Improvement Insights Blog
Automating Decision Trees to Collapse The Control Chart Learning Curve
Words like “variable” and “attribute” confuse new students of quality improvement. But they need to know that to choose the right control chart using a decision tree don’t they? Not if you automate chart selection for them. Here’s how I did it:
“I don’t know about the rest of you, but I got started back in 1990, and that’s when I learned about control charts. Then I started trying to teach people about control charts… guess what? You start talking about ‘variable’ and ‘attribute’ and ‘sample sizes’ and you can watch everybody’s eyes slowly glaze over; it’s just fascinating to watch.
“I was recently on a presentation at one of the big conferences and they were trying to do this and I was watching the chat bar because it was all [via] Zoom meetings. In the chat, people were saying, “This is where I get lost or confused” or whatever, right?
“When you’re trying to choose a control chart, the very first thing that freaks people out is this whole concept of ‘variable’ and ‘attribute’ and ‘counted’ and ‘measured’ and whatever it is…. Guess what? Back in 2006, I’d been trying to teach people that way and I kind of had a little decision tree built into the QI Macros. Because the QI Macros let you select data first, I realized since that was done then all you had to do was click a button and then I could analyze the data and see if it’s integers (which means it’s most likely counted or attribute) or it’s decimal (which means it’s always variable). You don’t have to know those two things, and so then it’s a question of sample size, right?
“Well, if you have one column of integers it’s probably a c chart. It could be an XmR chart, it could be days or something you measure, but most often small integers [is a c chart]. Or if you have [two columns of] integers and one column is smaller by a lot than the other column, it’s probably a p or u chart. I started thinking about this and I came up with the Control Chart Wizard: Just select data with your mouse, click a button, get a chart.
“So that next year in 2006… when I was training people, I used to try and teach them all about attribute and variable and how to choose sample sizes. I stopped doing that. I’d show them how to select data with the mouse, click a button, get a chart. Guess what? They could do it, and they very quickly became comfortable with the fact that they could draw control charts. Then we could start to teach them how to use a control chart. This was an experiment, but I found that it collapsed three or four hours out of trying to teach people about control charts because they could just get right to things, start doing their charts, start getting data. I knew I’d won when they pulled data down from their own server and started running control charts on their stuff; that was a big win.
“In the years since then, I’ve added things like the Chart Wizard because people would always ask me, “Jay, what do I do with my data?” Well, Chart Wizard will figure that out for you. Stat Wizard [answers] “What do I do with this data,” because who can remember all the rules about statistics? Statisticians can, but the rest of us… no not so much. Not if we don’t use it very often.
“Then I got the Improvement Project Wizard to start to figure out how to build entire improvement projects from scratch. In the control chart world, there’s one that will actually show step changes; I created a wizard to go look for trends and runs and say “There’s a process change.”
“I started to automate all the stuff that used to confuse people, and when I did that they could start making progress and getting results. That’s the goal. So do you really need to teach people everything there is to know about variable and attribute and sample size and everything else? Yes, but you can teach it after the fact. When you try and teach it before doing the chart it’s confusing. If you do it afterwards and they say, “Well, jay, why did it choose a c chart?” Then I can show them a decision tree and say “Well, it went here, went here, and did that… oh, okay.” Then the decision tree makes sense.
“So anyway, this is what I’m trying to think about: How do we automate more of the things that we do in Quality Improvement to make it easier for other people to learn so that we can transfer that knowledge more quickly? It really collapses the learning curve. I used to spend two or three days teaching this stuff and got it down to a day.
“So that’s my Improvement Insight for this week: Automate the things that we used to think you had to know beforehand, and do it all easily. Then you can teach them about it after the fact and they’ll learn it more quickly. Let’s create a hassle-free America; hassle free healthcare. Let’s go out and improve something this week.”