(Putting this back to the R-SIG mailing list, where it originates and belongs)
michael westphal <mi_westphal at yahoo.com>
on Wed, 17 Jun 2015 21:25:32 +0000 writes:
> Thanks for the advice. ?Since I have panel data, does it make sense to explore fixed effects or random effects models? Robust regression will not address the changes in countries over time. ?It is easy to compare fixed effects and random effects with each other, but how do I compare their outputs with the outputs of the robust regression? In this case, I would recommend using the robustlmm CRAN package ( http://cran.r-project.org/package=robustlmm ) of Manuel Koller (in CC) who has completed his Ph.D. thesis on the topic of robust mixed effects modeling. Best, Martin > Cheers, > Michael
> On Friday, June 12, 2015 10:48 AM, Rand R Wilcox <rwilcox at usc.edu> wrote:
> You might also look at the Elsevier book? Robust Estimation and Hypothesis Testing. It provides some additional options that might help.
> A couple of quick suggestions. Even when using a robust estimator with a high breakdown point, check on what happens when leverage points (outliers among the independent variables) are eliminated. Second, consider a method that allows a heteroscedastic error term.
> Third, take a look at robust smoothers. Illustrations on how to do this are in the book. My experience when working with various research teams is that they can be very important.
> Hope this helps.
> Rand
> Rand Wilcox
> Professor
> University of Southern California
> 3620 McClintock Ave
> Los Angeles, CA? 90089-1061
> For information about Understanding and Applying Basic Statistical Methods Using R, Wiley (in preparation), and other recent books, go to
> Dornsife.usc.edu/cf/labs/wilcox/wilcox-faculty-display.cfm
> and click on books.
> Or go to https://www.amazon.com/author/randwilcox
> ?
> ________________________________________
> From: R-SIG-Robust <r-sig-robust-bounces at r-project.org> on behalf of Martin Maechler <maechler at stat.math.ethz.ch>
> Sent: Friday, June 12, 2015 2:53 AM
> To: michael westphal
> Cc: r-sig-robust at r-project.org
> Subject: Re: [RsR] robust regression and fixed effects models
michael westphal via R-SIG-Robust <r-sig-robust at r-project.org>
>>>>>> ? ? on Wed, 10 Jun 2015 14:17:57 +0000 writes:
> ? ? > Hello:
> ? ? > I am using R 3.0.2.
> so you really should upgrade {unless you meant 3.2.0}... at
> least in a few days when? R 3.2.1 is released.
> ? ? > I have panel data on countries' renewable energy net generation (and installed capacity) over time.? I am regressing these dependent variables on various socioeconomic variables, as well as binary policy variables.? I have have done basic OLS, but I wanted to explore both fixed effects models, as there are likely significant country effects and robust regression, as Q-Q plots indicate that there are some strong outliers.? This might be a question of apples and oranges, but how do I compare the goodness of fit of the fixed effects models with the robust regression models?
> ? ? > Any help would be appreciated.
> Package? robustbase? which has function? lmrob()? with many good
> and modern options for robust regression
> *also* has a 'Suggests: fit.models' in its own description file,
> because the package 'fit.models' with its function fit.models()
> tries to take fits of basically the same model and
> produce "comparison output" from that.
> It's quite useful in situations like yours,
> and I plan to add an example of its use to the 'robustbase'
> package documentation.
> ? ? > [[alternative HTML version deleted]]
> ? ((Because you used "HTML" aka "rich text" / "formatted text"
> ? ? instead of simple plain text, your message ends up looking so
> ? ? messy as above ...))
> Martin Maechler,
> ETH Zurich
> _______________________________________________
> R-SIG-Robust at r-project.org mailing list
> https://stat.ethz.ch/mailman/listinfo/r-sig-robust