Quantcast
Channel: Statalist
Viewing all articles
Browse latest Browse all 65116

Bootstrapping standard errors with eteffects - one specific replication has missing values, does not converge

$
0
0
Hello Statalist,

I have recently upgraded to Stata 14.2 and am using the new eteffects command. I have been bootstrapping standard errors with eteffects successfully to this point; however, this specific model repeatedly will run for about 250 iterations and then never move forward for an hour or more, although Stata's "working" indicator (the spinning wheel at the bottom right) runs.

Code:
eteffects (uscnoer AGE3X i.female i.race i.ratehealth K6SUM4 i.degree i.marriedyes i.household i.poverty i.region i.urban i.employment i.recession, probit) (switch i.ivfour i.ivfive), vce(bootstrap, reps(1000) seed(51113))
I checked it step by step using the noisily command and found that the problem iteration returns the following notes:

Code:
note: 6790 missing values returned for equation 5 at initial values
note: 6790 missing values returned for equation 6 at initial values
note: 6790 missing values returned for equation 7 at initial values
note: 6790 missing values returned for equation 8 at initial values
note: 6790 missing values returned for equation 9 at initial values

#Iteration 0:   EE criterion =  3.2510106  (not concave)
Iteration 1:   EE criterion =  3.0020691  (not concave)
Iteration 2:   EE criterion =  2.8847664  (not concave)
Iteration 3:   EE criterion =  2.7962922  (not concave)
Iteration 4:   EE criterion =  2.7957563  (not concave)
Iteration 5:   EE criterion =   2.795542  (not concave)
Iteration 6:   EE criterion =  2.7953705  (not concave)
Iteration 7:   EE criterion =  2.7953534  (not concave)
Iteration 8:   EE criterion =  2.7953517  (not concave)
Iteration 9:   EE criterion =  2.7953516  (not concave)
Iteration 10:  EE criterion =  2.7953516  (not concave)
If left to its own devices, it will keep repeating the last value (2.7953516) for more than 100 iterations. The preceding runs of this model converge in one or two iterations at most, and I strongly suspect that the problem is the 6790 missing values alluded to in the notes - they never show up until the problem run. However, I'm not sure why these values are missing in this one case, or indeed what it refers to when it says that they are missing in equations 5, 6, 7, etc. The sample in the other runs of the model is 8,466 individuals, and there does not seem to be any variable that is missing for anywhere near 6,790 respondents.

Here is a sample of my dataset:

Code:
* Example generated by -dataex-. To install: ssc install dataex
clear
input float(uscnoer switch) byte AGE3X float(female race ratehealth) byte K6SUM4 float(degree marriedyes household poverty region urban employment recession switch ivfour ivfive)
0 0 55 0 0 3  0 0 0 2 5 2 1 2 2 0 0 0
1 1 58 1 1 2 -1 2 0 2 4 1 1 0 2 1 0 0
1 0 47 0 0 3  8 1 0 1 5 2 0 2 2 0 0 0
1 0 20 0 2 3 17 1 0 2 3 3 1 0 2 0 0 0
1 0 44 1 1 2  2 1 0 2 3 4 1 2 2 0 0 0
1 0 45 1 0 5 20 1 0 2 1 1 1 2 2 0 0 0
1 1 60 0 0 3  1 1 1 2 4 4 1 2 2 1 0 0
1 0 47 1 2 4 18 0 0 2 1 1 1 0 2 0 0 0
1 0 58 0 2 3  2 0 1 2 4 4 1 0 2 0 0 0
0 1 52 0 0 4  0 1 1 2 4 2 1 2 2 1 0 0
end
label values AGE3X H1560135X
label values K6SUM4 H1561688X
label def H1561688X -1 "-1 INAPPLICABLE", modify
When I set a different seed, the endlessly-iterating problem still occurs, but after fewer replications.

I would greatly appreciate any feedback the list can provide on strategies for diagnosing, understanding, and/or fixing this issue.

Sincerely,

Liz Wood

Viewing all articles
Browse latest Browse all 65116

Trending Articles



<script src="https://jsc.adskeeper.com/r/s/rssing.com.1596347.js" async> </script>