Error 504 when using Epctile with svy option

[Claire Connacher]

I'm using data from IPUMS USA, and I'm trying to find the median of a numeric variable for income with the person weight applied. So, I found the epctile command. I am using four different data sets with different subsets of the larger sample, and am using the IPUMS USA person weight (perwt). When I use the epctile command, it has worked fine with three of the datasets, but with one, it has given back the error "estimates post: matrix has missing values" when using the epctile command. There are no missing values though. I also tried the same syntax with other numeric variables in this particular dataset and it gives back the same error. This is the syntax I used and Stata's response:

 svyset cluster [pweight=perwt], strata(strata)

Sampling weights: perwt
VCE: linearized
Single unit: missing
Strata 1: strata
Sampling unit 1: cluster
FPC 1: <zero>

 epctile inctot, p(50) svy
estimates post: matrix has missing values
r(504);

I also included an example of the data for the variable I am using below. Any help with this issue would be appreciated--Thank you!
-Claire


Example of some data from the inctot variable using dataex:

* Example generated by -dataex-. For more info, type help dataex
clear
input long inctot
15100
12000
14000
36000
870
14000
30000
14000
870
40000
12000
7200
10000
870
12000
14000
12000
7200
14000
12000
7000
6400
7200
40000
18000
18000
98000
65000
3200
48000
19300
27000
20600
30000
55000
50000
70000
7200
70000
62000
0
27000
10000
58000
68000
29000
55000
9900
25000
110000
52000
43000
33000
43800
17000
22000
24000
7300
35000
29000
5400
30000
21000
40000
4600
20800
45000
4600
19600
80000
43500
79300
32000
40000
35700
10000
60000
40000
15000
23000
58000
40000
14180
30700
25500
25000
32000
40800
28000
25000
22000
40700
26500
29000
118000
9000
95600
50000
33800
9300
end

[Andrew Musau]

epctile is from http://staskolenikov.net/stata (FAQ Advice #12).

epctile from http://staskolenikov.net/stata
epctile -- estimation and inference for percentiles / / Author: Stas
Kolenikov, skolenik at gmaildotcom / epctile is an e-class version of
pctiles that produces / standard errors in addition to point estimates, /
including standard errors for complex survey data.

From the command's description, the command estimates point estimates and standard errors and for one reason or another, it is not able to do so with your final dataset. One reason in survey data as addressed in the following linked Stata FAQ is the existence of stratum with single sampling unit: https://www.stata.com/support/faqs/s...se-of-stratum/. So check out what is peculiar about your data. But essentially, you will get the error if you try to store the coefficient vector (b) and variance-covariance matrix (V) into Stata's system, and either or both of these contain missing values.

mat b= J(1, 3, .)
mat V= J(3, 3, .)
mat l b
mat l V
ereturn post b V

Res.:

. mat b= J(1, 3, .)

.
. mat V= J(3, 3, .)

.
. mat l b

b[1,3]
    c1  c2  c3
r1   .   .   .

.
. mat l V

symmetric V[3,3]
    c1  c2  c3
r1   .
r2   .   .
r3   .   .   .

.
. ereturn post b V
estimates post: matrix has missing values
r(504);

.

[Claire Connacher]

Thank you! I followed the article about strata with a single sampling units and found there were two strata with single sampling units in my data set, so I followed the solution using the option

singleunit(certainty)

to the command svyset, and that seemed to do the trick. Thanks!
-Claire