Counting Candy Debrief

Afiesha McMahon, IHI
David Williams, PhD

Afiesha McMahon: So we just play the candy count game, Dave. What can people learn about variation through this game?

David Williams, PhD: Great. So this game is really interesting because it makes a major point pretty simplistically. So when you're looking at the run chart, after you've called out all your M&M's, you see these dots that are varying. There may be 12 or 13 or 14 going up and down. And they seem pretty random. Right? As you're looking at them, they don't look like they were in any kind of pattern. But they're all hovering around a same range of numbers. So this is random variation. It's common cause variation. And it basically comes from M&M's being packaged by weight versus by count. And so there's naturally going to be a varying number of M&M's in each packet, varying by a couple, two or three M&M's each time. And that's common cost variation.

David Williams, PhD: Now, if it worked, you'll also see a couple of dots that are outside of that range. Right? These are outliers or special cause variation. And that's because your facilitator slipped in a different pack of M&M's. So maybe most of you were measuring the chocolate M&M's that have a certain amount. And then you had a snuck in pack of the peanut M&M's. And those are packed by weight as well, but they are bigger and heavier. So there's fewer of them. And that's a special cause. It's a different process or a different system that was identified because the data was out outside of the common cause variation that you saw.

Afiesha McMahon: So how did the lessons from the candy count game apply to real work?

David Williams, PhD: All right. So when you're starting an improvement project, or you're trying to improve something in your workplace, one of the things that we encourage you to try to do is to start to track data and to look at data. So you can have a sense of what is the normal variation of the process you're trying to improve. And this is important to help you figure out what to do. And so you want to start with a process that's stable, and that has common cause variation. If there are special causes, we want to try to figure out what those are and identify them, or remove them from the process, before we begin working on the process itself.

David Williams, PhD: One example that comes to mind, I remember a few years ago, I was working in England with pharmacists that were trying to figure out how to improve the process of doing medication reconciliation. And as we were looking at the data and we were sitting together, we realized that the data going along was had some normal variation from day to day. But then there were these independent spikes that were different.

David Williams, PhD: And what we discovered when we brought people who were knowledgeable about the process and looked at the data, is that the spikes were actually on the weekends where when we asked, "What's different about the weekends," we found there's a different process on the weekends. And so for us, it helped us to figure out that we needed to stratify our data and to think differently. Because during the week, there was one process that produced a very predictable level of performance. And then on the weekends, there was a different process that was producing a different level of performance. And so by looking at the data over time, we could differentiate the main process that we were trying to fix from this other process. And we removed that data, or the weekends, from that view so that we could focus in on the weekday process.

David Williams, PhD: So very similar, when we're looking at the M&M's, if you imagine, during the week we had the system that produced chocolate M&M's, and on the weekend we had the system that produced peanut M&M's. So in order to work on this process, we need to recognize we've got two different systems going on. And we may need to break those out and measure them separately so that we can understand how do we make this process work during the week, and how do we make this process work on the weekends. And see the improvement in both cases?

David Williams, PhD: One of the challenges that we find is that a lot of people will start doing improvement without thinking about measurement and without measuring the process that they're trying to improve. And so in many ways, it's really hard to differentiate what's common cause and normal random variation within a process. And what's special. And where this becomes critical is you don't want to act on events as if they are common cause when they're actually unique special cause event. And you don't want to act on common cause events as if they were special. So being able to see your data over time and differentiate what's normal random part of the process and what's an attributable special cause is important for you to then be able to figure out how to act on that improvement.