We continue our discussion with Design of Experiments part 5 in QuikSigma. In this video, we learn how to extract more useful information from a designed experiment at very little additional cost using multiple output variables. Check out the beginning of our series here.
A lot of the expense and difficulty of doing an experiment is setting up the input variables and it’s a great economy if you can have more than one output variable of interest measured against the same set of input variables. So here’s a sample data set. This is a hypothetical example out of building seat belts. The question is you’ve got two output variables. One’s engagement distance, and that’s how much seat belt the mechanism pays out before it locks up in a wreck, and the other is Paul insertion. The way that you lock it up, there’s a little gear mechanism, and you just jam a Paul in between the teeth. There’s a there’s a maximum spec for the amount of webbing that you could pay out and there’s a minimum spec for how for that Paul gets into the gear teeth within a few milliseconds.
Here are four input variables that might influence that, and this is the Paul to gap and the pivot lever height, surface finish, and I forgot what RLPC stands for. But I can just simply designate two different or more than two different output variables and I can then run both of them simultaneously against the same set input of variables. Now, let me add X here and one more here and then I’ll hit calculate. Here are my results.
Apparently, my model for Paul insertion is pretty good. You see the needles and sliders and I’m looking at PI here. Now, to switch over to ED, I then just use this drop-down box and everything changes across here except in the optimizer, I will simultaneously get both of my output variables. So, that’s very nice. Having gone to the trouble of building the model, I can then make any adjustments that I want here and watch simultaneously what it does to my to output variables.