Date: Wed, 24 Aug 2005 16:00:53 -0600
Reply-To: Nate Wojcik <nate.wojcik@gmail.com>
Sender: "SPSSX(r) Discussion" <SPSSX-L@LISTSERV.UGA.EDU>
From: Nate Wojcik <nate.wojcik@gmail.com>
Subject: Re: Nonparametric vs. Parametric
In-Reply-To: <S327809AbVHXVRF/20050824211710Z+236601@avas-mr06.fibertel.com.ar>
Content-Type: text/plain; charset=ISO-8859-1
well, i am not ignoring what you are trying to teach me, i am just
confused probably because we may misuse some of the statistical jargon
in our field of ecology. your input and insight are highly valuable.
i guess what i am trying to understand is how do i interpret a
significant result of a kolmogorov-smirnov test of normality? is this
not testing the sample residuals and errors? is this test important
in trying to understand the distribution of my residuals and/or
errors?
nate
On 8/24/05, Hector Maletta <hmaletta@fibertel.com.ar> wrote:
> Nate,
> You seem to have ignored my previous remark that the normality of your
> variable itself is not really that important. Parametric analysis applies
> whenever you may assume your sample is a random sample, and therefore the
> distribution of the differences between measures taken from many similar
> samples and the population measure is a normal distribution. Errors should
> be distributed normally, not the variables themselves. Otherwise, non
> normally distributed variables (such as income, for example) would be
> impossible tu analyze by means of samples.
> In the case of a linear model, you assume that a linear relationship exists
> in the population, but individuals in any sample may differ from it by an
> error term, and the normality of distribution of these errors or residuals
> is what you are concerned about, not the normality of the variable you are
> analyzing.
>
> Hector
>
> > -----Original Message-----
> > From: Nate Wojcik [mailto:nate.wojcik@gmail.com]
> > Sent: Wednesday, August 24, 2005 6:06 PM
> > To: Hector Maletta
> > Cc: SPSSX-L@listserv.uga.edu
> > Subject: Re: Nonparametric vs. Parametric
> >
> > hector,
> > i am looking at the distribution of the enitre sample
> > population, but it is still significantly non-normal.
> > although i have transformed this data, i am unable to reject
> > this null hypothesis. however, i have been able to reduce
> > the levels of kurtosis and skewness of the entire sample
> > populations to levels below 1.00 by log transforming the
> > data, which visually appears to be normally distributed. is
> > there a statistically ethical dilemma by continuing with a
> > parametric analysis if you know that you are failing to
> > satisfy the assumption of normality. my total sample size of
> > the sample population is 500 samples and contains very little
> > outlier samples.
> >
> > nate
> >
> > On 8/24/05, Hector Maletta <hmaletta@fibertel.com.ar> wrote:
> > > Without addressing the full scope of the questions, let me
> > recall once
> > > again that normality is not required of the variables
> > themselves, but
> > > of the residuals. A variable can have a skewed or pointed
> > or flattened
> > > distribution (relative to normal) and still be analyzable by
> > > parametric methods, provided the sample is a random sample and
> > > residuals are distributed normally. In fact what is theoretically
> > > needed is that the SAMPLING DISTRIBUTION is normal, i.e. that
> > > different results (e.g. different averages) obtained from
> > many similar
> > > samples show a normal distribution; the Law of Large Numbers states
> > > that under such conditions the average of all those
> > averages will tend
> > > to equal the population average if sample size is large.
> > This does not
> > > require all variables to have a normal distribution either at
> > > population or sample levels. However, an extremely skewed
> > distribution
> > > in a sample makes the resulting average more unstable (and
> > the standard error larger)because of outliers, but this is
> > quite a different problem.
> > >
> > > Hector
> > >
> > > > -----Original Message-----
> > > > From: SPSSX(r) Discussion [mailto:SPSSX-L@LISTSERV.UGA.EDU] On
> > > > Behalf Of Nate Wojcik
> > > > Sent: Wednesday, August 24, 2005 4:23 PM
> > > > To: SPSSX-L@LISTSERV.UGA.EDU
> > > > Subject: Nonparametric vs. Parametric
> > > >
> > > > i am currently analyzing data and i have come across a
> > few ideas
> > > > that i need clarification on. my design consists of a
> > single DV and
> > > > multiple IV's (with one IV containing greater than two
> > levels). my
> > > > hypotheses have been structured such that a multi-way ANOVA
> > > > (three-way) should be sufficient. however, i am unable
> > to satisfy
> > > > the assumption of a normally distributed population.
> > > > first of all, when determining the distribution and
> > variances for a
> > > > multi-way ANOVA, should the sample distribution be analyzed by
> > > > grouping factors?
> > > > second, since there are not any nonparametric equivalents for
> > > > designs with more than two IV's (to the best of my
> > knowledge, unless
> > > > a SRH extension works, but how?), how important is it to
> > meet all of
> > > > the assumptions of a three-way ANOVA.
> > > > furthermore, if i can transform my data (log and
> > > > log+constant) to allow both the skewness and the kurtosis to
> > > > approach zero, will this suffice a normally distributed
> > population.
> > > > are there any helpful cutoffs or limits?
> > > > finally, if i am working with pre- and post-burn treatments, can
> > > > these be viewed as repeated measures DVs? also, are there any
> > > > statistical techniques that allow you to compare the variance
> > > > explained with treatment response ratios (standardized data)?
> > > >
> > > > thanks
> > > >
> > > > nate
> > > >
> > > > __________ Informacisn de NOD32 1.1200 (20050823) __________
> > > >
> > > > Este mensaje ha sido analizado con NOD32 Antivirus System
> > > > http://www.nod32.com
> > > >
> > > >
> > >
> > >
> >
> > __________ Información de NOD32 1.1200 (20050823) __________
> >
> > Este mensaje ha sido analizado con NOD32 Antivirus System
> > http://www.nod32.com
> >
> >
>
>
| 
