comparing methods for ordinal response variables

[Kornbrot, Diana]

hi

SPSS has [at least] 3 different procedures: generalized linear model with cumulative logic link; regression, ordinal, aka PLUM; mixed -without specifying a subject variable [but i want to extend it later, so am checking]. syntax is below

concerns: different answers form the 3 procedures

1. all give same parameter values - that's good
2. generalised linear and PLUM are very similar, but not identical: give Wald chi-square as inferential BUT WALD values slightly different & hence se and confidence limits also slightly different
3. PLUM has very useful option to test for parallel lines and estimate scale if lines not parallel. Means may be obtained from 'cell information'
4. BUT PLUM does NOT give factorial main and interaction effects. Is there any way to do this in syntax? the TEST syntax seems geared only to contrasts, not to a complete factor
5. Generalized linear models gives inferential effects for main effects and interactions, but no means - just parameter estimates
6. MIXED uses F for inferential tests, and t-tests to estimate CLs [why one wonders]. the standard errors are MUCH bigger than for PLUM and Generalized linear
7. MIXED appears much less sensitive than Generalized linear and some effects are not significant
8. Generalized linear won't do means or comparisons for ordinal, not even via syntax, although it does it fine for binary logistics. WHY?
9. Advice on which procedure to trust and/or why MIXED gives such  very different inferential results much appreciate

___________

* Generalized Linear Models.

/GENLIN Response (ORDER=ASCENDING) BY cohort Dno Goal2catN HEI_TypeN (ORDER=ASCENDING)

  /MODEL cohort Dno Goal2catN HEI_TypeN cohort*Dno cohort*Goal2catN cohort*HEI_TypeN Dno*Goal2catN Dno*HEI_TypeN Goal2catN*HEI_TypeN cohort*Dno*Goal2catN cohort*Dno*HEI_TypeN cohort*Goal2catN*HEI_TypeN Dno*Goal2catN*HEI_TypeN

 DISTRIBUTION=MULTINOMIAL LINK=CUMLOGIT

  /CRITERIA METHOD=FISHER(1) SCALE=1 COVB=MODEL MAXITERATIONS=100 MAXSTEPHALVING=5 PCONVERGE=1E-006(ABSOLUTE) SINGULAR=1E-012 ANALYSISTYPE=3(WALD) CILEVEL=95 CITYPE=WALD LIKELIHOOD=FULL

  /MISSING CLASSMISSING=EXCLUDE

  /PRINT DESCRIPTIVES MODELINFO FIT SUMMARY SOLUTION.

* GENLINMIXED

  /FIELDS TARGET=Response TRIALS=NONE OFFSET=NONE

  /TARGET_OPTIONS DISTRIBUTION=MULTINOMIAL LINK=LOGIT

  /FIXED  EFFECTS=cohort Dno Goal2catN HEI_TypeN cohort*Dno cohort*Goal2catN cohort*HEI_TypeN Dno*Goal2catN Dno*HEI_TypeN Goal2catN*HEI_TypeN cohort*Dno*Goal2catN cohort*Dno*HEI_TypeN cohort*Goal2catN*HEI_TypeN Dno*Goal2catN*HEI_TypeN cohort*Dno*Goal2catN*HEI_TypeN USE_INTERCEPT=TRUE

  /BUILD_OPTIONS TARGET_CATEGORY_ORDER=ASCENDING INPUTS_CATEGORY_ORDER=ASCENDING MAX_ITERATIONS=100 CONFIDENCE_LEVEL=95 DF_METHOD=RESIDUAL COVB=MODEL PCONVERGE=0.000001(ABSOLUTE) SCORING=0 SINGULAR=0.000000000001

  /EMMEANS TABLES=cohort CONTRAST=NONE

   /EMMEANS TABLES=Dno COMPARE=Dno CONTRAST=PAIRWISE

PLUM Response BY cohort Dno Goal2catN HEI_TypeN

  /CRITERIA=CIN(95) DELTA(0) LCONVERGE(0) MXITER(100) MXSTEP(5) PCONVERGE(1.0E-6) SINGULAR(1.0E-8)

  /LINK=LOGIT

  /LOCATION=cohort Dno Goal2catN HEI_TypeN Dno*Goal2catN Dno*HEI_TypeN Dno*cohort Goal2catN*HEI_TypeN Goal2catN*cohort HEI_TypeN*cohort Dno*Goal2catN*HEI_TypeN Dno*Goal2catN*cohort Dno*HEI_TypeN*cohort Goal2catN*HEI_TypeN*cohort Dno*Goal2catN*HEI_TypeN*cohort

  /PRINT=CELLINFO FIT PARAMETER SUMMARY TPARALLEL.

____________________

best

Diana 

_______________

Professor Diana Kornbrot

University of Hertfordshire

College Lane, Hatfield, Hertfordshire AL10 9AB, UK

[phone]

[phone]

[phone] 

[hidden email]

http://dianakornbrot.wordpress.com/

http://go.herts.ac.uk/Diana_Kornbrot

skype:  kornbrotme_______________________________

===================== 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