[Date Prev][Date Next][Thread Prev][Thread Next][Date index][Thread index]

Re: st: question about weights in histograms

From   Steven Joel Hirsch Samuels <>
Subject   Re: st: question about weights in histograms
Date   Wed, 10 Oct 2007 13:03:06 -0400

Good call, Austin! All multiples of the original weight will give the same result for means and percentages.

On Oct 10, 2007, at 12:44 PM, Austin Nichols wrote:

Or, better, multiply the weight by the reciprocal of the smallest
increment ("units" in -codebook- output) and use that as a frequency
weight, which should give you identical point estimates to -svy-

use hhinc mars using cpsmar07
g double wt=round(marsupwt*100)
hist hhinc [fw=wt], xscale(log) percent

On 10/10/07, Steven Joel Hirsch Samuels <> wrote:

Round the weights to the nearest integer and use these as frequency
weights.  -histogram- is not a survey command and so is unaffected by
-svyset-.   The results will be nearly identical to  what you would
get if you estimated  the per cent in each category using -svy:
proportion-.  By the way, in most complex surveys the weights have
been trimmed to reduce total error: bias^2 + se^2.  So you cannot
expect 'unbiased' estimates even with the original weights.


On Oct 10, 2007, at 12:02 PM, Scott Cunningham wrote:

I am using 5% Census and March CPS data (United States) to construct
histograms.  These data include probability weights.  I know that
histogram can adjust for frequency weights, but apparently not
explicitly for probability weights that are not integers.  If I
-svyset- the data using the perweight, will -histogram- construct
unbiased estimates of the population?  Here is my code

. svyset [perweight=perwt]
. histogram tax, by(white, col(1)) percent
*   For searches and help try:
Steven  Samuels
18 Cantine's Island
Saugerties, NY 12477
Phone: 845-246-0774
EFax: 208-498-7441

*   For searches and help try:

© Copyright 1996–2017 StataCorp LLC   |   Terms of use   |   Privacy   |   Contact us   |   What's new   |   Site index