I am running deviation contrasts in csglm and cannot find a way to get confidence intervals (CIs) for the contrasts in the output (only gives p-values). I can get CIs for the category estimates (means) by specifying 'CINTERVAL' in  the /statistics subcommand, but this does not also produce CIs for the contrast estimates (i.e., in the Individual Test Results table). I see no way to specify CIs on the /test subcommand or elsewhere. Is there a way to request this that I have missed?

My syntax code:
CSGLM q13_AC_g BY partyid7
/PLAN FILE='H:\data\lia2022_v4.csaplan'
/MODEL partyid7
/INTERCEPT INCLUDE=YES SHOW=YES
/STATISTICS PARAMETER SE CINTERVAL TTEST
/PRINT SUMMARY VARIABLEINFO SAMPLEINFO
/TEST TYPE=F PADJUST=BONFERRONI
/EMMEANS TABLES=partyid7 COMPARE=partyid7 CONTRAST=DEVIATION (7)
/EMMEANS
/MISSING CLASSMISSING=EXCLUDE
/CRITERIA CILEVEL=95.

My results below (just the relevant parts). I got CIs for the Estimates table (this is great) by specifying in /statistics subcommand. I would also like to get CIs for the contrast estimates in the Individual Test Results (bottom table). Does anyone know if/how it is possible to also get CIs for the 'contrast estimates' in the  Individual Test Results table?

Many thanks,
Andrew

------------------------------
Andrew Crawford
------------------------------

#SPSSStatistics

Hi. I passed this on to one of our statisticians, who replied:

"I don't see a way to request CIs on comparisons of EMMEANs either. Really not sure why we didn't include CIs for those or for CUSTOM output. We do provide CIs for the parameter estimates in the model, so in some models there is a way to get these results, which is to reparameterize the model so that the parameter estimates provide the desired contrasts. For a one-way ANOVA like this, that's simple enough to do (assuming partyid7 is a numeric variable with values 1-7):

COMPUTE Dev1=(partyid7=1)-(partyid7=7).
COMPUTE Dev2=(partyid7=2)-(partyid7=7).
COMPUTE Dev3=(partyid7=3)-(partyid7=7).
COMPUTE Dev5=(partyid7=5)-(partyid7=7).
COMPUTE Dev6=(partyid7=6)-(partyid7=7).

CSGLM q13_AC_g WITH Dev1 Dev2 Dev3 Dev5 Dev6
 /PLAN FILE='H:\data\lia2022_v4.csaplan'
 /MODEL Dev1 Dev2 Dev3 Dev5 Dev6
 /INTERCEPT INCLUDE=YES SHOW=YES
 /STATISTICS PARAMETER SE CINTERVAL TTEST
 /PRINT SUMMARY VARIABLEINFO SAMPLEINFO

 /MISSING CLASSMISSING=EXCLUDE
 /CRITERIA CILEVEL=99.

The COMPUTE commands create the design or basis variables required to reproduce the deviation contrasts for the original partyid7 factor. The CILEVEL=99 specification changes the CI levels to give the same results as the Bonferroni method (which divides the original .05 associated with the 95% intervals by 5)."

------------------------------
Rick Marcantonio
Quality Assurance
IBM
------------------------------

Original Message:
Sent: Tue November 22, 2022 06:43 PM
From: Andrew Crawford
Subject: complex samples glm

I am running deviation contrasts in csglm and cannot find a way to get confidence intervals (CIs) for the contrasts in the output (only gives p-values). I can get CIs for the category estimates (means) by specifying 'CINTERVAL' in  the /statistics subcommand, but this does not also produce CIs for the contrast estimates (i.e., in the Individual Test Results table). I see no way to specify CIs on the /test subcommand or elsewhere. Is there a way to request this that I have missed?

My syntax code:
CSGLM q13_AC_g BY partyid7
/PLAN FILE='H:\data\lia2022_v4.csaplan'
/MODEL partyid7
/INTERCEPT INCLUDE=YES SHOW=YES
/STATISTICS PARAMETER SE CINTERVAL TTEST
/PRINT SUMMARY VARIABLEINFO SAMPLEINFO
/TEST TYPE=F PADJUST=BONFERRONI
/EMMEANS TABLES=partyid7 COMPARE=partyid7 CONTRAST=DEVIATION (7)
/EMMEANS
/MISSING CLASSMISSING=EXCLUDE
/CRITERIA CILEVEL=95.

My results below (just the relevant parts). I got CIs for the Estimates table (this is great) by specifying in /statistics subcommand. I would also like to get CIs for the contrast estimates in the Individual Test Results (bottom table). Does anyone know if/how it is possible to also get CIs for the 'contrast estimates' in the  Individual Test Results table?

Many thanks,
Andrew

------------------------------
Andrew Crawford
------------------------------

#SPSSStatistics

Original Message:
Sent: 11/28/2022 3:26:00 PM
From: Rick Marcantonio
Subject: RE: complex samples glm

Hi. I passed this on to one of our statisticians, who replied:

"I don't see a way to request CIs on comparisons of EMMEANs either. Really not sure why we didn't include CIs for those or for CUSTOM output. We do provide CIs for the parameter estimates in the model, so in some models there is a way to get these results, which is to reparameterize the model so that the parameter estimates provide the desired contrasts. For a one-way ANOVA like this, that's simple enough to do (assuming partyid7 is a numeric variable with values 1-7):

COMPUTE Dev1=(partyid7=1)-(partyid7=7).
COMPUTE Dev2=(partyid7=2)-(partyid7=7).
COMPUTE Dev3=(partyid7=3)-(partyid7=7).
COMPUTE Dev5=(partyid7=5)-(partyid7=7).
COMPUTE Dev6=(partyid7=6)-(partyid7=7).

CSGLM q13_AC_g WITH Dev1 Dev2 Dev3 Dev5 Dev6
 /PLAN FILE='H:\data\lia2022_v4.csaplan'
 /MODEL Dev1 Dev2 Dev3 Dev5 Dev6
 /INTERCEPT INCLUDE=YES SHOW=YES
 /STATISTICS PARAMETER SE CINTERVAL TTEST
 /PRINT SUMMARY VARIABLEINFO SAMPLEINFO

 /MISSING CLASSMISSING=EXCLUDE
 /CRITERIA CILEVEL=99.

The COMPUTE commands create the design or basis variables required to reproduce the deviation contrasts for the original partyid7 factor. The CILEVEL=99 specification changes the CI levels to give the same results as the Bonferroni method (which divides the original .05 associated with the 95% intervals by 5)."

------------------------------
Rick Marcantonio
Quality Assurance
IBM
------------------------------

Original Message:
Sent: Tue November 22, 2022 06:43 PM
From: Andrew Crawford
Subject: complex samples glm

I am running deviation contrasts in csglm and cannot find a way to get confidence intervals (CIs) for the contrasts in the output (only gives p-values). I can get CIs for the category estimates (means) by specifying 'CINTERVAL' in  the /statistics subcommand, but this does not also produce CIs for the contrast estimates (i.e., in the Individual Test Results table). I see no way to specify CIs on the /test subcommand or elsewhere. Is there a way to request this that I have missed?

My syntax code:
CSGLM q13_AC_g BY partyid7
/PLAN FILE='H:\data\lia2022_v4.csaplan'
/MODEL partyid7
/INTERCEPT INCLUDE=YES SHOW=YES
/STATISTICS PARAMETER SE CINTERVAL TTEST
/PRINT SUMMARY VARIABLEINFO SAMPLEINFO
/TEST TYPE=F PADJUST=BONFERRONI
/EMMEANS TABLES=partyid7 COMPARE=partyid7 CONTRAST=DEVIATION (7)
/EMMEANS
/MISSING CLASSMISSING=EXCLUDE
/CRITERIA CILEVEL=95.

My results below (just the relevant parts). I got CIs for the Estimates table (this is great) by specifying in /statistics subcommand. I would also like to get CIs for the contrast estimates in the Individual Test Results (bottom table). Does anyone know if/how it is possible to also get CIs for the 'contrast estimates' in the  Individual Test Results table?

Many thanks,
Andrew

------------------------------
Andrew Crawford
------------------------------

#SPSSStatistics