Modern robust statistical methods in Stata

[Eric Chameroy]

Hello everyone,

First, I would like to ask you for your patience as I am not only a greenhorn to Stata, but also to statistics in general, and hopefully in time, I'll become proficient with both.

I am interested in analyzing my data using robust methods such as those described in 'Introduction to Robust Estimation and Hypothesis Testing' (Wilcox 2012).

I am already aware of some of the trimming and winsorizing functions (e.g. trimmean, winsor, winsor2, etc.). However, are there any available functions to calculate ANOVA (or similar) using medians and/or trimmed means? Is there anyway I could use the just mentioned functions with the existing ANOVA function?

As for bootstrapping, is it possible to do the bootstrap-t method or percentile bootstrap? I know you can calculate a percentile-based CI, but that's not the same thing, right? I'm looking to use bootstrapping for comparing two or more groups.

Any feedback is greatly welcomed. Also, if you could recommend any references, that would be appreciated as well, especially if there is anything that caters to my background in biology/ecology.

Thanks everyone.

[Nick Cox]

http://www.amazon.com/Introduction-E.../dp/0123869838 is the particular book you cite.

Whatever led you to the commands (not functions) you cite -- all of which are user-written -- will unearth what's been done.

For example, Vincent Verardi has worked on various flavours of robust regression.

search

with keywords is how to find stuff made public.

I've not detected much interest in implementing many of the procedures suggested by Wilcox himself. He's written several overlapping books and many papers with original material, but their impact is hard to judge. Far from being "modern" his approach to data analysis seems very old-fashioned to me in several ways.

The main thrust within Stata, certainly from StataCorp, is really quite different, not to think of statistics in terms of lots of specific and ad hoc procedures, but to think in terms of broad families of models that may be appropriate for different kinds of data.

Bootstrapping in Stata seems rather well documented.

[Nick Cox]

That's Vincenzo Verardi, not Vincent.

. search verardi, author

Search of official help files, FAQs, Examples, SJs, and STBs

SJ-13-2 st0296 . . . . . . . . . . . . Semiparametric fixed-effects estimator
(help xtsemipar if installed) . . . . . . . . F. Libois and V. Verardi
Q2/13 SJ 13(2):329--336
describes Stata's implementation of Baltagi and Li's series
estimator of partially linear panel-data models with fixed
effects

SJ-12-4 st0278 . Robinson's square root of N consistent semipar. reg. estim.
(help semipar if installed) . . . . . . . . V. Verardi and N. Debarsy
Q4/12 SJ 12(4):726--735
presents Robinson's double residual semiparametric regression
estimator and Hardle and Mammen's specification test

SJ-12-2 st0252 . . . . . . . . . . A robust instrumental-variables estimator
(help robivreg if installed) . . . . . . R. Desbordes and V. Verardi
Q2/12 SJ 12(2):169--181
implements an instrumental-variables estimator robust
to outliers and allowing the usual identification and
overidentifying restrictions tests

SJ-12-2 st0259 . . . . . The S-estimator of multivariate location and scatter
(help smultiv if installed) . . . . . . . . V. Verardi and A. McCathie
Q2/12 SJ 12(2):299--307
provides the S-estimator of multivariate location and scatter

SJ-10-2 st0192 . . . . . . . . . . . Multivariate outlier detection in Stata
. . . . . . . . . . . . . . . . . . . . . . . V. Verardi and C. Dehon
(type findit mcd to obtain the mcd command) (no commands)
Q2/10 SJ 10(2):259--266
presents the minimum covariance determinant estimator

SJ-10-2 st0173_1 . . . . . . . . . . . . . . . . . . Software update for mcd
. . . . . . . . . . . . . . . . . . . . . . . V. Verardi and C. Croux
(mmregress, sregress, msregress, mregress, mcd if installed)
Q2/10 SJ 10(2):313
outlier option replaced by generate() option in mcd

SJ-9-3 st0173 . . . . . . . . . . . . . . . . . . Robust regression in Stata
. . . . . . . . . . . . . . . . . . . . . . . V. Verardi and C. Croux
(mmregress, sregress, msregress, mregress, mcd if installed)
Q3/09 SJ 9(3):439--453
provides alternatives to rreg and qreg for robust-to-outlier
regression; presents a graphical tool that recognizes the
type of detected outliers

[Dick Campbell]

Doing "ANOVA ... with medians" can be seen as a special case of quantile regression. Type help qreg for details and examples. Quantile regression is an example of the broad family of models to which Nick Cox refers.

[Eric Chameroy]

Thanks for the replies. I wasn't aware of quantile regression. After having looked into it, it may work for my situation.

Thanks again.