Ok with a "small amount" of non-normality?
Hi John, Greg et al. Thank you very much for these insights! This will greatly help! Yan
Yan Boulanger, Postdoctoral Visiting Fellow Ressources Naturelles Canada, Canadian Forest Service Centre de Foresterie des Laurentides 1055, rue du P.E.P.S. C.P. 10380, succ. Sainte-Foy Qu?bec (Qu?bec) Canada G1V 4C7 Tel. : +1 418 649-6859
-----Original Message----- From: John Maindonald [mailto:john.maindonald at anu.edu.au] Sent: 4 mai 2013 19:31 To: Greg Snow Cc: Boulanger, Yan; r-sig-mixed-models at r-project.org Subject: Re: [R-sig-ME] Ok with a "small amount" of non-normality? One recourse I'd temporarily forgotten is to examine the sampling distributions of the parameter estimates that are given by mcmcsamp(). You can check these for approximate normality, and you can derive credible intervals that do not depend on normality assumptions (but they will depend somewhat on the mcmcsamp() choice of prior). Also, long-tailedness or kurtosis at a crucial level in the design may I think lead to inefficient estimates, even though the sampling distributions of parameter estimates appear close to normal. John Maindonald email: john.maindonald at anu.edu.au phone : +61 2 (6125)3473 fax : +61 2(6125)5549 Centre for Mathematics & Its Applications, Room 1194, John Dedman Mathematical Sciences Building (Building 27) Australian National University, Canberra ACT 0200. http://www.maths.anu.edu.au/~johnm
On 05/05/2013, at 2:32 AM, Greg Snow <538280 at gmail.com> wrote:
To expand on John's answer, I would agree that simulation is the way to go. With simulation you can see how much your lack of normality affects the results that you are actually interested in (generally the results of a test or confidence interval). This post: https://stat.ethz.ch/pipermail/r-sig-mixed-models/2009q1/001819.html shows some examples of using simulation with mixed effects models to determine if the p-values are behaving properly and also show one method of adjustment (using simulations) for still doing the tests when the formulas based on the normal assumption don't work. On Fri, May 3, 2013 at 12:36 PM, Boulanger, Yan < Yan.Boulanger at rncan-nrcan.gc.ca> wrote:
Hi folks,
This may be more of a "philosophical"- student question. In Zuur et al.
(2009). "Mixed effects models and extensions in ecology with R", it
is mentioned on page 20 that "[...] we can get away with a small
amount of non-normality"
I'm little bit puzzled when I face this kind of affirmation in a textbook.
What is really "a small amount"? Of course, it depends on your
"judgement"... In my case, I have level0 and level1 residuals that
are unskewed and that show a relatively modest kurtosis (unbiased) of
about 2.5
- 3.0. My models are based on several tens of thousands of
individuals and normality tests (e.g., shapiro.test) always fail for
residuals. QQ-plot show these rather long tails which correspond to
"some" outliers (considering my data, there are several hundreds of
"outliers" in this case). Homoscedaticity, when considering or not
random effects, is not violated so I wondered if I could rely on
these model's estimates considering the non-normality of the
residuals. My judgement in this case would be that the departure from
normality is not that high and this might not be a problem. But, as
an ecologist, not a statistician, I have hard time to convince myself on this... Any thoughts?
Thanks
Yan
[[alternative HTML version deleted]]
_______________________________________________ R-sig-mixed-models at r-project.org mailing list https://stat.ethz.ch/mailman/listinfo/r-sig-mixed-models
-- Gregory (Greg) L. Snow Ph.D. 538280 at gmail.com [[alternative HTML version deleted]]
_______________________________________________ R-sig-mixed-models at r-project.org mailing list https://stat.ethz.ch/mailman/listinfo/r-sig-mixed-models