topic Re: QQ-Plot double x-axis in Discussions

QQ-Plot double x-axis

juliaaline — Thu, 30 Aug 2018 16:47:49 GMT

Dear Community,

Please help, how do I change the ugly default QQ-Plot in jmp so that it does not have two x-axis and a histgoram but only a normal x- and y-axis?

Re: QQ-Plot double x-axis

cwillden — Thu, 30 Aug 2018 17:03:28 GMT

In Distribution? I don't ever have 2 x-axes. Here's what mine looks like when I turn off the histogram and outlier box plot. Is this what you want?

If that's what you're after, just go to the red-arrow > Histogram Options > Histogram (uncheck it), and then to red-arrow > Outlier Box Plot (uncheck it). You can also make that the default from Preferences for the Distribution platform, though I'm not sure why you'd use Distribution if you don't want to see the histogram.

Alternatively, if you just want the plots separately, add a normal fit to data. Then, in the fit report for the normal fit, click the red arrow > Diagnostic Plot. You'll get a QQ plot with just the usual X and Y axes.

Re: QQ-Plot double x-axis

juliaaline — Thu, 30 Aug 2018 17:17:34 GMT

Thank you for your quick response! Thats what my plot looks like, too but with the histogram. But I do not have it in the distribution area but as a residual plot from a mixed model. Is there an easy way to get a normal QQ-Plot? Because I have never seen one like that in a paper. Do you think it is still useable like that or do I need to add a normal distribution to my data and create the QQ-Plot myself? That would be awful because I have to do 30 of these and do not have a lot of time to do that.

Re: QQ-Plot double x-axis

Mark_Bailey — Thu, 30 Aug 2018 17:21:35 GMT

The placement of the two scales is due to the fact that you modified the orientation of the plot by selecting Stack or change the Histogram orientation option.

I don't think that there is a way to achieve the same result with the default Histogram orientation. If uses a special type of scale. Right-click on the axis and select Axis Settings. You will see that the Type = Normal Probability.

Re: QQ-Plot double x-axis

juliaaline — Thu, 30 Aug 2018 17:21:18 GMT

Thank you for your quick response. My plot looks like that too but with the histogram. I cannot uncheck it because I am in the fit model section of a mixed model. I need to create residual plots like that. Is it a normal QQ-Plot with the histogram with the two y-axis'? I need 30 of these for my theses and feel like they do not look like these I have seen in papers. But is there a more efficient way than creating the QQ-Plots myself by adding a normal distribution to my data as a feel like that will take hours I unfortunately do not have left.

Re: QQ-Plot double x-axis

juliaaline — Thu, 30 Aug 2018 17:23:37 GMT

I did not chose stack.. By fitting a mixed model I cannot stack or unstack unfortunatley.

Re: QQ-Plot double x-axis

cwillden — Thu, 30 Aug 2018 17:29:16 GMT

Are you using the Mixed personality in Fit Model? If so, that is a JMP Pro feature, which I do not have access to. If you don't like the look of the QQ plot in Fit Model, I think you could get into Distribution with out nearly as much effort as you're thinking. If you have all 30 of your model fits in the same Fit Model report, you could probably save all of your conditional residuals to the data table at once by holding Ctrl as you do it for any one of them. If you are able to get that far, the rest is easy. You would plot all of your conditional residuals in Distribution and hold Ctrl as hide the histograms and box plots, or fit a normal distribution and turn on the diagnostic plot. Holding Ctrl is a powerful tool to easily do a repetitive task that works just about anywhere in JMP.

Re: QQ-Plot double x-axis

Jeff_Perkinson — Sat, 01 Sep 2018 07:49:34 GMT

If you don't want the histogram, you can use the Selection tool (the fat plus) by pushing the S key on the keyboard and then click in the scatterplot to select it and shift+ctrl click on the X axis to select it.

It will look like this:

Then Journal it and you'll get one that looks like this:

Re: QQ-Plot double x-axis

juliaaline — Thu, 30 Aug 2018 19:11:43 GMT

Thanks so much that looks great! But isn't it that on the x-axis hast to be the quantiles of the residuals or did I get there something wrong?

Re: QQ-Plot double x-axis

gzmorgan0 — Fri, 31 Aug 2018 06:03:12 GMT

When it is important to have really good looking graphs, like publications or training material, sometimes I like to save the residuals then do one of the following:

Distribution > Fit Continuous > Normal > Normal Quantile > Diagnostic Plot
Create a Character column named "id" where the constant value is the name of the model. Then use Fit Y by X where the residual is Y and id is X. Turn off show all graphs and select Normal Quantile and you can choose whether you want to fit Quantile by Actual or Actual by Quantile.

Below are the graphs resulting from the Scripting Index > Objects > Fit Mixed > Run Example > Save Residuals

Distribution, Fit Normal, Diagnostic PlotFit Y by X, Residual by id

Re: QQ-Plot double x-axis

juliaaline — Fri, 31 Aug 2018 07:38:29 GMT

Thank you so much that looks great and works fine with my jmp version! Just one last question, aren't normally the quantiles of the residual on the x-Axis and not the actual residuals?

Re: QQ-Plot double x-axis

gzmorgan0 — Sat, 01 Sep 2018 05:52:07 GMT

Julia,

You are correct that most Q-Q plots will use the sample data quantiles on the Y axis and the theoretical distribution quantiles on X-axis. See https://en.wikipedia.org/wiki/Q%E2%80%93Q_plot Also a Q-Q plot typically uses the same scale, since the quantiles of the sample data is plotted against the quantiles of the theoretical, hence, its name.

JMP plots are called probability plots, and the sample (Actual) data scale is used for one axis. The reason, I documented using Fit Y by X is because it allows you to to create an Actual by Quantile or a Quantile by Actual.

If you want a Q-Q plot as documented by the url above, you need to compute the standardized data values, or compute the sample quantile (0 to 1, sort the data and compute the cdf ) and use that data in Fit Y by X, and plot Actual by Quantile and then they will have the same scale.

However, I like the probability plot, because typically the consumer of the reported plot typically relates to actual sample data scale. Also, the key to any probability or Q-Q plot is to asses whether the data appoximates a linear relationship.

Because I first learned to use probability paper in physics prior to the computer age (yes in the late 60's we still used slide rules), plotting Quantile by Actual is my preference. However, I recommend you plot the data in a format that is expected or typical. If there is no "norm" or expectation then just add a blurb to your paper or report, describing the plot.

Hope that helps!

Re: QQ-Plot double x-axis

juliaaline — Sat, 01 Sep 2018 10:12:15 GMT

gzmorgan0,

thanks so much for your explanation, I now decided to take the normal probability plots. Just to make sure, on the y-axis there is the normal probability which is at the same time the normal quantiles, right? Also they are general normal quantiles and not in any way modified to fit my data? On the x-axis I put the residuals of a mixed model, so it would not be correct to call the plot a QQ-Plot but only a normal probability plot? Or are the terms not so strict and I could calll that a QQ-Plot, too? Because I feel like my supervisor called that a QQ-plot..

Thanks so much and best wishes,

Julia

Re: QQ-Plot double x-axis

gzmorgan0 — Sat, 01 Sep 2018 10:43:46 GMT

Yes, the Y-axis probability and quantiles are those of a normal.

Some people call these JMP plots quantile plots and they serve the same purpose as a Q-Q plot, to check if the distribution of residuals are approximately normal.

You should also plot residuals vs. predicted and residuals vs. x. The residuals should be randomly distributed for all levels of X and predicted.

Regards.