RE: Re: st: RE: AW: ratio function

["Roman Kasal" <kasal@trexima.cz>]

ok, for this purpose I agree, that is ok...but what about if I want to calculate SE of Mean in years 2009 and 2008 and then ratio with SE of the means?

the problem is that CI of
"svy: mean wage, over(year)" is not equal "svy: mean wage if year==2009"
for the year 2009 because of different degrees of freedom (SE's are equal), the first command gives wrong CI.

is any elegant solution to handle this in Stata with "nlcom" or do I have to calculate it manually?

thank you

-----Original Message-----
From: owner-statalist@hsphsun2.harvard.edu [mailto:owner-statalist@hsphsun2.harvard.edu] On Behalf Of Steve Samuels
Sent: Thursday, April 22, 2010 3:08 PM
To: statalist@hsphsun2.harvard.edu
Subject: Re: Re: st: RE: AW: ratio function

The degrees of freedom are correct.  See any sampling text.

Briefly: To identify a subpopulation, each observation in the sample
receives a 0-1  indicator variable d.   If X is the numerator variable
 and Y is the denominator variable, the numerator for the ratio of is
the sum *over the entire sample*  of Z_x = d *X and the denominator is
the sum of Z_y = d * Y.  The standard errors are based on variability
in the Z's, including the zero values.
By the way, the standard errors formulas are valid only if the
expected number of observations in a subpopulation is at least 20.

Steve

2010/4/22 Roman Kasal <kasal@trexima.cz>:
> thank you for the code, but I have found a problem:
>
> if I calculate over(foreign) the bound are enumerated with "e(N_psu)-e(N_strata)" degrees of freedom, but not for each foreign (degrees of freedom are for whole dataset) and this is wrong I assume.
>
> thank you
>
> -----Original Message-----
> From: owner-statalist@hsphsun2.harvard.edu [mailto:owner-statalist@hsphsun2.harvard.edu] On Behalf Of Steve Samuels
> Sent: Friday, April 02, 2010 2:58 PM
> To: statalist@hsphsun2.harvard.edu
> Subject: Re: Re: st: RE: AW: ratio function
>
> Roman
> Perhaps we misunderstand what you are asking for. I  We have been
> assuming that you  want the ratio of the means of two variables
> ("columns"?) measured possibly on the same person.  Perhaps you want
> the ratio of the means of one variable for two subpopulations.   Both
> analyses will ignore missing values.
>
> If this is not what you desire, then please demonstrate by hand what
> you do want on a small, non-survey data set.. Also I'd like to know
> which R function does what are asking for
>
>  The following do file computes the ratio of means with CI and then
> does the same for the log ratio and transforms to the original scale.
>
> -Steve
>
>
> **************************CODE BEGINS**************************
> capture program drop _all
> program antilog
> local lparm  el(r(b),1,1)
> local se    sqrt(el(r(V),1,1))
> local bound  invttail(e(df_r),.025)*`se'
> local parm  exp(`lparm')
> local ll  exp(`lparm'  - `bound')
> local ul  exp( `lparm' + `bound')
> di  "parm =" `parm'  "    ll = " `ll'  "   ul = " `ul'
> end
>
> sysuse auto, clear
> svyset _n
> svy: mean mpg, over(foreign)
> nlcom (myratio1: _b[Domestic]/_b[Foreign])   //ratio
> nlcom (myratio2: log(_b[Domestic]/_b[Foreign]))   // log ratio
> // Confidence interval of last -nlcom- on antilog scale
> antilog
> ***************************CODE ENDS***************************
>
>
>
> .
>
> On Fri, Apr 2, 2010 at 2:37 AM, Roman Kasal <kasal@trexima.cz> wrote:
>> I don't agree...so how to do it when you want to find out ratio between
>> years, male X female, ...? So there is no solution? Just to keep N,mean,
>> SE, degrees of freedom, N_strata, N_psu, .... and calculate it manually?
>> I think it is not appropriate solution, at least to have it as an
>> option. I think there is missing a lot with complex survey in Stata and
>> complex survey is needed for almost every survey research, even freeware
>> R-project is better equipped :(
>>
>> so have a hope Stata will get it soon....immediately we are buying it
>> again :)
>>
>>>
>>>
>> And it should.   Data (x,y) (1,2) (2,4) (3,6) (100,.)    will give an
>> entirely different view of the data if the unpaired observation is
>> included in a mean or ratio calculation.  Or consider data with x
>> missing in half the pairs and y missing in the other half; the ratio
>> of means would be meaningless.
>>
>> The formulas for standard errors for ratios  assume that the data are
>> paired. Formally, they are based on the residual MSE of a regression
>> of y on x through the origin. You cannot do that regression with
>> unpaired data.
>>
>> If your concern is missing data, the solution is to impute the missing
>> values before analysis.
>>
>> Steve
>>
>>
>> *
>> *   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/
>>
>
>
>
> --
> Steven Samuels
> sjsamuels@gmail.com
> 18 Cantine's Island
> Saugerties NY 12477
> USA
> Voice: 845-246-0774
> Fax:    206-202-4783
>
> *
> *   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/
>



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

*
*   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/