Bookmark and Share

Notice: On April 23, 2014, Statalist moved from an email list to a forum, based at

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

Re: st: Regression splines with survey data

From   Stas Kolenikov <>
Subject   Re: st: Regression splines with survey data
Date   Wed, 8 Jun 2011 12:24:48 -0500

The splines are splines... some are smooth, some are not, so the
choice is basically the matter of aesthetics (do you want your
scatterplots to be curvy or zigzagged?) Eventually splines enter a
regression as explanatory variables, so you'd be running -svy: reg-
with any of them. If you have a strong theory that suggests a
particular type of splines, then of course you should be using these
splines, but other than that, I don't see anything that would
distinguish one type of splines over others.

I am assuming you are familiar with methodological work on
semi-parametric models for complex survey data (Jean Opsomer; Sharon
Lohr). I cannot claim familiarity with that sort of stuff,
unfortunately. I would expect the appropriate asymptotics to be quite
complicated. Splines probably have  their own weird sub-n^{-1/2}
convergence rates, and survey data must have a joint growing
asymptotic rates of the finite population and the sample, which still
leaves room for growth either in number of strata, number of clusters
per stratum, number of units per cluster, etc. Marrying the two must
be a real mathematical nightmare.

On Wed, Jun 8, 2011 at 12:05 PM, Prisciandaro, James J
<> wrote:
> Hello list,
> I am hoping to do some analyses using regression splines with survey data. Specifically, the data have sample weight and stratification (and possibly cluster) variables that need to be taken into account. I was wondering which regression spline programs will work with "svy" in stata; I've noticed most regression spline programs are user contributed, and there are probably much more than I am aware of, but I am particularly interested in multivariable regression spline models by Royston ("mvrs"). I'm also aware of mkspline and bspline.
> If it weren't for the complex sampling aspect, I would be using a free equivalent of the MARS program (e.g., Earth for R).
> Thanks in advance,
> Jim
> *
> *   For searches and help try:
> *
> *
> *

Stas Kolenikov, also found at
Small print: I use this email account for mailing lists only.

*   For searches and help try:

© Copyright 1996–2018 StataCorp LLC   |   Terms of use   |   Privacy   |   Contact us   |   Site index