Re: st: FW: Max/Min and Growth rates among subsample(by year) of panel data

[Nick Cox <njcoxstata@gmail.com>]

Here is a trick which I like.

bysort fips_num: egen peakRGDP = max(RGDP) if year >2001 & year <2010
bysort fips_num (peakRGDP): replace peakRGDP = peakRGDP[1]

can be put in one line

bysort fips_num: egen peakRGDP = max(RGDP / (year >2001 & year <2010))

Here's how it works:

(year >2001 & year <2010) is true or false, evaluated as 1 when true
and 0 when false.

RGDP / (year >2001 & year <2010)  is thus RGDP/1 = RGDP or RGDP/0 = missing.

Normally, dividing by 0 is something you just know not to do, but here
it is totally deliberate: you know the result will be missing, but as
usual Stata will just ignore the missings, which is exactly what you
want.

max(RGDP / (year >2001 & year <2010))

therefore acts as if it were a conditional. You get a maximum, but conditionally

It's purely psychology, and accidental similarity of notation, but you
will know | as indicating a conditional, and here it's just a little
tired and leaning over slightly and is shown as /.

There are other ways to do it. Here's one:

bysort fips_num: egen peakRGDP = max(cond(year >2001 & year <2010, RGDP, .))

The -max()- function of -egen- takes an expression, which gives extra
scope here.

(You could also use -twithin()-, by the way.)

As for when the peak or trough occurs, that's the same trick

bysort fips_num: egen whenpeak = min(year / (RGDP == peakRGDP))

Note that this code copes with ties too. If two or more years tie for
peak, it takes the first. If you want another definition, go for
-max()- or whatever else makes sense.

I am bailing out here, but I think some of this carries over to the
rest of your problem.

Footnotes: I see no value in a -reshape- here.

The divide by zero trick is written up in
http://www.stata-journal.com/article.html?article=dm0055

Nick
njcoxstata@gmail.com


On 16 August 2013 17:31, Nickolas Lyell <nlyell@naco.org> wrote:

> I've been trying to analyze a large database of counties and their economic indicators from 1990-2013.
>
> I have it formatted long by fips number (a unique identifier for each US county) and year, with the indicators wide.
>
> I am trying to calculate several indicators within a subsample of years by county.  For instance, I would like to know the peak pre-recession level of real GDP (RGDP).  Since I don't want the 90's or post-recession data to show up, I would like to limit the maximum to 2002-2009.
>
> To get this value and its year as variables in the dataset, I did the following:
>
> sort fips_num year
> by fips_num: egen peakRGDP = max(RGDP) if year >2001 & year <2010 bysort fips_num (peakRGDP): replace peakRGDP = peakRGDP[1] by fips_num: gen peakRGDPyear = year if peakRGDP==RGDP bysort fips_num (peakRGDPyear): replace peakRGDPyear = peakRGDPyear[1]
>
> and the same thing for the trough value:
>
> sort fips_num year
> by fips_num: egen troughRGDP = min(RGDP) if year >peakRGDPyear bysort fips_num (troughRGDP): replace troughRGDP = troughRGDP[1] by fips_num: gen troughRGDPyear = year if troughRGDP==RGDP bysort fips_num (troughRGDPyear): replace troughRGDPyear = troughRGDPyear[1]
>
> which works, but feels clunky.  And neither does it help me to calculate annualized growth rates among subsamples of years.  I would like to find out, for instance the annualized growth rate of RGDP from 1990-peakRGDPyear, peakRGDPyear-troughRGDPyear, troughRGDPyear-2013.
>
> Is there a better way to do this?  Maybe I need to convert my data back to wide with respect to year and use rowmax and rowmin functions, but that also seems clunky.

*
*   For searches and help try:
*   http://www.stata.com/help.cgi?search
*   http://www.stata.com/support/faqs/resources/statalist-faq/
*   http://www.ats.ucla.edu/stat/stata/