[BlindRUG] qqplots, was Some Questions

Fri Jun 12 22:35:13 UTC 2015

Hi Tyler et al.,

This is a good discussion topic.
I do believe the qqplot() has its place; this is why it remains part of the output generated by UniDesc()

I think you're being a little harsh on the qqplot. Your comments re the usefulness are correct, but there are instance where the qqplot is an extremely useful tool, such as detecting significant effects in a 2^k factorial design with no replicates, or a partial fraction of one I plan to implement a VI() summarisation of the qqplot() for this reason alone.

Your comments re the hypothesis tests for normality are true, but they are true for practically all hypothesis tests and not just normality tests.

The key thing about the normality of a large data set is that it almost certainly means the central limit theorem kicks in and the need for normality is hugely reduced. In such circumstances, the need to test for normality is often redundant because we "know" what we should be expecting with enough forethought anyway. Our other methods of checking a distribution for the sample and therefore the population can be used instead of qqplot().

In cases where the sample size is small and the population's distribution is unable to be determined, the tests really should not be used as a scape goat for failing to have sufficient data in the first place.

Your comments re skewness and kurtosis are spot on, but the introductory student doesn't know about hypothesis testing when they first encounter the need to determine normality in their exploratory data analysis.

Finally, do be careful when using skewness and kurtosis as there are different measures for both notions. I've been very careful to explicitly state the particular test and measure for each in the UniDesc() output for this reason.

Jonathan

-----Original Message-----
From: BlindRUG [mailto:blindrug-bounces at nfbnet.org] On Behalf Of Ensor, Tyler via BlindRUG
Sent: Saturday, 13 June 2015 1:13 a.m.
To: Blind R Users Group
Cc: Ensor, Tyler
Subject: Re: [BlindRUG] Some Questions

I thought I would add my two cents on assessing normality. Maybe this will help someone. I can't see a Q-Q plot, and used to use hypothesis tests such as the Shapiro-Wilk or Kolmogorov-Smirnov. However, statisticians generally find these tests unhelpful, as they are too conservative with small sample sizes (i.e., very nonnormal data will still be deemed normal) and too liberal with large samples (i.e., very small deviations from normality will push the p value into the rejection region).

The approach I've since adopted is to look at the values of skewness and kurtosis. Skewness and kurtosis both have a standard error associated with them. If the skewness and kurtosis values are both within -2 or +2 standard errors of 0, then you can conclude that data are normal. This isn't a perfect approach, as it is somewhat arbitrary, but inspecting a Q-Q plot also lacks objectivity.

Tyler

On 6/12/15, Voica Gavrilut via BlindRUG <blindrug at nfbnet.org> wrote:

> Hello,

>

>>    1. Auto Load

>>    I want BrailleR package to be loaded automatically when starting R.

>> How can I do?

> For example I use the config file that already exists in the R

> installation - the R_HOME/etc/Rprofile. in this file I added a

> function to change my working directory and to load some of my useful

> scripts:

> .First <- function() {

> setwd("my_path")

> source("my_script1.r")

> #library("BrailleR")

> }

> Take care to the dot before the function name and the first letter

> capitalized, are part of the R config requirements.

>

> Good luck

> --

> G.Voica

>

> _______________________________________________

> BlindRUG mailing list

> BlindRUG at nfbnet.org

> http://nfbnet.org/mailman/listinfo/blindrug_nfbnet.org

> To unsubscribe, change your list options or get your account info for

> BlindRUG:

> http://nfbnet.org/mailman/options/blindrug_nfbnet.org/tyler.ensor%40mu

> n.ca

> The list archive can be viewed at:

> http://www.nfbnet.org/pipermail/blindrug_nfbnet.org

> More information and useful links about using R as a blind person can

> be obtained at:

> http://R-Resources.massey.ac.nz

>

> Look for help using R commands by reading the accessible e-book "Let's

> Use R Now" compiled by Jonathan Godfrey at:

> http://R-Resources.massey.ac.nz/lurn/front.html

>

_______________________________________________

BlindRUG mailing list

BlindRUG at nfbnet.org

http://nfbnet.org/mailman/listinfo/blindrug_nfbnet.org

To unsubscribe, change your list options or get your account info for BlindRUG:

http://nfbnet.org/mailman/options/blindrug_nfbnet.org/a.j.godfrey%40massey.ac.nz

The list archive can be viewed at:

http://www.nfbnet.org/pipermail/blindrug_nfbnet.org

More information and useful links about using R as a blind person can be obtained at:

http://R-Resources.massey.ac.nz

Look for help using R commands by reading the accessible e-book "Let's Use R Now" compiled by Jonathan Godfrey at:

http://R-Resources.massey.ac.nz/lurn/front.html
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://nfbnet.org/pipermail/blindrug_nfbnet.org/attachments/20150612/55a99670/attachment.html>