Thanks - I did check using summarize with weights, and other tests
(sktest), and qnorm/pnrom, and generally skewness is no problem, but
for some subsamples it may be. I am jconcerned that stratification is
lost by these views.
Could you please clarify the nlcom formula, please?
nlcom (_b[ht3] - 3*_b[ht2] * _b[ht] + 2*_b[ht])/(_b[ht2] - _b[ht] * _b[ht])^3/2
(ht being height, ht = ht^2 and ht3 = ht^3)
I suspect my translation is not correct as the coefficients I get are
not close to skewness (and is it the formula for skewness?).
Thanks again.
Richard
On Mon, Nov 3, 2008 at 3:26 PM, Nick Cox <[email protected]> wrote:
> Stas gives good advice, except that the implication of his last
> throw-away remark is incorrect. Far from being exotic, standard errors
> for skewness are often used, even if tacitly. Many people use tests for
> (non-zero) skewness, (non-Gaussian) kurtosis, or both, based on
> estimates of skewness and/or kurtosis and their standard errors. See
> e.g. -varnorm-. These tests are often attributed to Jarque and Bera by
> economists, on the usual grounds
>
> 1. they wrote about them
> 2. they are economists
> 3. econometricians seem to have a bias to naming tests after people,
> rather than using names to describe what they do (I owe this observation
> to Stas himself).
>
> These criteria do not include the more convincing grounds that Jarque
> and Bera invented or discovered them.
>
> More importantly, these tests are based on asymptotic ses when there is
> plenty of evidence that they are a lousy approximation for small
> samples.
>
> My own biases are that you want to examine for normality you are best
> off with -qnorm- and that if you want a test regardless you are better
> off with -omninorm- from SSC. But neither is adapted to the survey case.
>
>
> Nick
> [email protected]
>
> Stas Kolenikov
>
> On 11/3/08, Richard Palmer-Jones <[email protected]> wrote:
>> I hope this is not a stupid question, but how can I get skewness
>> estimates from svyset data? If it is, please enlighten me.
>
> If you want to eye-ball it, you can run just the regular -summarize
> [fw=], d-, and keep your fingers crossed that the algebra is the same
> with -fw- and -pw-. A better way would be to create the necessary
> moments of your variables, and make that nonlinear skewness formula
> out of them:
>
> g x2 = x*x
> g x3 = x*x
> svy: means x x2 x3
> nlcom (x3-3*x2*x+2*x)/(x2-x*x)^3/2
>
> or whatever the proper -nlcom- syntax is. I bet you've never seen
> standard errors on skewness coefficient in your life :))
>
> --
> 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/
>
*
* 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/
