meologit - problem with interpretation

[Janet Hill]

My dependent variable is a discrete scal in increments of 1 with several categorical independent variables with each patient measured twice so I thought that meologit would be an appropriate response. In the full dataset there are 302 patients. In this data subset mn is the patient identifier


----------------------- copy starting from the next line -----------------------

* Example generated by -dataex-. For more info, type help dataex
clear
input long mn byte x1 double(x16 rater)
53210 1  -7 2
98415 1   3 2
30352 1 -12 2
79393 1  -9 2
21324 1 -14 2
59239 1   0 2
99470 1  -3 2
54362 1  -2 2
89914 1  -9 2
14512 1   1 2
95465 1 -10 2
23362 1  -2 2
86785 1  -4 2
74015 1 -11 2
36960 1  -4 2
60797 1 -13 2
26959 1  -8 2
36257 2 -17 2
52235 2  -2 2
38264 2  -7 2
35697 2   0 2
33592 2  -5 2
50949 2 -15 2
61932 2  -7 2
25519 2  -4 2
53210 1  -7 1
98415 1   1 1
30352 1 -12 1
79393 1  -9 1
21324 1 -12 1
59239 1   0 1
99470 1  -5 1
54362 1  -1 1
89914 1  -6 1
14512 1  -1 1
95465 1 -13 1
23362 1  -3 1
86785 1  -4 1
74015 1 -11 1
36960 1  -4 1
60797 1 -13 1
26959 1 -10 1
36257 2 -16 1
52235 2  -5 1
38264 2  -8 1
35697 2  -1 1
33592 2  -5 1
50949 2 -13 1
61932 2 -10 1
25519 2  -4 1
end

------------------ copy up to and including the previous line ------------------

If I ignore the duplicate measurements then I get the following results:

. meologit x16 i.x1 if rater == 1
------------------------------------------------------------------------------
         x16 | Coefficient  Std. err.      z    P>|z|     [95% conf. interval]
-------------+----------------------------------------------------------------
          x1 |
          1  |      0.000  (base)
          2  |     -0.465      0.748   -0.622   0.534       -1.931       1.001
-------------+----------------------------------------------------------------
. meologit x16 i.x1 if rater == 2
------------------------------------------------------------------------------
         x16 | Coefficient  Std. err.      z    P>|z|     [95% conf. interval]
-------------+----------------------------------------------------------------
          x1 |
          1  |      0.000  (base)
          2  |     -0.207      0.760   -0.272   0.785       -1.697       1.283
-------------+----------------------------------------------------------------
. meologit x16 i.x1
------------------------------------------------------------------------------
         x16 | Coefficient  Std. err.      z    P>|z|     [95% conf. interval]
-------------+----------------------------------------------------------------
          x1 |
          1  |      0.000  (base)
          2  |     -0.354      0.532   -0.665   0.506       -1.396       0.689
-------------+----------------------------------------------------------------

The coefficients of x1 are the same order of magnitude. However when I try to allow for the duplicate measurements on the same subject they are an order of magnitude larger:

. meologit x16 i.x1 ||mn:, nolog
------------------------------------------------------------------------------
         x16 | Coefficient  Std. err.      z    P>|z|     [95% conf. interval]
-------------+----------------------------------------------------------------
          x1 |
          1  |      0.000  (base)
          2  |     -2.376      3.243   -0.733   0.464       -8.732       3.979
-------------+----------------------------------------------------------------

. meologit x16 i.x1 i.rater ||mn:, nolog
------------------------------------------------------------------------------
         x16 | Coefficient  Std. err.      z    P>|z|     [95% conf. interval]
-------------+----------------------------------------------------------------
          x1 |
          1  |      0.000  (base)
          2  |     -2.402      3.355   -0.716   0.474       -8.977       4.173
             |
       rater |
          1  |      0.000  (base)
          2  |      0.736      0.560    1.315   0.189       -0.361       1.834
-------------+----------------------------------------------------------------

I am surprised that the coefficient has changed that much as I have not observed this behaviour before when using meologit (or mixed) and I thought that only the 'variance' calculations would be significantly affected; so I have two questions:
1. Am I using meologit correctly
2. Are these results to be expected.

Thank you for any advice.

Janet