RobertNiles.com
About Robert
Reporting Help
Finding Data on the Internet
Questions for Candidates
Stats Lessons
Mean
Median
Percent
Per capita
Standard Deviation
Margin of Error
Data Analysis
Sample Sizes
Stats Tests
Help Board
Bookstore


A Mathematician Reads the Newspaper

Read about reporters who blew it when trying to write about stats. A mathematician leads you through the tricks, citing contemporary examples.
More information
-->

Home > Statistics Every Writer Should Know > The Stats Board > Discusssion

normal distribution and correction
Message posted by Ruth (via 194.82.103.73) on October 25, 2001 at 10:34 AM (ET)

I have to check if some data in SPSS is normally distributed and then apply transformations if not. I am not really sure how to do this . Can anyone help?


READERS RESPOND:
(In chronological order. Most recent at the bottom.)

Re: normal distribution and correction
Message posted by Tomi (via 154.32.143.90) on October 27, 2001 at 10:23 PM (ET)

I'm not an SPSS user, but it must have at least one of the following tests:

Kolmogorov-Smirnov.
Shapiro-Wilks.
Normal plot (gives straight line if normal).

Of course an experienced eye can ususally tell from a simple histogram.

A dead giveaway is if the distribution is not symmetrical, and this is where transformations can help. If the distribution has a long tail on the right, a logarithm, square root, cube root, fourth or higher root may be appropriate. If it has a long tail on the left, an exponential, square, cube, fourth or higher power may be appropriate.

Often it's trial and error - whichever of these transformations yields a distribution that your software accepts as normal. If there is a choice, go for the simplest transformation (lowest power).

By the way, if your original distribution is bimodal then you might as well give up now.


Re: normal distribution and correction
Message posted by Lena Kupriyanova (via 129.96.253.100) on November 6, 2001 at 1:08 AM (ET)

>>By the way, if your original distribution is bimodal then you might as well give up now.

Please, please, let me know what you mean by that! I need to know whether my original distribution is bimodal or not. Does it mean that if I transform the original data, bimodality will be more than obvious? Thanks!


Re: normal distribution and correction
Message posted by Tomi (via 213.133.214.2) on November 6, 2001 at 7:37 AM (ET)

If you have a bimodal distribution, it is not possible to transform it to a normal distribution.


Re: normal distribution and correction
Message posted by Lena Kupriyanova (via 129.96.253.100) on November 7, 2001 at 1:31 AM (ET)

Dear Tomi,

I understood this. Bimodal distribution => impossible to normalize. I am curious if it works all the way around, something like impossible to normalize => bimodal distribution (I doubt that). I do not need to make the distribution normal. I just need to test whether the distribution I have is bimodal or not.
Thanks a lot for your help!


Re: normal distribution and correction
Message posted by Tomi (via 154.32.143.43) on November 7, 2001 at 4:41 PM (ET)

I think I may have said this already, but here goes: As far as I am aware, you can only identify a bimodal distribution by looking at the histogram and seeing if two distinct peaks are visible.


Re: normal distribution and correction
Message posted by Tomi (via 154.32.143.43) on November 7, 2001 at 4:44 PM (ET)

Actually there is another way.

You can plot a cumulative frequency graph of your data. The typical shape is "shallow slope increasing to steep slope and getting shallow again and finally becoming horizontal" or more simply "shallow, steep, shallow, horizontal."

For a bimodal distribution the shape is "shallow, steep, shallow, steep, shallow, horizontal."



Your $5 contribution helps cover part the $500 annual cost of keeping this site online.

Niles Online Sites:RobertNiles.comTheme Park InsiderViolinist.com

RobertNiles.com™, the site, content and services 咀opyright 1996-2002, Robert Niles.
All rights reserved. Questions? Comments? Read my Privacy Policy, or E-mail me!