I intended the -subpop- to account for randomness in the sizes of the
two remaining subgroups, not to account for the fact that the third
subgroup was ignore.
I should have explained that the population size reported by Stata
is the sum of the probability weights for the sample members. That is
why I suspect that the weight variable is incorrect.
-Steve
On Fri, May 1, 2009 at 11:46 AM, Austin Nichols <[email protected]> wrote:
> Adebola Odunlami <[email protected]>:
>
> In re Steven's point (2): If one group is not asked the question, it
> is not a big deal IMHO. Using subpop or not will give the same answer
> when the variable is missing outside the subpop, in general.
>
> In re Steven's point (3): you might want the se or ci option in
> addition to row option, so that the row proportions are saved in e(b)
> and you can test hypotheses across rows, e.g.
>
> clear
> webuse nhanes2
> svyset
> svy: tab black diab, row ci
> test _b[p22]=_b[p12]
> svy, subpop(if age>59): tab black diab, row ci
> test _b[p22]=_b[p12]
> replace diab=. if age<=59
> svy: tab black diab, row ci
> test _b[p22]=_b[p12]
>
>
> On Fri, May 1, 2009 at 11:15 AM, <[email protected]> wrote:
>> Welcome to Stata! Reposting a question will not get it answered any
>> faster. See the Stata FAQ about why questions do not get answered.
>> In your case, your post is hard to read. Your questions are mixed up
>> with snippets of Stata code, and other statements.
>>
>> Important! Something appears wrong with your survey set up. You have
>> n = 5,191 observations, but Stata reports that the total population
>> size is only N = 3,061, You could be using the wrong probability
>> weight. Stop and fix this problem.
>>
>> To answer your questions.
>>
>> 1. If you have survey data, you must always -svyset- with the
>> appropriate setting. Consult your advisor if you do not understand
>> why, or look at Sharon Lohr, "Sampling, Design and Analysis", Duxbury
>> Books.
>>
>> 2. Your new race-ethnicity variable seems okay. You can
>> cross-tabulate it against the original variable to check.
>>
>> You have a potential problem if whites were not asked this question:
>> the survey weights might give misleading results if applied to
>> nonwhites alone. This is not an elementary matter and you should
>> consult someone who is knowledgeable about surveys for advice. Bring
>> the study documentation with you when you do this.
>>
>> 3. -svy: tab- with the tab() option doesn't do what you think it
>> does. Read the -help- to see why.
>>
>> I'm assuming that the new race-thnic variable is recoded 1 & 2.
>>
>> svy, subpop(if racethn==1), tab sex psymps, row
>> svy, subpop(if racethn==2), tab sex psymps, row
>>
>> -Steve
>>
>> On Fri, May 1, 2009 at 10:01 AM, Adebola Odunlami
>> <[email protected]> wrote:
>>> Hi all,
>>>
>>> I was hoping someone would please address my question below. Any help/insight is greatly appreciated.
>>> thanks in advance.
>>>
>>> I am relatively new to stata and I have a few related questions. I want to get proportion of units that answered a question( variable name = psysymps) by thier gender(sex) and race (race3cat) after I survey set the data. I tried "bysort race3cat: tab psysymps sex, row col chi2" but then realized the percentages were off and I needed to use svy commands. I then tried "svy: tab racet3cat sex, tab (psysymps) row se ci" but got a response that the table contains 0 in the marginals and therefore statsitics cannot be computed. I realized that one of the race categories (e.g. White) was not asked the question. So I created a new race category ("blackethn) that sets Whites to missing and others to either 1 or 2. Then I proceeded with my primary objective to see proportions of 'psysymps' by race and sex and tried "svy: tab blackethn sex, tab (psysymps) row se ci". This gave me the table below.
>>>
>>> My questions are: 1) must I always svy set before each code e.g svy: tab instead of just tab?; 2) was it okay to set Whites to missing as they were not asked this question and I dont want them counted as a denominator in my proportion?; 3) the table below still does not give me tabs of my variable by race and sex. how do I do this? Instead the table gave me---of those who answered yes to psysymps, what proportion of race/blackethn =1 were men and what prop were women? I would like to know how to code to get----what proportion of race/blackethn==1 men answered 'yes' to psysymps, compared to race/blackethn== 1 female (and also the proportion of race/blackethn ==2 men that answered yes to to psysymps, compared to race/blackethn== 2 female and associated chi2)
>>>
>
> *
> * 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/
