to speak. I.e., compute it yourself. > For AIC, you need N, k (the number of parameters fit by the model,
> including the intercept), and RSS (the residual sum of squares). For BIC, > things are a bit more complicated; so unless you really want it for some > reason, I'd stick with AIC (or the second order corrected version for
> smaller samples). > > http://en.wikipedia.org/wiki/Akaike_information_criterion> http://en.wikipedia.org/wiki/Bayesian_information_criterion
> > > This reminds me of something I've been meaning to suggest to the good > folks at SPSS: I think that AIC (and corrected AIC) would be very nice > additions to the CURVEFIT procedure.
> > > On the way home, I had another idea. If you use the MIXED procedure to run your ANOVA, it will spit out AIC, BIC, and a few other measures of fit. Here's an example:
http://www.angelfire.com/wv/bwhomedir/spss/mixed001.txt
HTH.
p.s.  AIC and related measures would STILL be nice additions to CURVEFIT!
View this message in context: http://spssxdiscussion.1045642.n5.nabble.com/AICinSPSStp3209580p3209741.html
Sent from the SPSSX Discussion mailing list archive at Nabble.com. ===================== To manage your subscription to SPSSXL, send a message to [hidden email] (not to SPSSXL), with no body text except the
command. To leave the list, send the command SIGNOFF SPSSXL For a list of commands to manage subscriptions, send the command INFO REFCARD


Also note that the default estimation
method in MIXED is REML (restricted maximum likelihood), which for models
with just a single residual error parameter produces results that are identical
to the leastsquares results from GLM or UNIANOVA.
Alex
It is worth pointing out that the MIXED procedure uses
maximum likelihood estimation instead of ordinary least squares estimation.
That is, the values of the parameters are estimated by maximizing
the likelihood function, not by minimizing the squared difference between
the observed and predicted values. The AIC and BIC estimated
via the MIXED procedure are partly based on the [log] likelihood
function.
Ryan
On Tue, Oct 12, 2010 at 6:31 PM, Bruce Weaver <bruce.weaver@...>
wrote:
Bruce Weaver wrote:
>
>
> Nabaneeta Saha wrote:
>>
>> Hi all,
>>
>> I am running a Univariate GLM. My single dependable variable is
>> continuous
>> and my independent variables are categorical. I have 4 independent
>> variables. I want to compare models of which combination of independent
>> variable best explain the response variable. Could anyone tell
me how
>> could I get the AIC or BIC values of the models in the output
in SPSS.
>>
>> Thank you
>>
>> Nabaneeta
>>
>>
>
> You'll have to "roll your own", so to speak. I.e.,
compute it yourself.
> For AIC, you need N, k (the number of parameters fit by the model,
> including the intercept), and RSS (the residual sum of squares). For
BIC,
> things are a bit more complicated; so unless you really want it for
some
> reason, I'd stick with AIC (or the second order corrected version
for
> smaller samples).
>
> http://en.wikipedia.org/wiki/Akaike_information_criterion
> http://en.wikipedia.org/wiki/Bayesian_information_criterion
>
>
> This reminds me of something I've been meaning to suggest to the good
> folks at SPSS: I think that AIC (and corrected AIC) would be
very nice
> additions to the CURVEFIT procedure.
>
>
>
On the way home, I had another idea. If you use
the MIXED procedure to run
your ANOVA, it will spit out AIC, BIC, and a few other measures of fit.
Here's an example:
http://www.angelfire.com/wv/bwhomedir/spss/mixed001.txt
HTH.
p.s.  AIC and related measures would STILL be nice additions to CURVEFIT!

