Negative values in margin command following xtgee

[Muhammad Faisal]

I used xtgee command in stata

xtgee counts week_center2 c.time2 i.flag2,family("poisson") corr(ar1) vce(robust) eform

and margin command on intervention (flag2[before=0,after=1]).

I was wondering why it is showing negative value for lower limit (-.1011584 ) as it is counts (0 to 8)?

Thank you





GEE population-averaged model Number of obs = 731
Group and time vars: id week_center2 Number of groups = 7
Link: log Obs per group:
Family: Poisson min = 89
Correlation: AR(1) avg = 104.4
max = 116
Wald chi2(3) = 0.65
Scale parameter: 1 Prob > chi2 = 0.8842

(Std. Err. adjusted for clustering on id)
------------------------------------------------------------------------------
| Robust
counts | IRR Std. Err. z P>|z| [95% Conf. Interval]
-------------+----------------------------------------------------------------
week_center2 | .9940413 .0083027 -0.72 0.474 .9779008 1.010448
time2 | 1.013084 .018704 0.70 0.481 .9770803 1.050415
1.flag2 | .7877304 .5323796 -0.35 0.724 .2094603 2.962467
_cons | .3142238 .0590433 -6.16 0.000 .2174186 .4541315
------------------------------------------------------------------------------


margin flag2

Predictive margins Number of obs = 731
Model VCE : Robust

Expression : Exponentiated linear prediction, predict()

------------------------------------------------------------------------------
| Delta-method
| Margin Std. Err. z P>|z| [95% Conf. Interval]
-------------+----------------------------------------------------------------
flag2 |
0 | .394614 .1283235 3.08 0.002 .1431045 .6461236
1 | .3108495 .210212 1.48 0.139 -.1011584 .7228574
------------------------------------------------------------------------------




Data example:
flag2 time2 week_center2 counts
0 0 -8 0
0 0 -7 0
0 0 -6 0
0 0 -5 1
0 0 -4 2
0 0 -3 0
0 0 -2 0
0 0 -1 0
1 1 0 7
1 2 1 0
1 3 2 0
1 4 3 0
1 5 4 0
1 6 5 0
1 7 6 1
1 8 7 0
1 9 8 2

[Nick Cox]

We ask that people use their full real name here.

Please read https://www.statalist.org/forums/help#realnames and https://www.statalist.org/forums/help#adviceextras for why that is the request and what to do about it.

[Muhammad Faisal]

I have requested to change my name. Thank you Nick!

[Phil Bromiley]

On your substantive question, the mean predicted value is positive. It then applies a standard deviation around that mean to give a confidence interval. The main thing we often look for in a confidence interval is whether it spans 0 which is very similar to asking does the random variable (here a predicted value) statistically significant difference from zero. And in this case, it does not (p=.139).

[Allison Anne]

In this Poisson model for margins is "Exponentiated linear prediction, predict()" predicting counts then for the two levels of flag2? Thanks.