Normal Probabilty Plot

[KeithB]

I was all set to say that the normal probability plot in the Statistics 1Var app was wrong because the data is plotted on the X-Axis rather than the Y. But I did some research and found that folks do it both ways...

Anyway, is there any way to put a straight-line fit on the plot? Or at least have access to the Y values so we could create our own? And while I am asking, minitab plots some confidence limits around the straight line to make it even easier to judge normality.

[Tim Wessman]

There is not a way to put a line currently.

I remember implementing NPP way back and had similar thought as you expressed. It is amazing the variety of ways people will present stat calculations and vary completely between different regions as to what is "correct".

Statistics is not math though, so what can you expect!

[KeithB]

Thanks. DSP guys are the same way. "Of course adding is the same as averaging!"

[grsbanks]

"There are three kinds of lies: lies, damned lies and statistics."

(att: Benjamin Disraeli)

[KeithB]

If anyone wants to make a Normal Probability plot with a fit:
1. Put your sorted (lowest to highest) data into a column of the Statistics 2Var App.

2. Use the make function to fill another column with the formula:
NORMALD_ICDF((X-0.5)/N))

With X going from 1 to N (The number of data points)

3. Plot your data.

4. Hit the Fit button.

Note: As Tim pointed out, there seem to be many ways to make this plot. I have also seen a slightly more complicated formula to generate the probabilities which comes from NIST's engineering statistics handbook: http://www.itl.nist.gov/div898/handbook/...rmprpl.htm

To create this alternate form:
Use the make facility to create NORMALD_ICDF((X-0.3175)/(N+0.365)) from 1 to N. In the first row insert NORMALD_ICDF(1-0.5^(1/N))
In the last row put in NORMALD_ICDF(0.5^(1/N))

Proceed as above.

[mark4flies]

I prefer the data axis to be Y when this plot is used alongside another plot such as a histogram or a box and whiskers plot. (I teach statistics for JMP. I am a SAS Institute instructor.) It doesn't matter much to me in HP Prime because you only see one plot at a time as long as it it labeled.
I think that a reference line is helpful for the interpretation and use of the plot. When the data are the Y, then the line has a slope equal to the standard deviation and a Y intercept equal to the mean. This form is useful in a one-way ANOVA to have over-laid normal quantile plots for each group. If the lines are offset vertically then the groups have different means. If the lines are not parallel Jen they have different standard deviations.
It is a simple graphic that adds much to the usefulness of the visual display of the data.

[mark4flies]

Forgot to say that the confidence curves for normality are usually based on the Shapiro-Wilk-Lillifors limit that might be too much for a calculator app. I can't remember what the basis is for the limits plotted inMINITAB.