Re: st: Questions about svy commands

[Steven Samuels <sjhsamuels@earthlink.net>]

Gaby,

For analyzing the children there are two issues:

1. Are the existing weights appropriate for the children?  To answer  
this I would need more information about the survey. How did children  
get into the sample?  As part of selected households? Is there a  
different weight for each household member, (males and females for  
example) or is the weight the same for all members of the household?   
Were the data post-stratified in any way? If there is just one weight  
for all members of the household, then use that.

2. Do you select just the children for an analysis data set, or do  
you analyze the entire set and use the -subpop- option? The second  
approach is the only one which will provide entirely correct standard  
errors, although often there will be little difference.  Austin  
Nichols showed how to create a data set for use with the -subpop-  
option that will be only a little larger than one containing only  
children.  See: http://www.stata.es/statalist/archive/2007-11/ 
msg00810.html .

3. Although -svy- does not work with -areg-, you can use -areg- with  
a -weight- option and with the proper PSU as the cluster variable.   
You will be unable to use the -strata- option, and this could  
potentially lead to estimated standard errors that are larger than  
the true ones. It will also artificially increase the degrees of  
freedom for error.  You can get around these by adding dummy  
variables for stratum into your model. If strata are defined by your  
“province” variable, then you have effectively done that.

4. If there are too many strata to add as dummies (and strata are not  
defined by your provinces), ignore the strata in the analysis, but  
adjust the degrees of freedom by hand. The proper degrees of freedom  
for error will be the listed d.f. minus the number of strata. You can  
compute correct confidence intervals, say 95% intervals, as follows:

4.1. Find the error degrees of freedom from the -areg- output WITH  
the the -cluster- option.  Suppose it is, df1 = 180.  If you had 80  
strata, the degrees of freedom should be df2 =180 - 80 = 100.

4.2. With 180 degrees of freedeom, the t-multiplier for a standard  
error would be 1.973, but this is too small. Compute the t-multiplier  
for the correct degrees of freedom  and 95% CI as  invttail 
(100,.025), or 1.9840.

4.3. You should INCREASE the nominal confidence level for -areg-, so  
that the t-multiplier with 180 d.f. is 1.9840.  What should the level  
be?  First find:  ttail(180,1.9840), or 0.02439.  The proper -level-  
is then: 1- 2x.02439=0.951.  So you should specify a -level-  
statement as “set level 95.12”.

You can find the proper level in one step by:

 di 1-2*ttail(df1 , invttail(df2,.025))
//finds level where df1 is the nominal degrees of freedom and df2 is  
the actual degrees of freedom =df1- n. strata.

-Steven

On Feb 9, 2008, at 7:04 AM, Ana Gabriela Guerrero Serdan wrote:


> Dear all,
>
> Sorry for these probably obvious questions. Have
> looked into the archives  but I'm still confused on
> the following issues:
>
> 1) I am using survey data (two-stages with
> stratification). I am looking at children less than
> five years old.  Can I apply svy set as usual to my
> sub-sample of children as follows?
>
> svyset [pweight= expweigh],  strata(AI05) psu( AI06)
>
>
> 2) I had initially done my analyis with linear
> ressions without the svyset, controlling for
> differences in provinces and cohorts, and clustering
> at the district level. I used areg as follows:
>
> areg Y X DummiesProvinces, vce(cluster district)
> absorb(mdate)
>
> What command can I use if I first set my data for
> svyset?
>
>
> Gaby Guerrero Serdan
>
> Deparment of Economics
> Royal Holloway, University of London
> TW20 OEX
> Egham, Surrey
> England, UK
> http://www.rhul.ac.uk/economics/About-Us/postgrads.html
> http://www.flickr.com/photos/49939890@N00/show/
>
> Tel: +44 7912657259
>
>
>        
> ______________________________________________________________________ 
> ______________
> Be a better friend, newshound, and
> know-it-all with Yahoo! Mobile.  Try it now.  http:// 
> mobile.yahoo.com/;_ylt=Ahu06i62sR8HDtDypao8Wcj9tAcJ
>
> *
> *   For searches and help try:
> *   http://www.stata.com/support/faqs/res/findit.html
> *   http://www.stata.com/support/statalist/faq
> *   http://www.ats.ucla.edu/stat/stata/

*
*   For searches and help try:
*   http://www.stata.com/support/faqs/res/findit.html
*   http://www.stata.com/support/statalist/faq
*   http://www.ats.ucla.edu/stat/stata/