Announcement

Collapse
No announcement yet.
X
  • Filter
  • Time
  • Show
Clear All
new posts

  • margins after melogit statistically different, but point estimate of one margin inside confidence interval of another

    Hello, I am working with a large data set on families (1/2 million families). I am using melogit to estimate the association of several covariates with my outcome.

    After running melogit, I run margins--for categorical covariate x, and continuous covariate z--and get the following results:

    Code:
    margins x, at(z = (33.1, 45.5, 55.4, 59.1, 66.3)) post
    
    -----------------------------------------------------------------------------------
                      |            Delta-method
                      |     Margin   std. err.      z    P>|z|     [95% conf. interval]
    ------------------+----------------------------------------------------------------
    _at#x             |
                 1 0  |    .205419   .0163226    12.58   0.000     .1734273    .2374108
                 1 1  |    .276733   .0198932    13.91   0.000     .2377429     .315723
                 2 0  |   .2836444   .0134719    21.05   0.000       .25724    .3100488
                 2 1  |   .3489734    .014977    23.30   0.000      .319619    .3783279
                 3 0  |   .3573103   .0164563    21.71   0.000     .3250566     .389564
                 3 1  |    .412324   .0171394    24.06   0.000     .3787313    .4459166
                 4 0  |   .3868972   .0192173    20.13   0.000      .349232    .4245624
                 4 1  |   .4369036   .0196641    22.22   0.000     .3983627    .4754446
                 5 0  |   .4467064   .0260632    17.14   0.000     .3956234    .4977894
                 5 1  |   .4855433    .026031    18.65   0.000     .4345235    .5365632
    -----------------------------------------------------------------------------------
    Note that for z = 5, the point estimates of the two groups are inside the confidence interval of the other.

    However, when I test the last two margins against each other, I get:

    Code:
    test _b[5._at#0bn.x] == _b[5._at#1.x]
    
     ( 1)  5._at#0bn.x - 5._at#1.x = 0
    
               chi2(  1) =   57.03
             Prob > chi2 =    0.0000
    I am aware of the discussion at https://www.statalist.org/forums/for...ns-overlapping, and elsewhere, that overlapping CI's do not mean that differences can not be statistically significant. However, I had thought, per Greenland et al. (2016), that when the point estimate of one group was inside the confidence interval of another group, that this meant that the two groups could not be different to a degree that was statistically significant.

    Any thoughts on this would be very much appreciated.

    Reference

    Greenland, S., Senn, S. J., Rothman, K. J., Carlin, J. B., Poole, C., Goodman, S. N., & Altman, D. G. (2016). Statistical tests, P values, confidence intervals, and power: a guide to misinterpretations. European Journal of Epidemiology, 31(4), 337–350. https://doi.org/10.1007/s10654-016-0149-3
Working...
X