Survival analysis

[Paologo Palogo]

Hi everyone I'm dealing with survial analysis. In my stset command, I introduce pweight variable. If I run the sts list by (group), I obtain negative values in the Fail column in latest time. Is it normal? Can any one give explanation. Thanks a lot

[Clyde Schechter]

I think the information you have provided is too scanty for anyone to respond. Please post back showing the following:

1. The complete, exact -stset- and -sts list- commands you used.
2. The complete, exact output that Stata gave in response to those.
3. Example data that reproduces the problem you are encountering. Use the -dataex- command to do this. If you are running version 17, 16 or a fully updated version 15.1 or 14.2, -dataex- is already part of your official Stata installation. If not, run -ssc install dataex- to get it. Either way, run -help dataex- to read the simple instructions for using it. -dataex- will save you time; it is easier and quicker than typing out tables. It includes complete information about aspects of the data that are often critical to answering your question but cannot be seen from tabular displays or screenshots. It also makes it possible for those who want to help you to create a faithful representation of your example to try out their code, which in turn makes it more likely that their answer will actually work in your data.

Then it may be possible to troubleshoot.

[Paologo Palogo]

Hello Clyde
Thank you much for replying. I'm analysing the effect of age (AGESORTIEBON) of school-leavers on their duration(BDurePEMPF) to access to a first job (PEMPF).
The data example is as followed:

* Example generated by -dataex-. To install: ssc install dataex
clear
input double(newid pond) float(PEMPF BDurePEMPF AGESORTIEBON)
1 2.0968 1 1.00e-07 1
2  1.303 0 1.00e-07 2
2  1.303 0       51 2
3  1.303 0 1.00e-07 1
3  1.303 0        9 1
3  1.303 1       58 1
4 1.2267 0 1.00e-07 1
4 1.2267 0        7 1
4 1.2267 0       14 1
4 1.2267 0       36 1
end
label values AGESORTIEBON AGESORTIEBON
label def AGESORTIEBON 1 "moins de 24ans", modify
label def AGESORTIEBON 2 "25 à 45 ans", modify
label var AGESORTIEBON "RECODE of AGESORTIE (RECODE of AgeSortie)"

My Stset command:
"stset BDurePEMPF [pweight = pond], id(newid) failure(PEMPF==1) scale(1)" and exact results:

id: newid
failure event: PEMPF == 1
obs. time interval: (BDurePEMPF[_n-1], BDurePEMPF]
exit on or before: failure
weight: [pweight=pond]

------------------------------------------------------------------------------
4,047 total observations
1,682 observations begin on or after (first) failure
------------------------------------------------------------------------------
2,365 observations remaining, representing
1,106 subjects
981 failures in single-failure-per-subject data
15,510 total analysis time at risk and under observation
at risk from t = 0
earliest observed entry t = 0
last observed exit t = 73


My sts list by age-group command: " sts list, f at(0 1 to 60) by(AGESORTIEBON)" and result:

failure _d: PEMPF == 1
analysis time _t: BDurePEMPF
id: newid
weight: [pweight=pond]

Beg. Failure
Time Total Fail Function
------------------------------------------------
moins de 24ans
0 0 0 0.0000
1 661.254 514.952 0.4289
2 641.839 21.8534 0.4484
3 618.604 15.8146 0.4625
4 601.123 17.9559 0.4785
5 583.167 12.2085 0.4895
6 567.984 12.1257 0.5004
7 554.176 9.7147 0.5091
8 544.461 14.7907 0.5225
9 528.444 5.8362 0.5277
10 521.29 11.8408 0.5385
11 509.449 15.9759 0.5529
12 493.473 25.6866 0.5762
13 463.18 12.3307 0.5875
14 450.849 5.573 0.5926
15 443.094 8.0867 0.6000
16 433.469 4.9618 0.6046
17 428.507 14.723 0.6182
18 413.784 4.3557 0.6222
19 409.429 10.0398 0.6315
20 397.242 9.7355 0.6405
21 387.506 13.4606 0.6530
22 372.335 6.1353 0.6587
23 364.533 11.2503 0.6692
24 350.009 20.6782 0.6888
25 327.621 10.7401 0.6990
26 316.88 5.98 0.7047
27 310.9 7.2058 0.7115
28 303.695 9.8878 0.7209
29 293.807 14.3111 0.7345
30 279.496 9.1963 0.7432
31 268.633 5.746 0.7487
32 261.584 5.0909 0.7536
33 256.493 8.3884 0.7617
34 243.535 8.7068 0.7702
35 234.829 10.3991 0.7804
36 219.521 9.2048 0.7896
37 208.649 4.9914 0.7946
38 203.658 4.6979 0.7993
39 197.198 8.1255 0.8076
40 187.406 2.6077 0.8103
41 183.235 4.6921 0.8152
42 175.307 4.6973 0.8201
43 169.306 2.7902 0.8231
44 166.516 12.0209 0.8358
45 150.913 4.6432 0.8409
46 146.27 2.9027 0.8440
47 141.993 2.0909 0.8463
48 139.902 9.3428 0.8566
49 124.308 6.811 0.8645
50 113.965 1.7105 0.8665
51 109.197 10.6788 0.8796
52 90.1379 4.0169 0.8849
53 82.7538 6.1561 0.8935
54 70.8769 4.5744 0.9004
55 62.3487 11.5107 0.9188
56 48.154 5.7038 0.9284
57 37.1428 2.3e-13 0.9284
58 37.1428 1.303 0.9309
59 34.158 1.8462 0.9346
60 32.3118 1.3182 0.9373
25 à 45 ans
0 0 0 0.0000
1 275.295 273.462 0.5028
2 265.81 13.7463 0.5285
3 252.064 21.1565 0.5681
4 230.907 8.6782 0.5843
5 222.229 9.6534 0.6024
6 212.576 4.4607 0.6107
7 206.888 7.3348 0.6245
8 199.553 11.2903 0.6458
9 188.263 4.2595 0.6538
10 184.004 3.1498 0.6597
11 180.854 3.3573 0.6660
12 177.496 4.4248 0.6743
13 173.072 7.9504 0.6893
14 164.121 4.7106 0.6982
15 157.999 6.2737 0.7102
16 151.725 7.0038 0.7236
17 144.721 2.9744 0.7293
18 141.747 1.5217 0.7322
19 140.225 2.8399 0.7376
20 136.067 1.2267 0.7400
21 134.84 2.9063 0.7456
22 131.934 3.2234 0.7518
23 128.711 5.9712 0.7633
24 122.74 9.0712 0.7808
25 113.668 5.0773 0.7906
26 108.591 2.8795 0.7961
27 105.712 4.6241 0.8051
28 101.087 1.88 0.8087
29 99.2075 3.0328 0.8145
30 96.1747 1.4375 0.8173
31 96.1747 0 0.8173
32 94.7372 1.5111 0.8202
33 93.2261 3.0729 0.8261
34 90.1532 2.4085 0.8308
35 87.7447 4.984 0.8404
36 81.4577 4.5651 0.8493
37 75.1307 1.5111 0.8524
38 73.6196 8.0246 0.8685
39 65.595 4.4178 0.8773
40 61.1772 1.6667 0.8807
41 57.6643 1.6 0.8840
42 53.2502 3.8284 0.8923
43 49.4218 1.625 0.8959
44 47.7968 2.7484 0.9018
45 47.7968 0 0.9018
46 47.7968 0 0.9018
47 45.0484 0 0.9018
48 43.3484 1.4375 0.9051
49 41.9109 5.0995 0.9166
50 36.8114 5.2818 0.9286
51 31.5296 1.3478 0.9317
52 27.4788 -5.7e-14 0.9317
53 23.0657 1.5111 0.9361
54 23.0657 -5.7e-14 0.9361
55 19.8879 -5.7e-14 0.9361
56 18.4007 2.9623 0.9464
57 15.4384 1.7123 0.9524
58 13.7261 3.6924 0.9652
59 13.7261 5.7e-14 0.9652
60 13.7261 5.7e-14 0.9652
------------------------------------------------
Note: Failure function is calculated over full
data and evaluated at indicated times; it
is not calculated from aggregates shown at
left.

[Clyde Schechter]

Looking at the commands, I see nothing wrong here.

In the output, there are indeed a few lines that have negative numbers in the fail column. But these negative numbers are extremely small, and I'm confident they represent rounding errors, the real number of failures being 0. I think this kind of rounding arises because you have non-integer numbers of failures, as a result of the pweights, and at a certain point this just happens. I doubt there is anything you can do about it. I wouldn't really worry about it. The most important part of the output is the failure function in the last column, and it looks quite normal.

[Paologo Palogo]

Dear Clyde

Thank you very much for this encouraging response. It will allow me to go ahead.

With very best regards

[Paologo Palogo]

Hello Dear Clyde
I hope you're doing well. Let me thank you again for your previous help with my data. I'am back with analysis issues now.
I run cox regression where one variable (SEXE1) interact twice with 2 others. How do I interpret the 1.SEXE1 Hazard Ratio (HR)? I know that if SEXE1 was in interaction with only one variable, I should interpret 1.SEXE1 HR as the effect of sex for the reference category of the other variable? Is it ok if I interpret my case as the effect of sex for the reference categories of the 2 variables (Parent_SAL and FILIERESEC2).
Please have a look in my command:

stcox AGECOX2 COHORT1 Niveau_DiplomeCOX2 Niveau_DiplomeCOX3 STATUECOLCOX1 NIVPERECOX2 NIVPERECOX3 NIVMERECOX2 NIVMERECOX3 TRAVETUDECOX1 TParcours4ACOX1 VIECO_ECOLCOX1 INTENTIONBONCOX2 INTENTIONBONCOX3 VIECOM_PROC2COX2 MobilResidCOX2 MobilResidCOX3 SEXE1##Parent_SAL SEXE1##FILIERESEC2, tvc(RELATIONCOX2 STAGECOX2 BDureCHOM BDureAFSSV) texp(_t>12)

with part of results

(Std. Err. adjusted for 618 clusters in newid)

Robust
_t Haz. Ratio Std. Err. z P>z [95% Conf. Interval]

main
AGECOX2 1.191452 .213162 0.98 0.328 .8390514 1.691861
COHORT1 .9213449 .1183526 -0.64 0.524 .7162753 1.185126
Niveau_DiplomeCOX2 .4079927 .0691601 -5.29 0.000 .2926594 .5687774
Niveau_DiplomeCOX3 .6227901 .1164696 -2.53 0.011 .4316744 .8985185
STATUECOLCOX1 .9675226 .1624553 -0.20 0.844 .696205 1.344575
NIVPERECOX2 .9625857 .1504265 -0.24 0.807 .7086295 1.307554
NIVPERECOX3 .8429818 .1786922 -0.81 0.420 .5563955 1.277182
NIVMERECOX2 1.066104 .1850291 0.37 0.712 .7586966 1.498066
NIVMERECOX3 1.271166 .300705 1.01 0.310 .7995456 2.020976
TRAVETUDECOX1 1.122697 .2308717 0.56 0.574 .7502759 1.679979
TParcours4ACOX1 1.036442 .2020377 0.18 0.854 .7073219 1.518704
VIECO_ECOLCOX1 .5235972 .1701031 -1.99 0.046 .27699 .9897614
INTENTIONBONCOX2 .9715032 .1470767 -0.19 0.849 .7220712 1.307099
INTENTIONBONCOX3 .9342495 .2256326 -0.28 0.778 .5819522 1.499817
VIECOM_PROC2COX2 .9608092 .2371753 -0.16 0.871 .5922682 1.558676
MobilResidCOX2 .8136275 .1362535 -1.23 0.218 .585976 1.129722
MobilResidCOX3 1.055082 .1754078 0.32 0.747 .7616819 1.4615
1.SEXE1 1.360155 .2259295 1.85 0.064 .9821976 1.883553
1.Parent_SAL 1.157758 .2454496 0.69 0.490 .7641164 1.754188

SEXE1#Parent_SAL
1 1 .8220984 .2337737 -0.69 0.491 .470841 1.435401

1.FILIERESEC2 1.523504 .4927045 1.30 0.193 .8082827 2.871601

SEXE1#FILIERESEC2
1 1 .984292 .4231782 -0.04 0.971 .4238044 2.286032

tvc
RELATIONCOX2 .8719985 .2710091 -0.44 0.659 .4742093 1.603472
STAGECOX2 .3038146 .3379903 -1.07 0.284 .0343286 2.688814
BDureCHOM 1.011204 .0043098 2.61 0.009 1.002792 1.019686
BDureAFSSV 1.008822 .0042843 2.07 0.039 1.000459 1.017254

Note: Variables in tvc equation interacted with _t>12.

Thanks a lot

[Clyde Schechter]

Yes, the hazard ratio for the SEXE variable is the effect of SEXE conditional on both Parent_SAL and FILIERESEC2 being zero.

I do not recommend interpreting interactions by hand, except in the simplest cases. Use -margins- instead: it will keep you from making mistakes, not to mention saving you a lot of time.

[Paologo Palogo]

Thank you too much Dear Clyde. This is helpful for me