Reply from Measure DHS
============
Dear Olalekan,
Any of these methods are acceptable,
depending on your
knowledge & experience of regressions.
Some analysts prefer to use the weights while others
do not. For
example, the main author of WP #31 always uses the DHS
weights, while
the author of the DHS Guide to statistics prefers not
to.
I suggest investigating the appropriateness of weights
in regression
analysis to come to a decision that you're comfortable
with.
=============
Method I - From DHS statistics guide:
"Use of sample weights is inappropriate for estimating
relationships, such as regression and correlation
coefficients".
Reference:
http://www.measuredhs.com/help/Datasets/sampling_weights.htm
Method II - From WP #31
"For all models, our analysis accounts
for complex DHS survey design to estimate efficient
regression coefficients and robust standard errors
adjusted for intra-cluster correlation, and
sampling weights are applied in accordance with
standard DHS procedures to ensure the
representativeness of the samples."
Reference:
Mishra, Vinod, et al. 2007. A study of the association
of HIV
infection with wealth in sub-Saharan Africa. DHS
working paper no. 31.
Calverton: Macro International.
http://www.measuredhs.com/pubs/pub_details.cfm?ID=638
Olalekan Uthman
--- Emma Slaymaker <[email protected]> wrote:
> Dear Olalekan,
>
> You asked about accounting for the survey design
> when analysing DHS
> data. You should account for the stratification and
> the clustering
> whenever you calculate standard errors- if you do
> not, the standard
> errors will not be correct. I always include the
> weighting information
> since I usually want to know if associations are
> likely to be true of
> the population the survey is designed to represent
> (with weights) and
> not just the survey sample (without weights). Once
> you have accounted
> for the stratification and clustering you don't
> really lose anything by
> also accounting for the weights since you are
> already working with
> robust standard errors.
> I don't know of a reason why it would be
> inappropriate to use sample
> weights in a regression analysis- but that may well
> be ignorance on my
> part.
> Hope this helps,
> Emma
>
> >>> Olalekan Uthman <[email protected]> 30/07/07
> 15:49 >>>
> Dear Stata users,
>
> I am trying to run a logistic regression using
> Measure
> DHS survey data. Quoting from DHS statistics guide:
> "Use of sample weights is inappropriate for
> estimating
> relationships, such as regression and correlation
> coefficients.
>
> I am not sure whether to ignore the psu and strata
> also when running the logistic regression.
>
>
> Can I just use
>
> xi: logistic bmi i.edu
>
> instead of:
>
> svyset psu [pw=weight], strata(strata)
>
> xi: svy: logistic bmi i.edu
>
>
>
>
____________________________________________________________________________________
> Looking for a deal? Find great prices on flights and
> hotels with Yahoo!
> FareChase.
> http://farechase.yahoo.com/
> *
> * 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/
>
____________________________________________________________________________________
Be a better Heartthrob. Get better relationship answers from someone who knows. Yahoo! Answers - Check it out.
http://answers.yahoo.com/dir/?link=list&sid=396545433
*
* 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/
