Several Questions Ordinal Regression and Interactions

[Mwenza Blell]

Hi.
I am using SPSS 14.0 to perform analyses on data collected for my dissertation
and have a few questions/problems. I am using ordinal logistic regression (OLR)
with a complementary log-log link function (the highest level of my outcome is
the most probable as it’s the biggest group – I also tried out one of my models
with logit and the fit was better with clog-log).

Question 1: Can someone tell me whether there is an easy way to compute an odds
ratio from the coefficients of OLR with complementary log-log? I found a paper
which showed a confusing equation which seemed to be for an odds ratio from OLR
with complementary log-log link but I did not understand the equation and wasn’t
sure that I was interpreting what it was correctly.

I suspected there might be interactions between factors and so I used the
function to test for interactions (Analyze>Regression>Ordinal>Location> Specify
Model: Custom >Build Term(s): Interaction). I got two different significant
interactions. Each interaction was made up of two dichotomous categorical
variables each coded 0 and 1, for each interaction there were four lines in the
output table representing the combinations 0-0, 0-1, 1-0, and 1-1 as well as
lines representing 0 and 1 of each of the “main effect variables.” The odd thing
is that everything seems to change depending whether I put the interaction first
in the dialog box or if put one or both of the main effect variables in first.
The parameter estimate output doesn’t just change order, the meaning of the
output changes, though other things stay the same (model fitting chi-squared,
model p, pseudo R-squared statistics, goodness of fit chi-squared).  In one of
the ones I have done, there is a significant result for one category which moves
to a different category when I change the order and the coefficient, SE, and p
value are the same. Also, I end up with 5 redundant/reference categories (which
move around when the order changes as well, sometimes applying to the main
effects, sometimes to levels of the interaction). When I put both of the
significant interactions in the model together and change the order of entry
even more dramatic changes happen in the parameter estimate box.

Question 2: Why does putting the variables in a different order change the
output so much? Is a particular order of entry correct? If not, how do I
interpret the differences in the output?

Question 3: Why so many redundant categories? How do I interpret coefficients
when there are multiple redundant categories for the same variable? (Previously
I have treated the redundant category as the reference category and interpreted
with reference to it e.g. if the categories of the independent were blue, black,
and red and red was the redundant one, I interpreted the coefficients of both
blue and black as relating to the increase or decrease in the probability of
being in a higher category of the dependent variable compared to red. This might
have been incorrect, I am not sure.)

Please, fellow list members, answer any of these that you can at your earliest
convenience, I have searched everywhere for explanations and found none.

Thanks very much,

Mwenza Blell

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

[Mark A Davenport MADAVENP]

All,

 

I need to solve this probelm fast.  Hope some of you have answers to this.

 

Using versions pre-16, I could use the Text Import Wizard to read in a fixed width text file (suffix .dat) then, in the last step, tell it to write a syntax file so I can read a like file later on.

This runs perfectly in v14 and v15.  In v 16, the steps and options look to be the same.  When I click the final OK, I get the data set fine.  However, when I try to run the syntax generated by the data wizard, it gives me a blank file data file and the following errors:

 

GET DATA
  /TYPE=TXT
  /FILE='C:\SPSS_Course\Dataset_1.dat'
  /FIXCASE=1
  /ARRANGEMENT=FIXED
  /FIRSTCASE=1
  /IMPORTCASE=ALL
  /VARIABLES=
  /1 Age 0-1 F2.0
  Ethnic 2-9 A8
  Gender 10-10 A1
  College 11-40 A30
  itme1 41-41 F1.0
  item2 42-42 F1.0
  item3 43-43 F1.0
  V8 44-45 2X
  item4 46-47 F2.0
  item5 48-48 F1.0
  V11 49-48 0X.

>Error.  Command name: GET DATA
>(2265) Unrecognized or invalid variable definition.
>This command not executed.
CACHE.
EXECUTE.

>Error # 105.  Command name: EXECUTE
>This command is not valid before a working file has been defined.
>This command not executed.
DATASET NAME DataSet1 WINDOW=FRONT.

The weird thing is, if I do the exact same thing but DO NOT ask the Wizard to change the fixed comuln break lines, I get the following and the syntax runs fine.

 

GET DATA
  /TYPE=TXT
  /FILE='C:\SPSS_Course\Dataset_1.dat'
  /FIXCASE=1
  /ARRANGEMENT=FIXED
  /FIRSTCASE=1
  /IMPORTCASE=ALL
  /VARIABLES=
  /1 V1 0-6 A7
  V2 7-16 A10
  V3 17-19 A3
  V4 20-29 A10
  V5 30-43 A14
  V6 44-48 F5.0.
CACHE.
EXECUTE.
DATASET NAME DataSet3 WINDOW=FRONT.

 

Of course, my columns aren't right, in this case.  I seem to recall asking about this once before but can't fine the response in the archives.  What gives?

***************************************************************************************************************************************************************
Mark A. Davenport Ph.D.
Senior Research Analyst
Office of Institutional Research
The University of North Carolina at Greensboro
336.256.0395
[hidden email]

'An approximate answer to the right question is worth a good deal more than an exact answer to an approximate question.' --a paraphrase of J. W. Tukey (1962)

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