Normal distribution - Stata version 13.

[Thomas Rasmussen]

Hello I'm working with a dataset that include 69 pilots, it's a intervention project for my master degree.

I need to make sure if my data are normal distribution ore not.

I already tryed to draw my variable by using:

kdensity VARIABLE, normal
histogram VARIABLE, normal by(Randomizering)
qnorm VARIABLE

My issue is that some of my data do not look normal distribution, but I'm not shure how to test it statistically.
I tried to use:

sdtest VARIABLE, by(Randomizering)


But I'm not sure if it's the right method?

I hope someone can help.



Sincerely, Thomas.

[Bomar]

Hello may be you can try these two commands: swilk variablename (Shapiro-Wilk) or sktest variablename (Skewness/Kurtosis)

Regards.

[Maarten Buis]

You can add confidence intervals to your QQ-plot, which can help you assess how much deviation from Gaussiantiy you might expect just through sampling. This is done using quenv, which you can install by typing ssc install quenv in Stata.

[Scott Merryman]

That should be qenv

[Maarten Buis]

Thanks Scott, I should have known better

[Clyde Schechter]

Others have given you good advice about how to do this. Let me just prompt you to question why you are doing it in the first place. If you are contemplating a regression analysis, remember that it is the residuals that need to be normal, not the outcome variable. Also, even with respect to the residuals, the normality requirement is not a major obstacle to correct inference. If your sample is reasonably large and your distribution is reasonably symmetric, inferences based on OLS regression estimation will be pretty robust. In particular, the usual statistical tests for normality are sensitive to departures from normality that are too small to matter, for practical purposes, for the robustness of regression. It has sometimes been said of these tests that they are like sending a rowboat out into a storm to determine if it is safe for the Queen Mary to sail.

[Marcos Almeida]

Really, a thoughful and concise explanation, this one done by Clyde. I felt the same but could not find such a (clever) way to put the matter into words. The metaphor of the rowboat seems to me precise enough, worth being used (with reference and context, of course) in a lecture!

[Clyde Schechter]

To be clear: I did not originate the metaphor of the rowboat. It has appeared previously on Statalist (perhaps the last time was during the old listserve, I'm not sure.). Nick Cox, if I recall, has cited it often. I do not know if it originated with him.

[Nick Cox]

The original reference alluded to here I take to be

Box, G.E.P. 1953. Non-normality and tests on variances. Biometrika 40: 318-335.

but strictly (despite the promising title) Box was aiming at a different target, carrying out tests on variances before tests on means.

To make the preliminary test on variances is rather like putting to sea in a rowing boat to find out whether conditions are sufficiently calm for an ocean liner to leave port!

I agree with Clyde, however. Testing for normality is often unnecessary and irrelevant to the objective. Nevertheless looking carefully at marginal distributions often uncovers behaviour that should be borne in mind for the analysis.

[Marcos Almeida]

Indeed, it was clear to me. Now, even more clearer and contextualized.