Thanks very much. I probably wasn't too clear in my original post
I am examining the relationship between HIV testing behaviors (whether
people plan to take the test, etc) and a number of socio-demographic
and epidemiological variables. At a minimum, I wanted to understand
the distribution of all my predictors before running crostabs
procedures.
I know svy: tab can produce both the estimated proportions (just like
the %s displayed in frequency tables) and the crosstab estimates. And
both of these allow one to make assumptions about the underlying
population.
Hence the question -- whether it is appropriate to use the svy:
estimation commands once I am dealing with such a survey sample... or
to revert to non-survey commands.
thanks - CY
----------------------
On Tue, Jul 29, 2008 at 12:36 PM, Stas Kolenikov <[email protected]> wrote:
> On 7/25/08, Chao Yawo <[email protected]> wrote:
>> I am using svy commands to analyze a DHS dataset.
>>
>> As a usual prerequisite, I want to run some descriptive statistics on
>> my sample. I can use the regular tabulate or fre commainnd to produce
>> frequency distributions.
>>
>> However, I realized that svy has a "tabulate" or "proportions" option
>> that could produce frequency distributions/estimates per variable. I
>> run both and realized slight differences between the two frequencies
>> outputs.
>>
>> Which one should I use - I am leaning towards using the one with the
>> svy: prefix.
>>
>> I would appreciate any thoughts and pointers.
>
> Well as Steven said, what is it exactly that you want to figure out?
> If you want to see whether you have cells with zero or low counts,
> then either -tab- or -svy : tab- will do. If you want to get any idea
> of the underlying population, you MUST use -svy-.
>
> Let's think through a grocery shopping example. Suppose somebody
> looked at your fridge and counted how many gallons of milk you have
> there, how many eggs, the total weight of vegetables, etc. If they
> want to figure out a diet of a given person, then that's all the data
> they need. If they wanted to figure out what's available in your
> grocery store, or what's a diet of an average person, then there is
> more work to do: they need to figure out how often you buy any
> particular food. May be you are a vegetarian, and skip the meat rows
> in your supermarket -- so your fridge will not provide any information
> about meat consumption, and estimates of protein intake based on your
> fridge only will be biased. The "how frequently" question is what you
> also know as sampling weights, based on inverse probabilities of
> selection.
>
> So if you want something that's specific to your sample, you can have
> a go without -svy- options. Will that be interesting to anybody?
> Probably not. Whichever summaries you want to produce out of your data
> will only be interesting to the extent that they describe the
> population -- and then you need to use the survey design information.
>
> --
> Stas Kolenikov, also found at http://stas.kolenikov.name
> Small print: I use this email account for mailing lists only.
> *
> * 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/
