Test for Normal Distribution

[tja26]

I have a data set and I want to know whether it is Normally distributed or not. How can I do this easily? I would prefer to be able to do it in Excel as I don't have any advanced statistical packages like SPSS.

Thanks

Tim

[Aaron Blood]

Look at my example...

http://www.xl-logic.com/xl_files/formulas/distrib.zip

[tja26]

Thanks for you post. I have had a look through the Excel but I'm not exactly sure how to test for a Normal distribution. Would you suggest that I plot my distribution and compare it to a Normal distribution shape and qualitatively decide if it's Normally distributed? Are there any statistical tests that give you a definite answer?

Thanks

Tim

[Aaron Blood]

I believe determining whether or not something fits a normal distribution is a qualitative assessment.

There might be some tests... but in the end it's still gonna be a judgement call. There is no definitive yes/no answer that I'm aware of.

But I'm no statistician... maybe there are some rules of thumb? I'm recollecting here from a grad-stats course I took like 10 years ago...

But keep in mind, before you even test your data for a distribution, you realize you're making an assumption that the data is in fact random! I believe what you really need to do is first analyze the data to confirm randomness THEN you can see if it fits a normal distribution. And yes, there are standard tests for randomness.

As an example... you can't just take the daily movement in a stock price and say "It's Normal" because that would assume first that the daily price is random (it's not, it's a random walk, and yes there's a big difference, the change in the price is more likely the random variable).

Make sure you understand what you're talking about statistics wise before jumping to a conclusion here...

[Wigi]

I can run Stata 9 on my pc. Has tests for normality:
from its website: tests available:
* Shapiro–Wilk
* Shapiro–Francia
* skewness and kurtosis test (D'Agostino, with and without Royston correction)

Don't know if it can be done with Excel.

Wigi

[Aaron Blood]

Just an update...

I did a quick Google on it and yeah, looks like there are tests...

Kolmogorov-Smirnov was one that I'm not familiar with. (Probably needed the next level course for that one.) A Shapiro-Wilk test is still not full proof. It gives evidence for certain types of non-normality but does not guarantee normality.

Again, I believe these tests all assume that you know what you're doing (ie. you've already done the tests for randomness).

Some Links:
http://www.statsdirect.com/help/para...ethods/swt.htm
http://www.le.ac.uk/biology/gat/virt...ats/normal.htm