Many thanks to David, Nikos, Scott and Nick who all responded to my post
about a colleague's request to find the maximum drop in a variable
during a given time period.
Reading the answers was very instructive for me, especially as I tried
and failed to solve the query before I posted.
Unfortunately, none worked "out of the box", but were close enough that
my colleague had now been able to work out her own solution.
In summary:
- Nikos appeared to work fine, but failed when there were two
consecutive values that were equal. This we fixed by replacing
by day:gen tri=-1*(change<0)+(change>0)
with
by day:gen tri=-1*(change<=0)+(change>0)
- Nick's was good on brevity, and neat enough not to take much figuring
out. But it failed when the drop began with the first observation in a
day. This my colleague says can be fixed by replacing
egen max = max(pressure) if _spell, by(panel _spell)
with
gen max = pressure if _seq ==1
but I can't see why it wouldn't fail now if the maximum pressure wasn't
the first observation of the day...
- David's answer we liked lots, particularly for all the helpful
comments! It appears to produce the right answer every time, but
rounded to a whole number (although in my sample data in the email,
pressure was all integer, in the real data it is float). This we fixed
by dropping the "int" from
by year month day runno: gen int extreme = pressure[_N]
- Scott's I'm afraid we got a bit lost on. It doesn't always work right,
but we can't figure out why it fails. (Sorry).
Thanks again to all of you for taking the time to help on this. And
apologies for listing the failures above like they were problems in your
code. In fact, all the code worked fine for the simplified data I sent
in my original email, and the task I set was a little unfair - expecting
you to write code in anticipation of idiosyncrasies in a real dataset I
didn't show you!
Chris.
*
* For searches and help try:
* http://www.stata.com/support/faqs/res/findit.html
* http://www.stata.com/support/statalist/faq
* http://www.ats.ucla.edu/stat/stata/
