Statalist


[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]

st: Permbands : question about bord effects


From   "emilie.josserand" <emilie.josserand@laposte.net>
To   statalist@hsphsun2.harvard.edu
Subject   st: Permbands : question about bord effects
Date   Wed, 15 Jul 2009 14:57:03 +0200 (CEST)

Dear all,
 
I try to understand how permbands are performed, and there is an incoherence between what I understand and my results.
Here is what we find in the help :
 
In many applications it may be hypothesised that yvar is independent of xvar.
permband first smooths yvar^moment on xvar using a locally linear smoother
(running). The process is then repeated reps(#) times, with the order of
the yvar values permuted at random at each replicate. The minimum and maximum
values of the resulting smooths at each value of xvar form the "permutation
band". This band and the original smooth are plotted against xvar. 
If yvar^moment and xvar are independent we would expect the original smooth to
wander between the lower and upper boundaries of the permutation band. If
it crosses a boundary somewhere, there may be evidence of dependence. Based
on 50 replications and the default span, the proportions of values expected 
outside the permutation band have been evaluated and are formulated as a 
hypothesis test. For more details see Royston and Wright (1998).

I use permband with a variable which is supposed standard normal, and independant of age.
My syntax is : 
permband Z_ml age, moment(1)
What I don't understand is why the permbands are always larger at the extreme values of age,
whereas the permutation are made with the same values whatever the age : 
Is there an age-grouping for performing permutations ?
Thanks, 
Emilie

 Créez votre adresse électronique prenom.nom@laposte.net 
 1 Go d'espace de stockage, anti-spam et anti-virus intégrés.


*
*   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/



© Copyright 1996–2014 StataCorp LP   |   Terms of use   |   Privacy   |   Contact us   |   What's new   |   Site index