nav[aria-label="Primary Navigation"] { padding: 0; & ul { list-style: none; width: 100%; display: flex; flex-direction: row; justify-content: start; align-items: start; gap: 30px; padding: 0; & li { margin: 0; } & ul li { list-style: none; } } }

Quick way to generate Power Curves for Paired t test?

JerryFish

Need a quick turnaround answer:

Is there a quick way in JMP to generate a power curve, similar to the one below (generated in Minitab)?

I'm sure I could do it with Formulas and GraphBuilder (and JSL), but client is asking if we do it. (Almost sounds like a gotcha question.)

Thanks again for the help!

Accepted answers

All comments

Phil_Kay

Hi Jerry,

You can use DOE > Design Diagnostics > Sample Size and Power > Two Sample Means. Enter your alpha, std dev and sample size. Leave Difference to detect and Power empty. This gives you a plot of power vs difference to detect.

I don't think there is a quick way to generate the overlaid graph. But you can generate separate plots and then use the trick of right-click in the graph > Edit > Copy Frame Contents. And then Paste Frame Contents into the other plot.

Regards,
Phil

Phil_Kay

Their power curves look different. I assume this is because they are calculating for a one-sided t-test.

And just to say, I have had a lot of my time wasted by Minitab users. As you alluded to, a lot of them are just looking for any small reason not to go with JMP. JMP can be a threat to them if their expertise is based on Minitab. They don't want a tool that makes it faster or easier to discover useful insights from data.

Mark_Bailey

Phil is mistaken this time. (A rarity!) Because this case is a paired t-test, you use the sample size calculator for the one sample test of the mean. You are effectively testing if the paired differences (mean( Y2-Y1 ) versus 0) are different from zero. You are not comparing samples from two populations (mean( Y2 ) versus mean( Y1 )) .

MikeD_Anderson

Another note on Minitab - there will be a difference between the sample sizes, etc. between JMP and Minitab. There is a difference in the assumptions made during the power calculation so JMP tends to predict 1 or 2 more runs than Minitab. I can't find my notes on what the difference is specifically - maybe someone else can chime in. However, here are some great referecnces that I use when discussion Power and Sample Size and JMP vs Minitab.

Clay's awesome white paper on the topic: https://www.jmp.com/content/dam/jmp/documents/en/white-papers/wp-jmp-powersample-104887.pdf
A how-to for means from the ed team (note the use of crosshairs - used this when I taught the subject and GF): https://www.jmp.com/content/dam/jmp/documents/en/academic/learning-library/04-sample-size-and-power-for-testing-means.pdf
The complementary how-to from the ed team for proportions: https://www.jmp.com/content/dam/jmp/documents/en/academic/learning-library/04-sample-size-and-power-for-testing-proportions.pdf

The white paper on ditching Minitab: https://www.jmp.com/content/dam/jmp/documents/en/white-papers/wp_moving_minitab_to_jmp.pdf

Best,

M

MikeD_Anderson

Here's something I did a long time ago that might be helpful... Forgive the crudeness of it. I think I did it during my green belt certification to help me wrap my head around the F Sample Size Function and never came back around to tidy it up.

Sampling Calculation Figure.png

Sampling Calculation Figure.jmp

JerryFish

Thank you very much for your quick replies, Phil, Mark, and Mike. All replies were helpful!

I think the idea of using a one-sample mean is correct for this question. However, the power for a given difference and sample size is higher in the original Minitab plot than what JMP is giving me. It looks like it has to do with whether we are testing for mean <>0 (assumed by JMP), or mean >0 (as shown in the Minitab plot).

Is there a simple way to talk JMP into doing a mean > 0 test?

MikeD_Anderson

You might try the animation for the 1-sided test it gives you the options to do the one-sided tests (look at the bottom of the platform). It's also interactive!

If you want to win against Minitab you go after them on interactivity, graphics, and (if they are process people) the process capability platform.

Also on the sample size difference - see my previous reply. There is a fundamental difference in the assumptions they use that causes a difference between JMP and Minitab. I'll see if I can find the notes from my conversation with tech support on the topic.

JerryFish

Thanks for the tip on copy frame contents, Phil. One last question: When the original graph is generated, it contains a black line. I can't find a way to change the color of the line (or distinguish it in any other way). So when I past the frame contents to the 2nd graph, it is tough to distinguish without color coding. A legend would be even better.

Any suggestion on how to color code? Or how to export the points comprising the line to a data table, so I could pull it into Graph Builder?

Mark_Bailey

Right-click in the plot frame and select Customize?

JerryFish

Doggone it. Don't I look like the idiot? I had clicked on Customize, but then hadn't hit the Template button. (Guess it wasn't that obvious to me.)

Moving too fast and not paying attention.

Thanks Mark.

jiancao

Is there a simple way to talk JMP into doing a mean > 0 test?

JMP's Sample Size and Power for one-sample mean tests in DOE doesn't have the option to show the power curve for a one-sided test. But you can trick it by doubling the alpha value. For example, enter 0.10 instead of 0.05 to generate the power curve for one-sided test on the mean with std dev=2 and n=20.

6-2-2017 2-29-31 PM.png 6-2-2017 2-01-07 PM.png