stcox and missing standard errors due to what type of collinearity?

[Jim Nebus]

I am not getting standard errors for many antecedents when using stcox. Results at bottom for the simplest run of a single antecedent.. I have read "Why does stcox sometimes produce missing standard errors" FAQ by Mario Cleves. However, my results do not seem to match any of his four cases. In particular, my hazard rate of 16.8 is not approaching infinity or zero (case 1), nor is it close to one (cases 2, 3, or 4). My question is what may be the cause of this collinearity?

log using "C:\Offline\MeasuringPower\StataAnalysis\Ctest1.sm cl", replace
---------------------------------------------------------------------------------------------------
name: <unnamed>
log: C:\Offline\MeasuringPower\StataAnalysis\Ctest1.smc l
log type: smcl
opened on: 23 May 2015, 08:56:37

. use "C:\Users\jnebus\Desktop\Stata Runs May 2015\CISPA May 4.dta"

. stset time, id(act_id) failure(Evnt) exit(time .)

id: act_id
failure event: Evnt != 0 & Evnt < .
obs. time interval: (time[_n-1], time]
exit on or before: time .

------------------------------------------------------------------------------
1252 total observations
217 ignored because act_id missing
45 observations end on or before enter()
------------------------------------------------------------------------------
990 observations remaining, representing
45 subjects
51 failures in multiple-failure-per-subject data
990 total analysis time at risk and under observation
at risk from t = 0
earliest observed entry t = 0
last observed exit t = 22

. stcox TwoDaysPriorTotalProxInten

failure _d: Evnt
analysis time _t: time
exit on or before: time .
id: act_id

Iteration 0: log likelihood = -194.13979
Iteration 1: log likelihood = -192.21861
Iteration 2: log likelihood = -191.32051
Iteration 3: log likelihood = -191.08012
Iteration 4: log likelihood = -190.72658
Iteration 5: log likelihood = -190.14112
Iteration 6: log likelihood = -190.44729 (backed up)
Refining estimates:
Iteration 0: log likelihood = -191.04468
Iteration 1: log likelihood = -190.69538
Iteration 2: log likelihood = -190.60049

Cox regression -- Breslow method for ties

No. of subjects = 45 Number of obs = 990
No. of failures = 51
Time at risk = 990
LR chi2(0) = 7.08
Log likelihood = -190.60049 Prob > chi2 = .

--------------------------------------------------------------------------------------------
_t | Haz. Ratio Std. Err. z P>|z| [95% Conf. Interval]
---------------------------+----------------------------------------------------------------
TwoDaysPriorTotalProxInten | 16.88659 . . . . .
--------------------------------------------------------------------------------------------

.
. clear

.
. log close
name: <unnamed>
log: C:\Offline\MeasuringPower\StataAnalysis\Ctest1.smc l
log type: smcl
closed on: 23 May 2015, 08:56:42
---------------------------------------------------------------------------------------------------

[Steve Samuels]

Welcome to Statalist, Jim!

I suggest that you create the risk sets, then get the summary statistics for each failure and for the non-failures. Perhaps something will pop out.

stsplit, at(failures) riskset(rset)
bys rset _d: sum TwoDaysPriorTotalProxInten

Please add the link to the Cleves reference and read FAQ 12 to learn how to format results and code with the Forum editor.

[Steve Samuels]

There can be other causes of missing standard errors besides those that Mario listed. Your hazard ratio of 16.9, although not infinite, is larger than I've ever seen, suggesting an anomaly. The LR test also has zero degrees of freedom. Try all the methods besides the default of computing standard errors using the vce() option.