How to change histogram bin centering and/or spec limit location?

Question

Hi JMPer Community,

(W10, enterprise, JMP Pro 15.0.0)

Does anyone know how to change the bin centering for histograms in JMP? That, or how to adjust the spec limit location in a process capability analysis histogram.

The image below is a histogram from a process capability analysis where we have an upper specification limit of 3 units. Our data does not have any values that are larger than 3.0. When the histogram is generated and is shown with the upper spec limit, it can appear that there are out of spec events (see purple bin below). The USL is shown on the graph exactly at a horizontal value of 3. The bin that contains the values of 3.0 for this measurement are all falling in the bin 3.0 to 3.1. The Nonconformance report correctly states that 0% of the observations fall above the USL, yet the customer might look at this graph and make an erroneous conclusion.

Is there a way to re-center the bins for histograms, so that the bin for the values of 3.0 was actually centered on 3.0 and not centered at 3.05? I looked through the Preferences and couldn't find anything. Also, no searches through the community or online have shown a similar concern from other users.

I can't change the USL or measurement values, as this would lead to incorrect statistics and calculations in the process capability analysis. It's really just a visual thing, but it might appear less of issue for the customer if the bin that is at a spec limit fell at the same location as the spec limit. Below is an example where I changed the USL to 3.05, just to show what I'm trying to describe (preferably, it would be centered at 3.0).

Seems like this could be a simple check box in preferences where one can choose to center the bin or not. Or, maybe it's more complicated than I think.

Thanks!,

DS

SDF1 · Answer

Hi @Dan_Obermiller,

I appreciate your suggestions and example. Unfortunately, JMP is not responding in the same fashion as with your example. For me, and perhaps this is because there are several thousands of data points that take on fractional values, JMP will not recenter the bars unless I dramatically change the binning size.

As an example, the below figures are bin sizes of: 0.2 (l), 0.05 (r), 0.02 (bl). For the 0.05 binning, you can barely see the bin that contains the counts at the USL of 3.0. For the binning of 0.02, you can't even see that bar anymore as it's completely covered by the USL line. At this point, I feel that it's deterring from the overall value of the analysis rather than helping.

Below, is an example of what I'd like to do using IGOR Pro software from Wavemetrics. In this software, it stops the bar graphs at the largest value of the distribution (red, shaded bar graph) rather than starting them at the largest value as JMP does. When generating the binning in IGOR, you have the option of choosing "bin-centered x-values", which generates the same distribution, just shifted as can be seen by the black bar graph. I have added in the USL as the blue line by hand. These graphs are static and can't easily be changed like in JMP.

I was hoping that JMP might have an option like this were it would always center the bars or left justify them as a preference setting. It could be that with the dynamic graphing capabilities of JMP it's a lot more difficult to do this kind of justification of the bars in a histogram. Maybe something for the future?

I do think it's a little strange that JMP graphs bars of a given value ABOVE the actual value. In my data, the values range from 0.0 to 3.0, so instead of having all bars graphed within this range, it actually defaults to graph such that the largest value bar starts at the largest value rather than ending there. Perhaps the difference is as you explain, that the right edge is NOT included in the binning with JMP, whereas it appears to be for IGOR.

There doesn't appear to be an easy solution to this at the moment.

Thanks,

DS

SDF1 · Answer

Hi @txnelson,

Thanks again for your suggestions. Definitely a possibility, but unfortunately, with the continuous data that I have, which can be tens of thousands of rows, it's not the most feasible solution, even with a script, as the continuous nature of the data could change from reporting period to period.  Appreciate the suggestions!

Thanks,
DS

Dan_Obermiller · Answer

The grabber tool does indeed change the centering of the histogram bars. In my silly example, there is a maximum of 40, which is where I set the spec limits. The original distribution from Process Capability looks like this:

This is the problem you are seeing with your data. To make this easier, look at the bar that starts at 20. The bar is 2.5 units wide and is centered at 21.25.

Using the grabber and dragging LEFT (not up or down) I get this picture. Notice that the bar that was at 20 now starts around 18 or so. The width is still 2.5 units wide and is centered near 20. You can see that this causes the histogram to now have something less than the LSL of 0. The USL "problem" is less pronounced, but still there a little bit.

Now I use the grabber and drag UP to change the width of the bars. I don't know how wide they are, but they are clearly much skinnier.

Just my opinion, but I think this is a much better picture to let customers know that you were right at the spec limit.

Notice that with the bar centered at 40 as it is in my example, there is a little bit of the bar over the USL due to the width. Setting the center to be less than 40 will not work because the right edge of the bars are NOT included in the bar, the left edge is. By default JMP uses the histogram widths as 40<=X<45 (or whatever the width is). This is a pretty standard approach to graphing data. So in order to include all of the data, you are getting a bar that is above your USL.

So just centering the bars will not resolve the problem that you see, but it can lessen it. The issue really becomes what happens at the right-edge of a bin. Right-edges are not included, so having a bin end at 40 will require another bar to start at 40. The bar starting at 40 includes the data.

Finally, I do not see is how to gain PRECISE control on the centering of the bars. The bin width can be controlled precisely in the Distribution platform.