Help in interpreting coefficients in an ordered regression

[William Rossi]

Dear all,
I am studying the effect of alcohol consumption on self reported health (SRH). SRH takes 4 values: 1(Very bad or bad), 2(neither good nor bad, 3(good), 4 (very good) health. The other regressors are prevalently dummy variables or categorical variables such as age or instruction.
I have already run different specifications both with regress, mlogit and ologit and the preferred specification (which has lower AIC and BIC) is the partial proportional odds model.

Below there is the code of the model with just some of my regressors (I tested the model even with more variables for health status, familiy relationship and sports but the results do not differ much). For spce reasons I cut some of the regressors from the output that I am showing, if necessary I can add the complete output.

Generalized Ordered Logit Estimates                  Number of obs  =  207,957
                                                     Wald chi2(106) = 72035.97
                                                     Prob > chi2    =   0.0000
Log pseudolikelihood = -190330.26                    Pseudo R2      =   0.2127

                                (Std. err. adjusted for 107,996 clusters in family)
-----------------------------------------------------------------------------------
                  |               Robust
              SRH | Coefficient  std. err.      z    P>|z|     [95% conf. interval]
------------------+----------------------------------------------------------------
1                 |
       Drunk_year |   .5748232   .0222563    25.83   0.000     .5312017    .6184446
          disease |  -3.397611   .0691435   -49.14   0.000    -3.533129   -3.262092
           smoker |   .0710709   .0120859     5.88   0.000     .0473829    .0947589
                  |
           female |    .024049   .0217941     1.10   0.270    -.0186666    .0667645
          foreign |  -.0230205   .0182156    -1.26   0.206    -.0587224    .0126815
life_satisfaction |   .3624923   .0056357    64.32   0.000     .3514465    .3735381
            _cons |   2.442286   .0960758    25.42   0.000      2.25398    2.630591
------------------+----------------------------------------------------------------
2                 |
       Drunk_year |   .2306524   .0134163    17.19   0.000     .2043569    .2569478
          disease |  -1.734568   .0132714  -130.70   0.000    -1.760579   -1.708556
           smoker |   .0710709   .0120859     5.88   0.000     .0473829    .0947589
                  |
           female |  -.1618877   .0115242   -14.05   0.000    -.1844747   -.1393007
          foreign |  -.0016579   .0082389    -0.20   0.841    -.0178058      .01449
life_satisfaction |    .276836   .0038655    71.62   0.000     .2692598    .2844123
            _cons |  -.8607331    .047251   -18.22   0.000    -.9533432   -.7681229
------------------+----------------------------------------------------------------
3                 |
       Drunk_year |  -.1008911    .016192    -6.23   0.000    -.1326268   -.0691553
          disease |  -1.296835   .0177248   -73.16   0.000    -1.331575   -1.262095
           smoker |   .0710709   .0120859     5.88   0.000     .0473829    .0947589
                  |
           female |  -.2563435   .0132369   -19.37   0.000    -.2822873   -.230399
          foreign |   .0198676   .0082947     2.40   0.017     .0036103    .0361249
life_satisfaction |   .2282275   .0055746    40.94   0.000     .2173015    .2391536
            _cons |  -4.065911   .0595232   -68.31   0.000    -4.182574   -3.949247
-----------------------------------------------------------------------------------

The main coefficient of interest is Drunk_year which is a dummy which takes value of 1 if the respondent has drunk at least once in the past year. I would interpret it as follows:

1. Since in the first panel the coefficient is positive it means that it is more likely that the respondent will be in neither good nor bad, good or very good health categories if he drinks.
2. The third panel coefficient is negative, therefore it means that drinking makes it more likely to be in a lower category, which is reasonable.

From these results it looks like that drinking increases the probability of having medium values of health (neither good nor bad) bu
Am I right? Probably I have some difficulties in interpreting these results since the coefficient for Drinking behaves the opposite with respect to the coefficient for having a disease. I would really appreciate an help in untangling this issue.
Kind regards,
William