I'm working with a dataset of several million observations identified
by, among other things, string variables. I have a list against which
I check these to determine if they belong to a certain category. So
far, so good.
What I would like to do is catch typos, so that "Republic of Frrance"
gets caught by "Republic of France" or whatever. Simon Moore had a
similar request
(http://www.stata.com/statalist/archive/2008-08/msg00467.html); like
him, I occassionally have multiple words per string, but the kind
responses to his post assume (if I read them correctly) that there are
just a few likely substitutions, while I have a couple hundred "red
lion" equivalents and no idea of what the likely typos for them are.
The Giuliano code might work, though, even if I don't understand its
internals. Is Levenshtein distance generally considered the best way
to search for typos? What edit distance is generally considered
appropriate?
Thanks so much in advance.
*
* For searches and help try:
* http://www.stata.com/help.cgi?search
* http://www.stata.com/support/statalist/faq
* http://www.ats.ucla.edu/stat/stata/
