If I understand this correctly, individuals for whom all raters agree
cannot
be included in a -clogit- analysis. That seems to rule it out absolutely
as a way of examining the structure of agreement and disagreement.
There are many ways to look at the data, depending on what data
generation process
you have in mind. It's not clear to me that a binary
response inescapably implies a logit analysis. Missings aside, four
ratings
for each individual imply that your data can be collapsed with loss of
information to a table of the frequencies of 16 joint outcomes:
0000
0001
0010
0011
...
1111
which may allow structure to be discerned. Here "0000" means all four
raters
assign "0", and so forth.
Nick
[email protected]
Margaret R Grove
To clarify further where I think the problem may lie:
PHREG output notes "Number of Observations Read 2300" and "Number of
Observations Used 2000" (300 have missing values for the dependent
variable)
CLOGIT notes that 1710 observations (496 groups) were dropped because of
all negative or all positive outcomes and our final number of
observations is 290!
With this I wonder if comparing the two methods makes sense and which
method (PHREG or CLOGIT) is preferable (if any)?
*
* For searches and help try:
* http://www.stata.com/support/faqs/res/findit.html
* http://www.stata.com/support/statalist/faq
* http://www.ats.ucla.edu/stat/stata/
