mai mai,
references for fracpoly cited in the stata manual are:
Royston and Altman 1994 Applied Statistics 43: 429-467 as well as
several references in the Stata Bulletin. Maybe these can help you.
If you do a search on fractional polynomial regression in pubmed or
google scholar you will find other relevant publications. I do not
consider it "data mining", but an approach to explore non linear
relationships.
Good luck, Tim
On 12/18/07, mai mai <[email protected]> wrote:
> Hi,
> I have couple of questions on Fracpol, I tried finding answers through
> help and the internet but was unsuccessful. The results of my model are
> posted below.
>
> 1) How acceptable is the use of fractional polynomial in the
> literature. Is it considered data mining?
>
> 2) What is the structure of the best relation?
> Is it: y=b0 + b1 X^-.5 + X^-.5*ln(X) ?
>
> 3) Why is X = (dist+2.860107421875)/1000)? why the addition of 2.8
> and division by 1000?
>
> 4) What is the exact meaning of deviance? Is lower (more negative)
> better or worst?
>
> . fracpoly reg b dist, adjust (no) compare
> ........
> -> gen double Idist__1 = X^-.5 if e(sample)
> -> gen double Idist__2 = X^-.5*ln(X) if e(sample)
> (where: X = (dist+2.860107421875)/1000)
>
> Source | SS df MS Number of obs = 57
> -------------+------------------------------ F( 2, 54) = 3.45
> Model | .014450698 2 .007225349 Prob > F = 0.0390
> Residual | .113204826 54 .002096386 R-squared = 0.1132
> -------------+------------------------------ Adj R-squared = 0.0804
> Total | .127655524 56 .002279563 Root MSE = .04579
>
> ------------------------------------------------------------------------------
> y | Coef. Std. Err. t P>|t| [95% Conf. Interval]
> -------------+----------------------------------------------------------------
> Idist__1 | .04901 .0190668 2.57 0.013 .0107834 .0872366
> Idist__2 | .007933 .0031175 2.54 0.014 .0016829 .0141832
> _cons | -.0443816 .0171458 -2.59 0.012 -.0787569 -.0100063
> ------------------------------------------------------------------------------
> Deviance: -192.87. Best powers of distance among 44 models fit: -.5 -.5.
>
> Fractional polynomial model comparisons:
> ------------------------------------------------------------------------------
> dist df Deviance Res. SD Dev. dif. P [*] Powers
> ------------------------------------------------------------------------------
> Not in model 0 -186.025 .047745 6.848 0.174
> Linear 1 -188.380 .047192 4.492 0.247 1
> m = 1 2 -189.268 .046825 3.604 0.193 .5
> m = 2 4 -192.873 .045786 -- -- -.5 -.5
> ------------------------------------------------------------------------------
> [*] P-value from deviance difference comparing reported model with m = 2 model
>
>
> Many thanks for your help
> *
> * 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/
>
> *
> * 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/
>
*
* 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/
