 |
[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]
Re: st: Comparing frequencies with pweight
Erick, I'd still like to know whether your data are from a sample,
and, if so, what design.
There is nothing wrong with the first table. Chi square is
proportional to sample size; take any table, double the sample size
and chi square will double. You do have a large N, but that's
reality; there is no need to "adjust" for it. What matters is the
percentages that are being compared. The two waves definitely do
differ w.r.t. the outcome (22.5% vs 13.9%) You are asking very
different questions if you do and do not weight by number of
clients. So, you must decide what your research question is.
-Steve
On Nov 15, 2008, at 10:38 PM, Erick Guerrero wrote:
Thanks for the reply Steve, yes, x is an attribute of orgs - for
instance - dummy of whether they are for profit or not.
So, tab forprofit wavenumber (pweight= number of clients)
my total sample for both waves is 1118 orgs
I basically want to run a chi-square or a t-test between an org
attribute and wavenumber but weight by number of clients.
I tried the svyset - (pweight=number of clients), and then use svy:
tab forprofit wavenumber - but I get an ackward result.
or the
*Normal Chi-square has a huge N*
. tab pfp wavenum, row col chi
+-------------------+
| Key |
|-------------------|
| frequency |
| row percentage |
| column percentage |
+-------------------+
| wavenum
pfp dummy | 4 6 | Total
-----------+----------------------+----------
0 | 527 430 | 957
| 55.07 44.93 | 100.00
| 86.11 77.48 | 82.01
-----------+----------------------+----------
1 | 85 125 | 210
| 40.48 59.52 | 100.00
| 13.89 22.52 | 17.99
-----------+----------------------+----------
Total | 612 555 | 1,167
| 52.44 47.56 | 100.00
| 100.00 100.00 | 100.00
Pearson chi2(1) = 14.7018 Pr = 0.000
. tab pfp wavenum [pweight= nuosatcl], row col chi
pweight not allowed
*Using survey option - reports low column percentages.
*
. svyset
pweight: nuosatcl
VCE: linearized
Single unit: missing
Strata 1: <one>
SU 1: <observations>
FPC 1: <zero>
. svy:tab pfp wavenum
(running tabulate on estimation sample)
Number of strata = 1 Number of obs
= 1121
Number of PSUs = 1121 Population size
= 696235
Design df = 1120
-------------------------------
| wavenum pfp dummy | 4
6 Total
----------+--------------------
0 | .4576 .3995 .857
1 | .048 .095 .143
|
Total | .5056 .4944 1
-------------------------------
Key: cell proportions
Pearson:
Uncorrected chi2(1) = 21.5824
Design-based F(1, 1120) = 5.1446 P = 0.0235
*For MEANS*
. svy: mean latinclients, over(wavenum)
(running mean on estimation sample)
Survey: Mean estimation
Number of strata = 1 Number of obs = 1088
Number of PSUs = 1088 Population size = 673329
Design
df = 1087
4: wavenum = 4
6: wavenum = 6
--------------------------------------------------------------
| Linearized
Over | Mean Std. Err. [95% Conf. Interval]
-------------+------------------------------------------------
latinclients |
4 | 11.36198 .9582807 9.481693 13.24227
6 | 17.05415 1.98646 13.15642 20.95188
--------------------------------------------------------------
. test [latinclients]4 - [ latinclients]6=0
Adjusted Wald test
( 1) [latinclients]4 - [latinclients]6 = 0
F( 1, 1087) = 6.66
Prob > F = 0.0100
.
.
. Erick
Steven Samuels wrote:
Martin's suggestions of -svy: tab- , with "organization'
designated as the psu, is probably the right one. However, I'm not
sure. With your specification "x" and "y" must be attributes of
the organizations. One must be "wave". What is the other? -
pweight- will weight the organizations by the number of clients.
So, essentially you are classifying clients by the attribute of
their organization and ignoring organization , except to compute
standard errors. For example, if you classify organizations as
"big" and "small", -svy: tab- would compare the proportions of
clients who are in big organizations in 1995 and 2005 Is this what
you want?
Are the 618 organizations a sample of a population? If so, please
describe the design, for you need to take into account other
aspects of the design. Final question: what do you mean by "adjust
for N"?
-Steve
On Nov 14, 2008, at 12:24 PM, Martin Weiss wrote:
Have you checked out -help svy: tabulate twoway-?
HTH
Martin
-------- Original-Nachricht --------
Datum: Fri, 14 Nov 2008 09:52:33 -0600
Von: Erick Guerrero <[email protected]>
An: [email protected]
Betreff: st: Comparing frequencies with pweight
Hi, I need code for Stata to do descriptives - chi-squares with
p-weights and adjust for N.
I am trying to compare percentages between two waves (1995 and
2005) on
618 organizations.
The tab x y, col row chi with pweight (number of clients in each
organization) does not work, and when I use fweight gives me an
inflated
N. What is the proper way to compare these frequencies in
organizations
and weight by number of clients and adjust for N in stata?
Thanks,
Erick
*
* 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/
--
Martin Weiss
Hackersteigle 3
72076 Tübingen
00497071/793535
00491784597218
Der GMX SmartSurfer hilft bis zu 70% Ihrer Onlinekosten zu sparen!
Ideal für Modem und ISDN: http://www.gmx.net/de/go/smartsurfer
*
* 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/