Harvardx course Practical Improvement Science in Health Care with the Institute for Healthcare Improvement.
So as we’re getting enough data, we have the ability to start using more sophisticated tools. And when we get 12 data points in a run chart, we can begin to use four rules to help us differentiate between common cause variation and special cause variation. We can actually look at and try to differentiate between what’s random and is just normal, and what is a non-random signal of change. So I’m going to walk you through these four rules. And you’ll see examples about how they look, so you kind of get a sense of how to apply them to your data. All of these rely on us having that median, which is something you put into your line chart after you have 12 data points.
So, the first rule is something we call a shift. This one is where you identify six or more data points in a row, above or below that median. So a couple things to think about though when you’re looking at it, so it sounds simple but every once in a while, there’ll be a case where you’ll see a data point falls on the median. If that happens, don’t include that one in the count, also don’t interrupt it. But any six data points in a row, above or below the median, equals a shift. And this would be a signal that something nonrandom has occurred in your process.
So the second rule is something we call a trend. Now I want to explain what a trend is, because trend is a word that we hear quite often in the news. Or people talk about it when they’re in meetings. And we actually have an operational definition for what a trend is. So in order to have a trend on a run chart, we’re looking for five or more data points either going up or going down. And we need exactly five or more in order for us to call it a trend. Now there’s one thing, kind of like when we’re talking about rule number one, you want to take into consideration. If two data points are the same, they fall on the same number, the same line, we just count those as one. We don’t let that interrupt our trend. And we don’t count both dots. But if we have one, two, three, three, four, five, that counts as a trend. It doesn’t interrupt it.
So rule number three, this is too many or too few runs on your run chart. So data should predictably bounce above and below the median. And depending on how many data points you have, you can use a table to determine how often is normal. To do this, you need to know that a run is one or more consecutive data points on one side of the median. And so now you could either circle the number of runs you have, or you can count the number of times your data crosses the median and add one to it. Now, go to the table and see what the range of runs should be for the amount of data in your run chart.
And then the last rule is rule number four, which is an astronomical point. This is where you see a data point that’s far out from the normal process that you’re looking at. And it stands out to you. It’s a point that’s likely something that happened. It was a problem in your data collection. It was a very recognizable attributable event. And it just doesn’t match the rest of the data that you see. If any of these signals, any of these rules, trigger within your run chart — are present within your run chart —something’s happening. There’s an attributable cause. And that’s something that we want to understand more about. So hopefully it’s a result of your improvement work. But if not, it’s something we want to try to learn about and figure out how to avoid.
If this was helpful, share and include me @DaveWilliamsATX. Sign up here to receive a monthly email from me that includes all my blog posts and other Improvement Science resources I think you’d appreciate.