Dear all,
I've estimated a survival model using stcox with stratification, using the following code:
stcox log(litigiousness) log(size) log(diversification) c.log(litigiousness)#c.log(portfolio85) i.year ib1.dcountry, strata(sector) vce(cluster id) nohr
I am interested in displaying the margins of the interaction between litigiousness and portfolio85, both continous variables. I used this command:
margins, at(log(litigiousness)=(-4.736198(0.9)0) log(portfolio85)=(0 4.758966))
This is the result:
Predictive margins Number of obs = 4,449,257
Model VCE : Robust
Expression : Predicted hazard ratio, predict()
1._at :log(litigiousness) = -4.736198
log_port~o85 = 0
2._at : log(litigiousness) = -4.736198
log_port~o85 = 4.758966
3._at : log(litigiousness) = -3.836198
log_port~o85 = 0
4._at :log(litigiousness) = -3.836198
log_port~o85 = 4.758966
5._at : log(litigiousness) = -2.936198
log_port~o85 = 0
6._at : log(litigiousness) = -2.936198
log_port~o85 = 4.758966
7._at : log(litigiousness) = -2.036198
log_port~o85 = 0
8._at : log(litigiousness) = -2.036198
log_port~o85 = 4.758966
9._at : log(litigiousness) = -1.136198
log_port~o85 = 0
10._at : log(litigiousness) = -1.136198
log_port~o85 = 4.758966
11._at : log(litigiousness) = -.236198
log_port~o85 = 0
12._at : log(litigiousness) = -.236198
log_port~o85 = 4.758966
------------------------------------------------------------------------------
| Delta-method
| Margin Std. Err. z P>|z| [95% Conf. Interval]
-------------+----------------------------------------------------------------
_at |
1 | .0204764 .0091551 2.24 0.025 .0025327 .0384201
2 | .006142 .00274 2.24 0.025 .0007717 .0115123
3 | .0184538 .0082302 2.24 0.025 .0023229 .0345847
4 | .0071035 .003165 2.24 0.025 .0009002 .0133067
5 | .016631 .0074034 2.25 0.025 .0021206 .0311414
6 | .0082154 .0036587 2.25 0.025 .0010445 .0153863
7 | .0149883 .0066639 2.25 0.025 .0019273 .0280492
8 | .0095014 .0042327 2.24 0.025 .0012054 .0177974
9 | .0135078 .0060021 2.25 0.024 .0017439 .0252717
10 | .0109887 .0049006 2.24 0.025 .0013836 .0205938
11 | .0121735 .0054095 2.25 0.024 .0015711 .022776
12 | .0127088 .0056783 2.24 0.025 .0015794 .0238381
------------------------------------------------------------------------------
Using marginsplot I get:
I am concerned with respect to two things:
1) why do I get the same p-values in the results' table of the margins command?
When I use webuse stan3 from the manual and I try to reproduce a similar exercise (with stratification), I also got equal p-values in the margins' table.
2) how do you interpret the large confidence intervals (CIs)? How can I explain the fact that the CIs for the two levels of log_portfolio85 overlap along different values of log(litigiousness)?
Thank you very much in advance,
Julia
I've estimated a survival model using stcox with stratification, using the following code:
stcox log(litigiousness) log(size) log(diversification) c.log(litigiousness)#c.log(portfolio85) i.year ib1.dcountry, strata(sector) vce(cluster id) nohr
I am interested in displaying the margins of the interaction between litigiousness and portfolio85, both continous variables. I used this command:
margins, at(log(litigiousness)=(-4.736198(0.9)0) log(portfolio85)=(0 4.758966))
This is the result:
Predictive margins Number of obs = 4,449,257
Model VCE : Robust
Expression : Predicted hazard ratio, predict()
1._at :log(litigiousness) = -4.736198
log_port~o85 = 0
2._at : log(litigiousness) = -4.736198
log_port~o85 = 4.758966
3._at : log(litigiousness) = -3.836198
log_port~o85 = 0
4._at :log(litigiousness) = -3.836198
log_port~o85 = 4.758966
5._at : log(litigiousness) = -2.936198
log_port~o85 = 0
6._at : log(litigiousness) = -2.936198
log_port~o85 = 4.758966
7._at : log(litigiousness) = -2.036198
log_port~o85 = 0
8._at : log(litigiousness) = -2.036198
log_port~o85 = 4.758966
9._at : log(litigiousness) = -1.136198
log_port~o85 = 0
10._at : log(litigiousness) = -1.136198
log_port~o85 = 4.758966
11._at : log(litigiousness) = -.236198
log_port~o85 = 0
12._at : log(litigiousness) = -.236198
log_port~o85 = 4.758966
------------------------------------------------------------------------------
| Delta-method
| Margin Std. Err. z P>|z| [95% Conf. Interval]
-------------+----------------------------------------------------------------
_at |
1 | .0204764 .0091551 2.24 0.025 .0025327 .0384201
2 | .006142 .00274 2.24 0.025 .0007717 .0115123
3 | .0184538 .0082302 2.24 0.025 .0023229 .0345847
4 | .0071035 .003165 2.24 0.025 .0009002 .0133067
5 | .016631 .0074034 2.25 0.025 .0021206 .0311414
6 | .0082154 .0036587 2.25 0.025 .0010445 .0153863
7 | .0149883 .0066639 2.25 0.025 .0019273 .0280492
8 | .0095014 .0042327 2.24 0.025 .0012054 .0177974
9 | .0135078 .0060021 2.25 0.024 .0017439 .0252717
10 | .0109887 .0049006 2.24 0.025 .0013836 .0205938
11 | .0121735 .0054095 2.25 0.024 .0015711 .022776
12 | .0127088 .0056783 2.24 0.025 .0015794 .0238381
------------------------------------------------------------------------------
Using marginsplot I get:
I am concerned with respect to two things:
1) why do I get the same p-values in the results' table of the margins command?
When I use webuse stan3 from the manual and I try to reproduce a similar exercise (with stratification), I also got equal p-values in the margins' table.
2) how do you interpret the large confidence intervals (CIs)? How can I explain the fact that the CIs for the two levels of log_portfolio85 overlap along different values of log(litigiousness)?
Thank you very much in advance,
Julia
