Re: st: svyset with raked weights

[Steven Samuels <sjsamuels@gmail.com>]

Fernando, you can safely use the raked weights only for statistics about the population as a whole, that is all three regions together. If the age and sex, occupation,  industry, and industry distributions are not the same in each region, then the raked weights might give biased results for estimates for  any single region.

You ascertain if this is a problem by estimating age/sex, occupation, and industry distributions in each region, with -svy: prop- . First  -svyset- with pweights equal to the design probability weights-assuming that they are among the variables in the data set.

If you can get region-specific age-sex, occupation, or industry proportions from a recent census, then you can rerake the original probability weights. To do this,  use - survwgt rake-, which is contain in Nick Winter's -svr- package (from SSC). An alternative is  John D'Souza's -calibrate-, also available from SSC.

If you don't have such external information, then I'd suggest just using the original design probability weights for your region-specific analyses.


Steve


On Mar 8, 2011, at 1:11 PM, Fernando Terrés wrote:

Steve,
1. Stratum variables are municipalities (66) that were sampled to get the phone numbers. Some of these strata are small, and are not used in the statistical analysis. The interest lies in groups of municipalities, which can be grouped in regions (20), and autonomous areas (3). The main regional interest is on these three areas.
2. The purpose of the study is mainly descriptive of working conditions (univariate and multivariate, and multiple correspondence analysis). I think that for this purpose I could simply use the raked weights, but I would like to be able to make inferences (if it is not complicated).
3. Primary sample units are households. Secondary sample units are employees in the household, but with the peculiarity that only one individual is selected in each household (selection probabilities are given in terms of total number of individuals in each household and total number of employees in the household).
4. Primary sample units were selected at random. The individual to interview was also selected at random, if more than one lived in the household.
Many thanks.





On Mar 8, 2011, at 11:30 AM, Steve Samuels wrote

Fernando-
Please provide more information.

1. Name of the stratum variable
2. Purpose(s) of the survey (e.g. to estimate descriptive statistics; to fit regression models; to test hypotheses).
3. Names of the primary sampling unit (unit selected at first stage) and, if available, units selected at subsequent stages.
4. Whether any of the primary sampling units were selected with certainty-i.e. not by random numbers. In the study documentation, these might also be called "self-representing" units.


Steve


Steven J. Samuels
Consultant in Statistics
18 Cantine's Island
Saugerties, NY 12477 USA
Voice: 845-246-0774
Fax:   206-202-4783 
sjsamuels@gmail.com



On Mar 8, 2011, at 10:48 AM, Fernando Terrés wrote:

I have a question on setting the correct design for a multi-stage stratified design. It is an official household survey, were the strata are formed by using (known) population census data, and only one individual is randomly selected among the (number known) eligible individuals (employees) in each household. The dataset also includes raked weights (they call it post-stratification weights), for population totals of four relevant variables (age and sex -6-, occupation -8-, industry -4-, province -3-).
The question is which svyset must I use with Stata in order to get correct standard errors?
Thanks in advance.


-- 
Fernando Terres
Lecturer. PhD candidate
CERPIE/Research
Universitat Politecnica de Catalunya
Diagonal, 647 - 10
08028 Barcelona (Spain)
+34.934.010.708
fernando.terres@upc.edu
http://cerpie.upc.edu
*
*   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/