Why does Stata graph box not ignore missing values _

[Duyen Luu]

hi Stata Gurus,
I am trying to check the outliners in my data. For example, the total asset (ta) of european banks in bank-quarter level

. list ta in 10/20

+----------+
| ta |
|----------|
10. | . |
11. | . |
12. | . |
13. | . |
14. | 1.27e+09 |
|----------|
15. | 1.30e+09 |
16. | 1.30e+09 |
17. | 1.31e+09 |
18. | 1.36e+09 |
19. | 1.39e+09 |
|----------|
20. | 1.41e+09 |
+----------+

. sum ta

Variable | Obs Mean Std. dev. Min Max
-------------+---------------------------------------------------------
ta | 2,748 1.62e+08 3.91e+08 152426 2.67e+09

. codebook ta

-------------------------------------------------------------------------------------------------------
ta (unlabeled)
-------------------------------------------------------------------------------------------------------

Type: Numeric (float)

Range: [152426.03,2.673e+09] Units: .01
Unique values: 2,744 Missing .: 6,300/9,048

Mean: 1.6e+08
Std. dev.: 3.9e+08

Percentiles: 10% 25% 50% 75% 90%
1.7e+06 5.4e+06 2.0e+07 7.6e+07 4.5e+08

I tried to detect the outliners by graphing boxplot and all of the observations which are NOT missing values are graphed as Outliners.
. graph box ta - seems to assign numeric value to missing observations and include them when graphing
https://goetheuniversitaet-my.sharepoint.com/:i:/g/personal/d8_6qmtk8o_goetheuniversitaet_onmicrosoft_com/EaGD2KcyY9NJtobXIPjZ0sUBmN54uKuBxdhK8CYA-06ssQ?e=unJuZe

I tried to tell Stata to graph only non-missing value but nothing really works
. graph box ta if ta >= 152426 *Graph only the observation with value starting from the min(ta)
. graph box ta if ! missing (ta)

All suggestion are appreciated,

Best

[Ken Chui]

Welcome to Statalist!

I think there is some confusion in the terms here. Those values with a number in the box plot are "outside values" (some may call them "outliers") and they are NOT missing. A numeric missing values in Stata looks like a period (.). In your data, cases 10 to 13 are missing. Missing values are not plotted. If you look at the summary table, the maximum is 2.67e+9. And if you look at the box plot, the highest data point is also labeled as 2.67e+9. So, all those points with values are real data.

If you do not want to see those extreme points, you can add -nooutsides- as an option after the graph box command like:

graph box ta, nooutsides

But notice that this does not change the data. Only the graph. Those extreme values are still there.

[Nick Cox]

When a variable ranges from 152426 to 2.67 billion you'll find that working on logarithmic scale is a good idea. A box plot is fairly useless for such a variable. unless you do that.