Narrow scale of axis in sts graph (survival data)

[Sebastian Geiger]

Hello everyone,

I have another question for which I could not find an answer by myself: I'm trying to create a graph for a survival date analysis (-sts graph-). The command works fine but since there are some outliers with a very high value for the duration variable the graph extents very far to the right. I tried the option -xscale()-, but despite the option is accepted by Stata it does not change the look of the graph. In line graphs, I solve this problem by using the -if- option. In this case, however, the -if- option would also affect the calculation of the Kaplan-Meier survival function, which I would like to avoid. Is there any way to accomplish my goal?

Here is my full command für the graph:

[CODE
]#delimit ;
sts graph, ci
title("")
xtitle("Elapsed time (days)", size(small)) ytitle("Probability of reaching the respective duration", size(small))
xlabel(0(200)1000, labsize(small))
xscale(range(0 1000))
ylabel(, labsize(small) angle(horizontal))
legend(off)
;
#delimit cr
[/CODE]

Thanks in advance!

Best regards,
Sebastian

[Steve Samuels]

It would help if 1) you'd show the graph (graph export as png file); Upload attachment (button to left of "A" on right); and 2) tell us the values of the outlying points, which we cannot see.

[Sebastian Geiger]

I'm using test data to write my do file. Thus, the actual values will definitely differ to some extend, but here is the output of the -sum- command for the dependent variable.

. sum unempl_epi if nrU==1  & jahr>=2010, det

     Duration of unemployment episode, with combining (value in
                all spells, use unempl_first)
-------------------------------------------------------------
      Percentiles      Smallest
 1%            2              1
 5%            8              1
10%           17              1       Obs               1,932
25%         52.5              1       Sum of Wgt.       1,932

50%        145.5                      Mean           494.0316
                        Largest       Std. Dev.      1252.806
75%        359.5          11317
90%          992          11757       Variance        1569522
95%         2023          12673       Skewness       5.532403
99%         7460          12824       Kurtosis       39.13846

I hope the percentiles give you a better indication about what I mean with outliers.

Here is one of the graphs I'm trying to create:


It would be perfectly fine if the x axis ranges just up to, let's say, 1000.

[Steve Samuels]

You can restrict the graph with a new stset:

​stset studytime, fail(fail) exit(time 6000)
sts graph, ci

[Sebastian Geiger]

Thank you Steve for your answer. Doesn't this also affect how the graph looks like, because the unemployment spells are treated as if they end after 6,000 days even if they are longer?!

[Steve Samuels]

It's just like cutting off the original graph you show above at 6,000 days. The graph looks different, but the values shown are the same as those on the original curve. Your statement that "unemployment spells are treated as if they end after 6,000 days even if they are longer" is contradictory. The longer spells do end after 6,000 days, but when is not shown. An alternative which will show the entire range is

xscale(log)

A disadvantage is that will squish the left hand size of the curve.

[Sebastian Geiger]

Thank you Steve for your help. I checked the graphs again and you are right.