This has been overtaken somewhat by later postings, but a few comments:
Somewhat like Jeph, I think I understood Leny's previous posting. Unlike
Jeph, I think my solution does what Leny asked for. Whether that's best
is another matter.
Jeph's solution appeals as making most use of the data. In practice
growth rates calculated from close measurements will be noisy, and I'm
not sure a mean of growth rates is going to be the best bet. In
principle a harmonic mean is preferable, although that will choke on any
growth rates measured as zero. Or you average just by measuring growth
over longer spans, as Leny initially signalled.
You might find the two solutions complementary, rather than
contradictory.
Nick
[email protected]
-----Original Message-----
From: [email protected]
[mailto:[email protected]] On Behalf Of Jeph Herrin
Sent: 24 November 2008 22:02
To: [email protected]
Subject: Re: st: RE: coding problem: looping through a list of ID's
I think my solution does what you want, but better in that
the best approximation to the increase in height per 365 days
is the average increase/day between visits, multiplied by
365.
Stata stores the dates of visit as consecutive integers counting
number of days. Thus, the difference in height between visits
divided by the difference in dates will be the change in height
per day between those visits. Do this for each visit, and then
average over visits. It only needs two lines of code, per my
solution, though to get in years you will want to multiply by
the number of days in a year.
bysort ID (dov) : gen rate=(height-height[_n-1])/(dov-dov[_n-1])
bysort ID : egen average=mean(rate*365.25)
If you truly want to ignore the information in the visits that
are less than a year apart, and only look at the growth between
visits that are "closest" to one year from the first visit, try
this:
* put date in years
gen year=dov/365.25
* get gap to nearest whole years from first visit
bysort ID (dov) : gen gap = abs(round(year-year[1])-(year-year[1]))
* use the visit if closer than neighbor visits to one year
bysort ID (dov) : gen use = gap<gap[_n-1]&gap<=gap[_n+1]
where I've made the last inequality soft to deal with potential ties.
Now just keep the ones you're using, and apply solution above, only
now the denominator is assumed to be 1, though of course that's just
a rough approximation:
keep if use
bysort ID (dov) : gen rate = height-height[1]
bysort ID : egen average=mean(rate)
where the second line gives you the average for the subject over
all of their years.
hope this helps,
Jeph
Leny Mathew wrote:
> Thanks Nick & Jeff for your suggestions. I think that I was perhaps
> not clear enough with what I was trying to do. The following is the
> data for one of the patients:
>
> pt_id dov pt_ht age
> 2 24-May-01 141 12.92519
> 2 31-May-01 141 12.94441
> 2 11-Sep-01 141 13.22718
> 2 11-Dec-01 145 13.47701
> 2 21-Feb-02 146 13.67467
> 2 2-May-02 147 13.86685
> 2 11-Jul-02 149 14.05903
> 2 21-Nov-02 152 14.42416
> 2 21-Jan-03 152.3 14.59163
> 2 10-Apr-03 153.7 14.80851
> 2 1-May-03 153.4 14.86616
> 2 1-Jul-03 153.8 15.03363
> 2 9-Sep-03 154.8 15.22581
> 2 18-Nov-03 154.8 15.41798
> 2 20-Jan-04 156 15.59094
> 2 18-Mar-04 157 15.75017
> 2 20-May-04 156 15.92313
> 2 15-Jul-04 157 16.07687
> 2 16-Sep-04 157 16.24983
> 2 11-Nov-04 158 16.40357
>
> I'm trying to find out how much this patient grew in a year. So the
> way I thought of it is to find out the interval of time that
> approximates an year and then compute the difference in height between
> those two points. So, taking the difference between _n and _n-1 will
> not suffice but it has to be _n and _n-i when i goes from 1 to _N
> within patient.
> For example the first time point is 24 May 01 and the time that
> approximates a year is 2 may 02. So, the height increase would be
> (147-141).
> Then the next one would be from 2 May 02 to 1 May 03 and so on.
> The patient number 3 has 105 visits from 1988 to 2000, so the number
> of intervals would be more compared to ID 2. I would most probably end
> up using the most recent one year height increase for analysis
> purposes, but that doesn't make this any easier.
>
> I hope this clarifies things better.
*
* 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/
