Using Stata to identify data errors

[Welch Suggs]

I have a large and messy dataset of the form

* Example generated by -dataex-. To install: ssc install dataex
clear
input str1 Product byte(Year FirmA FirmB FirmC)
"P"  1  .  0 19
"P"  2  . 26 12
"P"  3  8 14 12
"P"  4  . 12  0
"P"  5  0 12  .
"P"  6 12 33  .
"P"  7 18 22  .
"P"  8 12 13  .
"P"  9 16 17  .
"P" 10 12 13  .
"P" 11  9 18  .
"P" 12 10 15  .
"P" 13 11 11  .
"P" 14 12 27  .
"P" 15 13 40  .
"P" 16 11 31  .
"P" 17 15 23  .
"P" 18 17 30  .
"P" 19 18 20  .
"P" 20  8 21  .
"P" 21 13 23  .
"P" 22 17 25  .
"P" 23 21 23  .
"P" 24 27  .  .
end

...in which there there are many more products and firms. I know from other information that Firm C discontinued production in year 4, so the subsequent null values are legitimate. But I also know that the nulls and zeroes for Firms A and B are data errors. Are there any strategies to identify the anomalous values such as by looking for nulls or zeroes that are adjacent to nonzero values? Perhaps by excluding data runs?

TIA!

[Andrew Musau]

See this FAQ by Nick Cox and Vince Wiggins on how to identify runs in a time series.

http://www.stata.com/support/faqs/da...-observations/


Or alternatively, install tsspell from SSC and read the help files

ssc install tsspell
help tsspell

You then just have to check conditions such as

list if runA< 2 & FirmA==.
list if runA< 2 & FirmA==0

etc.

[Welch Suggs]

Thank you!