9 hours 3 minutes
Hi, guys. Welcome to improve phase design of experiments. I'm Catherine MacGyver, and today you'll have an awareness of design of experiments. So I want you to know about this tool because it's one that people frequently think of when they think about lean and six Sigma projects and improvements. However,
if you actually see a full factorial design of experiment done in your career, unless you work in research and development, consider yourself very lucky. So one of my business partners and then he actually kind of specialized in higher level
lean six Sigma in organizations and process improvement. And in the time that he and I were working on this
pushing towards the world class. So we're not talking
the choosen three signal levels. He and I actually didn't go in unless they had a signal level of higher than 4.5. We only did this twice in our entire career, so that being said, it's not something that you're going to see quite a bit of. But if you remember back to Dilbert at the beginning,
it's something that pops up is a popular
idea for what Lean and six Sigma people? D'oh!
So with that a design of experiment is a controlled study. So you're going tohave your control or your placebo or your baseline, and you're going to measure it. Then you're going to test each individual factor and measure the input of each factor or the influence of each factor. So
this is a statistical tool.
It is used to evaluate multiple factors on the dependent variable. Preferably this is going to be a mini tha one relationship. So where you have multiple independent factors and a single dent dependent, vector dependent factor, it can be done
with many too many. It is much more difficult to pull out
your statistical significance. So if you remember back to our hypothesis testing when you have a minute of many. So this is ideal in a minute, a one. So where you have, you know, four of your independent variables or your exes to one dependent variable, which is the output of your process.
Design of experiments is a branch of applied statistics, so this isn't something that is exclusively lean and six Sigma thing. You actually see this quite a bit in pharmaceutical testing because it also identifies interactions between different variables or different components. So, like I said, if you work in a research and development environment,
you're probably very familiar with seeing this
because we're going to test each item individually.
So when we're looking at a full factorial experiment, the first thing that you need to do is isolate each hypothesized factor independently, and you're going to want to document What are those independent variables? How did you isolate them? So how do you know that when you're testing this single variable,
it's not going to have any other factors with any of the other variable? So that's where one of the challenges
enough olfactory all comes in and why they're not really done all that frequently. So for the sake of this example, we're going to say that our X one independent variable is going to be the number of people working on a production line, and our X two independent variable is going to be the number of shifts worked
on that same thing. So for experiment number one, we're going to say that we have one employee
who works one shift, and that's going to be our no hypothesis. So if you remember back to hypothesis, development and hypothesis testing are null. Hypothesis says that if we change an independent variable, there is no change to the outcome.
So then if we say we're running experiment number two on we're going to say are variable is now we're gonna have to employees working one shift. We're gonna measure that outcome. Um, then let's say experiment number three, we're going to go back to one employee because that was our baseline working two shifts
and measure that outcome. And then for experiment number four, we're gonna say to employees working two shifts and measure our outcome.
So once we test each one of the assignments of these variables and for this module, I'm talking about positive and negative because buying Eri is easy to understand. When we're looking at pectoral experiments.
Then we're goingto look at all of the results from those experiments and identify you. Which combination is closest to our objective statement
from our project charter. So if we say that we want to increase our production by 25% as we look through the results, what are those? Which one of these gets us closest or there.
So from an organization standpoint, you're going to want to create a table that has your experiment numbers. So it's easy to keep track of which is which your independent variables your results. You're going to be here. Why or you're dependent variable. And then what are the assignments for each of those independent variables? In this case,
I just said binary, no variable invariable.
But if you were to say test with three different employees now you have three different assignments for X one are independent variable. So when we talk about full factorial experiments, they're not very common because they're very time consuming. They can be very expensive, depending on the variables that you want to test.
They can be very difficult to tease out
each of the hypothesized variables. However, one of the benefits is that you can be very certain about which one of the combinations of your independent variables will get you closest to your objective statement on gun. You can also start identifying,
um, interactions between the variable. So if you were to say, look at experiment number two, where we see two employees working together, but we don't see double the output
that would make us say OK, perhaps there's an interaction between those variables with each other.
If you were to ever see a design of experiments, the one thing that I want you to take away from this specific module is how to know if you're doing enough experiments. So that's kind of the kicker. In doing a deal, we is making sure that you are doing the right number of experiments.
I mean, of course, identifying your independent variables and how are you going to test them?
But a really easy way for you to know if your facilitator is doing the right number of experiments is your base number is the number of assignments for that variable? So in this case, we said by an eerie we said Yes, no variable, no variable, which means our base is going to be to your exponent
is the number of independent variables.
So with that, I would expect to see a table of eight experiments because we have a two to the third power for our number of experiments that we need to be done.
Full factorial are very powerful tools. When you get to use them on, they're able to identify which of your solutions is most effective towards your objectives thing that they can help you enter. Identify interactions between each of your variables
so they're very helpful when you're starting to get into
prioritizing your solutions. But
that's only if you get to see them. So with that, our next module is actually going to be about designing pilots on how, from a more practical standpoint, the solutions that you're identifying can be tested.
I'll see you guys there.