Bear-y Interesting: Moving from Easy DOE to Custom Design with Gummy Bears (2025-US-PO-2570)

Welcome to our talk today entitled, Bear-y Interesting: Moving from Easy DOE to Custom Design with Gummy Bears. I'm Ryan Lekivetz, a director at JMP.

I'm Rory Lekivetz, a student at Northwoods Elementary School.

Okay, so what are we looking at here today? We were really interested in how do different factors affect gummy bears? Now, in particular, our experimenter, Rory here, had previous experience with Easy DOE. But she really wanted to try Custom Design because she heard that was kind of the harder version is Custom Design. But I'll say what we're going to actually see here is that the design was still easy, the analysis less so. And also some lessons about objective responses being hard for home experiments. Now I want you to tell us, how did you decide to do this experiment?

So last year at discovery, I took home a pack of gummy bears, and I put in hot water to see how it would change the texture. And so it actually made it a lot softer. So I wanted to make an experiment out of that.

Ah. So how about we tell them what were the factors that we picked in our experiment here?

So we picked two brands and the temperature, which was freezer, room temperature, and hot, the time two minutes and five minutes. So, how long you put it in temperature, the color, and because I thought that maybe the different pigments would make it softer or harder. And the air exposure, because overnight would make the gummies a lot harder. However, it was just like fresh, like from the bag.

Yeah. Or if we left it out. Yeah. I think you figured the temperature was going to make a difference going in, didn't you? Yeah. Oh, and I'll say actually, so the hot we had it under like a lamp. So we had a desk lamp that gets pretty warm. So it was enough to generate some heat. I think because we tried, did we try boiling water at first, and we just ended up with a melted gummy bear. Now for our responses. So what was dad's idea first? So I thought, because usually we have subjective responses, I thought, "Well, this is going to be great. Let's try an objective response." So here was dad's idea. So we had a nine-volt battery, we had like a little buzzer. So you can see in the picture there there's a buzzer as well as a button. And so the idea was going to be to see how much weight can we put on a gummy bear until that buzzer goes off.

I think, "What did we find when we were trying to do that?" It was really hard to get something to balance on top of a gummy bear, to be able to add enough weight for it. So, what we probably could have made it work, I think it would have been too time-consuming, but at least we have this set up for a future experiment, right where we can use that buzzer. So dad thought he was being so clever with dad's attempt, but that was a fail. So what did we end up looking at our responses?

So we ended up looking at the enjoyment scale from 1 to 10. The squishiness, which the larger value means harder, the chewiness, which the larger value means chewier.

Right. And I think that we have a baseline of seven. Is that right? So I think that you were kind of saying if it was fresh out of the bag, maybe that was kind of the seven. I think one of the other things I asked you going in, did you think that squishiness and chewiness were going to be related? So it kind of makes sense. And as statisticians, you know, that's where we might think of that as being correlated. So now let's talk about the DOE. I think you were able to find Custom Design, how about the factors in response? I think I didn't have to tell you anything. Did you just enter that? Now, how about the model? So I think because the model came through with just the main effects, is that what you wanted for your model?

So I added the interactions because I really wanted to see how the interactions will affect the gummy bears.

Okay. So yeah, I mean, so she was able to find in Custom Design, if you go to the model terms, you can say interaction second. So she actually discovered that without any kind of prompting. And then likewise, so what about the run size selection. So how did you decide on the number of runs?

So I decided on a number of runs because...

I think that was the default.

And well, it was 45 runs. That means that we have to test 45 gummy bears.

Did you think you could eat 45 gummy bears?

Yes.

Yes. In fact, we actually bought two 5-pound bags of gummy bears for this. Now I will say, so we didn't really talk about the design diagnostics, but of course, dad went in after. And I did look at the different design diagnostics to make sure. Okay. Did this seem like a reasonable design? And here, for instance, you can see the color map on correlations. And I'll say that looked pretty good for those main effects and two-factor interactions. And likewise, when I looked at the power analysis, you know, I could see at least for that, that signal-to-noise of two, like the default settings. It looked like as long as we had some big effects, we would have some reasonable power to be able to detect them in her experiment. Okay, so what did we do? I think we went into the kitchen. It took us a few hours, I think, to actually get through all those gummy bears.

But, you know, so Rory was able to actually create the table, create the JMP data table, and enter the responses directly in there. And now, if you've ever seen Easy DOE, the beauty of Easy DOE is that it takes you through step by step, all within the one workflow. But so what was the problem here? Custom Design created this data table you put in your responses, but now you actually didn't really know what to do. Now, part of that was though she hasn't done a lot of analysis and really any analysis outside of Easy DOE. So even ideas of table scripts and things like that were foreign to her. But I think we did start. So what did we look, so in Graph Builder, do you remember?

So we looked at the different responses to see.

The different responses relative to like those number of factors. And so I think she was able to actually quickly kind of grasp that idea of the, you know, being able to drag pieces into Graph Builder. But now I will say ultimately, I actually did bring her to load design and Easy DOE. So on Easy DOE, if you haven't seen it on that first screen, it's a little bit hidden, but in the factor table, there's actually an option there to load design. So to make this easier for Rory, I was able to load the design back into Easy DOE, so she was able to use the analysis there. Now I'll also say we ended up just looking at enjoyment. So this goes back to that question. What did we find about the squishiness and chewiness? Was there a relationship between them?

Yes. So if the chewiness had a high number and the squishiness would also have a high number. So I didn't really make sense to test both of them.

Yeah. And so I think here's a Graph Builder file. So what did we ultimately find. So, like with the squishiness or chewiness, what was the important thing there?

The important thing was the temperature. So that was the only factor that really mattered that much for the squishiness and chewiness.

Yeah. So ultimately, we ended up going. So Rory did Easy DOE. And so I think what's displayed on the bottom left there was for the enjoyment scale, what Easy DOE picked as the best model. But I think what the dad asked you, so I think, what did we find there? It was time. So time, temperature, and there was a really big interaction too. What was the big interaction?

Brand and temperature.

Right. And so that's why the brand was actually chosen there. Now I'll say so, dad actually asked you as well because I thought it'd be a more interesting analysis. And so in my own analysis, when I had went through some of the other modeling platforms, I had actually wanted to add color. And do you remember how you added color to your model?

So you just tap on the color one, and it will add it to the model.

Yeah. And so you can see here, so even the one color you know was marginally significant if we were to add it back in. And so in fact that's what I had asked Rory to do. And so you can see in the final model, you know, the PValue is 0.08. But I think it makes it more interesting if we were to take a look at having color in there as well. But yeah, so ultimately, what were our big things? The main ones are really temperature, time, and the brand and temperature. So I think what it looked like, so maybe the color and the color air exposure, is that right? But also one of the things I did ask Rory, I said, "If you were going to explain to one of your friends, so what really is that? What does that mean to have like a brand and temperature interaction?"

So a branding temperature interaction means that if it was one brand, then the temperature will be different depending on which brand it is. So per haribo the temperature didn't change that much. And for the other one albanese the freezer had a big difference, and it was a lot better for freezer.

Yeah. And so you can see here too, this is also where we see that apparently Rory has some certain preference for different colors as well. So that's why I thought it was interesting to try to add that into there. And so with that, you know, so some final thoughts, the Custom DOE was actually a much easier transition. So I think when we had originally planned on this experiment, I thought there was going to be much more of that story here, that transition from Easy DOE to Custom Design. So it was less so on the design side, but more on the analysis. But I think from my own perspective, maybe we'll have to try to throw in some constraints or other types of variables that you can't do an Easy DOE in the future. But so analysis really was that tougher part, but again, at least now we do have that option to load design in Easy DOE. And what was the other ultimate thing, what did you discover from after doing this experiment? How do you eat your gummy bears?

[inaudible 00:11:24]

Yeah. So since doing this experiment, she does indeed actually enjoy freezer gummies the most. And so with that, thank you for your time. Thank you for joining us here today. And leave us some comments as to what you might want to see for experiments next. I think you have some ideas in your own mind, but it would be curious to see what people think.