Tests of Model Effects vs. Parameter Estimates using GEE

Moderators: statman, Analyst Techy, andris, Fierce, GerineL, Smash

Nicolas

Tests of Model Effects vs. Parameter Estimates using GEE

Postby Nicolas » Wed Nov 19, 2008 1:04 pm

Hi everybody,

I ran a GEE (Generalized Estimation Equations) in SPSS 16 with 2 within-subject variables and a few between-subject variables. Now, I'm not sure how to interpret the output of the "Tests of Model Effects" vs. the "Parameter Estimates". A dichotomous between-subject variable is not significant regarding the "Tests of Model Effects", but significant regarding the "Parameter Estimates". I know this is due to interaction-effects, but which of the two output I have to interpret?

I'm pretty clueless, any help is really appreciated!

Thx,
Nicolas
statman
Administrator
Posts: 2757
Joined: Tue Jun 12, 2007 12:08 pm
Location: Florida, USA

Postby statman » Wed Nov 19, 2008 3:32 pm

So am I. Can you insert a graphic of the outcome tables then perhaps I can help
See the note below

NOTE: Please read the Posting Guidelines and always tell us your OS, the SPSS version and information about your study and data!

Statman
Statistical Services
Nicolas Berkowitsch
Posts: 4
Joined: Wed Nov 19, 2008 8:23 pm

Output of Tests of Model Effects vs. Parameter Estimates

Postby Nicolas Berkowitsch » Thu Nov 20, 2008 1:01 pm

Hi statman,

The following variables were included:
dependent variable.
- EPS: "which earnings per share do you announce"? A [=1] or B [=2]?
within-subject variables:
- sitSV: 5 different situations; extend of opportunity costs choosing A over B, [5=1.20$, 4=0.90$, 3=0.60$, 2=0.30$, 1=0.00$]
- phase: the 5 situations were asked twice (after receiving additonal information)
between-subject variables:
- c_SV: centered scale variabe, assessing attitude toward honesty
- framing: subject was assigned to either positive[=3], negative[=2], or to no framing at all [=1]
- sex: female[=2]/male[=1]
- fach: economics[=2]/psychology[=1]
Nicolas Berkowitsch
Posts: 4
Joined: Wed Nov 19, 2008 8:23 pm

Postby Nicolas Berkowitsch » Thu Nov 20, 2008 1:01 pm

The following link (next post) will guide you to the outputs of the tables, saved either as a *.pdf (tables of the "Model Effects" and Parameter Estimates" included) or as *.spv (the output file itself)
Nicolas Berkowitsch
Posts: 4
Joined: Wed Nov 19, 2008 8:23 pm

Postby Nicolas Berkowitsch » Thu Nov 20, 2008 1:02 pm

http://www.psychologie.uzh.ch/cognisoz/files/

Thx, for having a look at this
Nicolas
Last edited by Nicolas Berkowitsch on Fri Nov 21, 2008 9:53 am, edited 1 time in total.
statman
Administrator
Posts: 2757
Joined: Tue Jun 12, 2007 12:08 pm
Location: Florida, USA

Postby statman » Thu Nov 20, 2008 7:21 pm

These are displaying two different results so can differ but first, what type of model did you apply? Can you post your syntax?
Nicolas Berkowitsch
Posts: 4
Joined: Wed Nov 19, 2008 8:23 pm

Syntax to GEE

Postby Nicolas Berkowitsch » Fri Nov 21, 2008 7:57 am

Hi statman,

I ran a binomial GEE with a logit link, first-order autoregressive correlation structure with a robust estimator was assumed due to the design.
Here is the SPSS-Syntax:

________

* Generalized Estimating Equations.
GENLIN choice_EPS (REFERENCE=FIRST) BY sitSV framing phase sex fach (ORDER=DESCENDING) WITH c_SV
/MODEL sitSV c_SV framing phase sex fach sitSV*framing sitSV*phase sitSV*sex sitSV*fach
sitSV*c_SV framing*phase framing*sex framing*fach framing*c_SV phase*sex phase*fach phase*c_SV
sex*fach sex*c_SV fach*c_SV sitSV*framing*c_SV sitSV*phase*c_SV INTERCEPT=YES
DISTRIBUTION=BINOMIAL LINK=LOGIT
/CRITERIA METHOD=FISHER(1) SCALE=1 MAXITERATIONS=100 MAXSTEPHALVING=5 PCONVERGE=1E-006(ABSOLUTE)
SINGULAR=1E-012 ANALYSISTYPE=3(WALD) CILEVEL=95 LIKELIHOOD=FULL
/EMMEANS TABLES=sitSV SCALE=ORIGINAL
/EMMEANS TABLES=framing SCALE=ORIGINAL
/EMMEANS TABLES=phase SCALE=ORIGINAL
/EMMEANS TABLES=sex SCALE=ORIGINAL
/EMMEANS TABLES=framing*phase SCALE=ORIGINAL
/REPEATED SUBJECT=ID WITHINSUBJECT=phase*sitSV SORT=YES CORRTYPE=AR(1) ADJUSTCORR=YES COVB=ROBUST
MAXITERATIONS=100 PCONVERGE=1e-006(ABSOLUTE) UPDATECORR=1
/MISSING CLASSMISSING=EXCLUDE
/PRINT CPS DESCRIPTIVES MODELINFO FIT SUMMARY SOLUTION (EXPONENTIATED) WORKINGCORR.

________

Thx,
Nicolas
statman
Administrator
Posts: 2757
Joined: Tue Jun 12, 2007 12:08 pm
Location: Florida, USA

Postby statman » Fri Nov 21, 2008 3:01 pm

Since I don't get into the logic model too often my inclination is that 'overall' phase isn't significant (as the Type III table), however, when split out and compared between P1 & P2 with P2 as the reference then P1 is (the Parameter table). Again, I might be thinking statistically and not based on the model you actually applied.

Sorry, that's about it

Who is online

Users browsing this forum: No registered users and 1 guest

cron