Notice: On April 23, 2014, Statalist moved from an email list to a forum, based at statalist.org.
[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]
RE: st: finding means and percentiles with mim
From
"Lachenbruch, Peter" <[email protected]>
To
"'[email protected]'" <[email protected]>
Subject
RE: st: finding means and percentiles with mim
Date
Fri, 26 Mar 2010 09:06:16 -0700
Here is a short experiment using centile, sqreg with default (20) reps and sqreg with 1000 reps.
The intervals are comparable, but the ones with 1000 bootstrap replications are sometimes shorter and sometimes longer than the ones with 20. A similar comparison holds for the centile command. So my previous post can't be so definitive.
. centile ck252 if _mj==0,c(10 25 50 75 90)
-- Binom. Interp. --
Variable | Obs Percentile Centile [95% Conf. Interval]
-------------+-------------------------------------------------------------
ck252 | 414 10 151.5 115.9086 185.2382
| 25 325 281.8514 394
| 50 1031.5 778.7953 1288.068
| 75 5168.75 3481.268 6165.873
| 90 14443 10497.57 19204.41
. sqreg ck252 if _mj==0,quantile(10 25 50 75 90)
(fitting base model)
(bootstrapping ....................)
| Bootstrap
ck252 | Coef. Std. Err. t P>|t| [95% Conf. Interval]
---------+----------------------------------------------------------------
q10 |
_cons | 152 16.20559 9.38 0.000 120.1443 183.8557
---------+----------------------------------------------------------------
q25 |
_cons | 325 29.48077 11.02 0.000 267.0489 382.9511
---------+----------------------------------------------------------------
q50 |
_cons | 1032 130.1559 7.93 0.000 776.1494 1287.851
---------+----------------------------------------------------------------
q75 |
_cons | 5142 658.7705 7.81 0.000 3847.039 6436.961
---------+----------------------------------------------------------------
q90 |
_cons | 14404 2132.311 6.76 0.000 10212.46 18595.54
--------------------------------------------------------------------------
. sqreg ck252 if _mj==0,quantile(10 25 50 75 90) reps(1000)
(fitting base model)
(bootstrapping [I deleted the dots]
----------------------------------------------------------------------------
| Bootstrap
ck252 | Coef. Std. Err. t P>|t| [95% Conf. Interval]
---------+----------------------------------------------------------------
q10 |
_cons | 152 18.3516 8.28 0.000 115.9258 188.0742
---------+----------------------------------------------------------------
q25 |
_cons | 325 28.69129 11.33 0.000 268.6008 381.3992
---------+----------------------------------------------------------------
q50 |
_cons | 1032 137.8826 7.48 0.000 760.9607 1303.039
---------+----------------------------------------------------------------
q75 |
__cons | 5142 749.7526 6.86 0.000 3668.193 6615.807
---------+----------------------------------------------------------------
q90 |
_cons | 14404 1837.045 7.84 0.000 10792.88 18015.12
--------------------------------------------------------------------------
Tony
Peter A. Lachenbruch
Department of Public Health
Oregon State University
Corvallis, OR 97330
Phone: 541-737-3832
FAX: 541-737-4001
-----Original Message-----
From: [email protected] [mailto:[email protected]] On Behalf Of Martin Weiss
Sent: Friday, March 26, 2010 8:48 AM
To: [email protected]
Subject: AW: st: finding means and percentiles with mim
<>
" In looking at the confidence intervals, the ones produced by sqreg/qreg
are slightly shorter than the ones produced by centile."
How does the fact that we are comparing analytic (-centile-) and
-bootstrap-ped (-sqreg-) standard errors play into your considerations?
HTH
Martin
-----Ursprüngliche Nachricht-----
Von: [email protected]
[mailto:[email protected]] Im Auftrag von Lachenbruch,
Peter
Gesendet: Freitag, 26. März 2010 16:41
An: '[email protected]'
Betreff: RE: st: finding means and percentiles with mim
Martin and I have had an interchange off-line on this. I'd like to
summarize my interpretation.
The centile command uses a single observation to estimate the percentile.
The sqreg or qreg command uses a weighted combination of the observations.
So I would expect them to differ. In looking at the confidence intervals,
the ones produced by sqreg/qreg are slightly shorter than the ones produced
by centile.
It may be easier to talk about centile to a client, but the issue of short
confidence intervals is important to me. Also, for multiple imputation, you
can't use centile, so that tips the balance for me.
Thanks very much to Martin. I appreciate his good comments on all issues.
Tony
Peter A. Lachenbruch
Department of Public Health
Oregon State University
Corvallis, OR 97330
Phone: 541-737-3832
FAX: 541-737-4001
-----Original Message-----
From: [email protected]
[mailto:[email protected]] On Behalf Of Martin Weiss
Sent: Thursday, March 25, 2010 9:24 AM
To: [email protected]
Subject: AW: st: finding means and percentiles with mim
<>
Those percentiles are slightly off, however:
*************
clear*
set obs 1000
set seed 234232
gen x=rgamma(2,2)
sqreg x, quantiles(10 25 50 75 90) reps(2)
centile x, centile(10 25 50 75 90)
*************
HTH
Martin
-----Ursprüngliche Nachricht-----
Von: [email protected]
[mailto:[email protected]] Im Auftrag von Lachenbruch,
Peter
Gesendet: Donnerstag, 25. März 2010 16:40
An: '[email protected]'
Betreff: RE: st: finding means and percentiles with mim
I agree with the message. However, what I am looking for is a descriptive
statistic, so the command sqreg y, quantile(10 25 50 75 90) will give me
the percentiles that I want.
Here is some output.
. mim:sqreg lck ,quantile(10 25 50 75 90)
Multiple-imputation estimates (sqreg) Imputations = 20
Minimum obs = 432
Minimum dof = 386.9
-------------------------------------------------------------------------
lck | Coef. Std. Err. t P>|t| [95% Conf. Int.] FMI
---------+----------------------------------------------------------------
_cons | 5.01997 .118927 42.21 0.000 4.78621 5.25373 0.008
---------+----------------------------------------------------------------
/q25 | 5.79006 .081047 71.44 0.000 5.63076 5.94937 0.017
/q50 | 6.95493 .132365 52.54 0.000 6.69472 7.21513 0.030
/q75 | 8.52631 .170016 50.15 0.000 8.19204 8.86058 0.050
/q90 | 9.56167 .132408 72.21 0.000 9.30136 9.82199 0.045
--------------------------------------------------------------------------
Tony
Peter A. Lachenbruch
Department of Public Health
Oregon State University
Corvallis, OR 97330
Phone: 541-737-3832
FAX: 541-737-4001
-----Original Message-----
From: [email protected]
[mailto:[email protected]] On Behalf Of Martin Weiss
Sent: Wednesday, March 24, 2010 12:48 PM
To: [email protected]
Subject: RE: st: finding means and percentiles with mim
<>
" the command qreg is supported by mim, so I think I can use it."
Note the cautionary tale in this thread, though:
http://www.stata.com/statalist/archive/2009-11/msg01343.html
HTH
Martin
*
* 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/
*
* 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/
*
* 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/
*
* 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/
*
* 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/