It is also possible to transform to logs,
do the density estimation and then
back-transform the density function using
a standard probability result. This
is an option in -mdensity- from SSC
and you could do it with official
Stata with a few commands before and
after. The results in practice might
well resemble some adaptive method.
In my experience the back-transformation
is only helpful visually if the distribution
is moderately skew. With strong skewness
it is often more practical to transform to
logs and stay in log space. In the case
of income people in this field seem very
used to think about income on a log scale
anyway.
Nick
[email protected]
Ramani Gunatilaka
> This is not a programming question but one on theory. I would
> be grateful for any views about this.
> I am trying to decompose income inequality using DiNardo,
> Fortin and Lemieux's (1996) semi-parametric methodology
> (Econometrica 64(5), 1001-44) as applied to the kernel
> density function.
> However, DiNardo et al, Deaton (Analysis of Household Surveys
> - 1997) and D'ambrosio (Review of Income and Wealth 47(1),
> 2001) transform the data into log-form and work with that.
> This is because the kernel estimator has difficulties
> handling densities that have a high degree of asymmetry as is
> the case with income data. Densities close to normality are
> the easiest for the kernel estimator to estimate.
> Now my question is this:
> Would I be able to get round this problem of the kernel not
> performing well if I were to estimate adaptive kernel density
> (akdensity) on the original data?
> The adaptive kernel density curve, though smoother than the
> kernel density function, is also skewed or asymmetric.
> Wouldn't this matter?
> Would someone know of a useful reference in this regard?
*
* For searches and help try:
* http://www.stata.com/support/faqs/res/findit.html
* http://www.stata.com/support/statalist/faq
* http://www.ats.ucla.edu/stat/stata/
