Re: st: -svyset- parameters for survey with multiple waves

[Steve Samuels <sjsamuels@gmail.com>]

Without a PSU variable, I see no way that you can set up or do a correct survey analysis. 

I'm not familiar with Gallup surveys, so, from your description, can't really tell how to set up the data even with PSU information. (Were PSUs re-used?) But in one publication that I looked at (*), Gallup statisticians themselves quoted bounds of error and design effects that ignored the clustering: this is not an option for valid inference.  

I suggest that you search the study manual for guidance on how to estimate standard errors or do tests for inferential analyses; in the absence of that information, contact Gallup statisticians and ask them to supply you with a version of the data that includes the PSU variables or, at least, with design effects that reflect the clustering by PSUs.


Steve



(*) http://www.gallup.com/poll/161675/remittances-payments-money-transfers-behaviors-south-asians-indonesians.aspx page 68



Steve

On Nov 5, 2013, at 3:20 AM, Mohamad Mahmoud wrote:

Hi,
I'd appreciate your insight on this, specifically on how to set up a
survey with multiple waves.
thanks,
Mohamad

On Mon, Nov 4, 2013 at 1:40 PM, ksgksg <ksgksg@gmail.com> wrote:
> Hi,
> I am working with a Gallup survey conducted across 20 countries ("wp5"
> is the country identification variable ), each country was sampled in
> 7 independent waves, that is respondents in each wave are different.
> The survey has around 1000 respondents per country per wave. Each
> respondent is assigned two weights, 1) "wgt" is a within sample weight
> that adds up to the 1000 per wave. 2) "projwt" which is the weight for
> the respondent within the country and adds up to the eligible
> population sample for each country in each wave. My objective is to
> identify certain cross-country effects over the time of the different
> waves.
> 
> I am new to analyzing surveys, so I need guidance in correctly
> declaring the survey design for Stata using -svyset-. Specifically:
> 
> a) I am not sure which variable to use as PSU and strata given that
> the data only provides weights but not PSUs. The survey says that it
> was conducted by first selecting the PSUs within each country to be
> statistically representative. The survey manual says the design was
> based on 3 stages of selection: 1) PSUs consist of clusters of
> households which were stratified by population size. 2) random routes
> used to select households within each PSU. 3) Respondents selected
> randomly within each household (there is a variable "wp12" identifying
> the number of potential eligible respondents per households ).
> 
> b) How can I capture the fact that the survey has multiple waves per
> each country in -svyset- so that the population estimate is not (7 *
> the sum of population across countries) but rather the true
> population.
> 
> Thank you in advance for your kind assistance and time.
> Best,
> 
> Mohamad Mahmoud
> *
> *   For searches and help try:
> *   http://www.stata.com/help.cgi?search
> *   http://www.stata.com/support/faqs/resources/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/faqs/resources/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/faqs/resources/statalist-faq/
*   http://www.ats.ucla.edu/stat/stata/