So given the sort of understanding of why people leave jobs. What organizations then want to do is go and start looking back. Okay, who's leaving? And where are they leaving from? And how can we use this to help us think about what we want to do about it? And so the first thing the organizations often end up looking is just kind of where they are leaving from. When the big questions is managers. Do we see the attrition as associated with different managers? And as we saw right in the beginning, spoiler alert. Yes. Kind of every organization that does this, the big insight they come back with this. Wow there's huge difference across our managers in terms of their rate of attrition. Now obviously there's going to be some sort of natural variation. I think Kade is talking with you about the fact that in any data set you're just going to see random variation. There's a point at which you don't want to read too much into it. If you look at it, you will see that there are some managers who are often real outliers, to the extent to which you think this is more than random variation. And that does tell you something about where to start dealing with your attrition problem. You also as I say, want to think about how we're hiring people. Are there people that we're hiring who are particularly likely to leave? And do we want that to inform our hiring processors to whether or not we want to look for slightly different attributes? Obviously, we want to understand all their functions, types of work, types of projects that are really leading to attrition. What can we do to lessen their demands? And so on. Is it our high performers, is it our low performers that leave? And there's also geography. Is it the fact that, for example, commute distances can be a big driver of time? While I'm on that subject, one of the interesting and challenging issues of people analytics is worrying about adverse impact. And so, suddenly within the US and I think many jurisdictions, there are important regulations about discrimination. In the US those regulations are not just about deliberate discrimination by race, ethnicity, gender and so on. But they're also about putting in place procedures that have the affect of disproportionately screening out certain groups. Also, as you think about do we want to favor people within a certain commuting distance. You want to be worried about what does this mean about kind of are we segregating it. Are there certain races that this is going to make life more difficult for? And indeed I would say throughout doing people analytics on a practical level. At a certain point you probably want to just chat with a lawyer to think about where might you get into trouble implementing some of these things, and where might you. That aside, there are still all sorts of exciting things you can do in predicting attrition. Now there are some organizations these days that look at social network behavior. And so, there's a belief when somebody suddenly starts updating their LinkedIn profile. That's often a sign that they're thinking about other opportunities. Is that a point at which you want to intervene? And so, yeah. All sorts of fun things you can do if you're creative to try and understand who's likely to leave and how we can intervene. In terms of doing this analysis. Again, there are a variety of different things that you can do. And at the lowest level is take on different groups and types of workers and calculate what percentage of them leave every year. These kind of standard attrition statistics that you see, percent turn over. Okay, and again, I'd definitely encourage you to do tests for statistical significance to understand which of these differences are meaningful and which are not. I think a big concern with these straight attrition percentages is they really don't take into account the fact that people are much more likely to leave early in their tenure. As you can certainly imagine certain groups, for some reason a couple of people leave. They're then going to fill these with new workers. Those new workers are themselves much more likely to leave. And so, it may end up looking like you have a big attrition problem in one group. Just because on average, workers tend to have been there less time. A better approach is actually to look at this on a cohort basis. So produce statistics saying okay, what proportion of people last three months, last six months, last a year, and so on. And compare those across groups and managers. And so on. Even better, kind of the same sort of approach that we talked about before with predicting performance is to think about what are all of the various different processes, or all the different predictors that might affect attrition. And throw them into a multivariate regression, where you're trying to predict things like who stays six months, who stays a year. So multivariate structure helps enable you start untangling those. Is it age? Is it education? Is it experience level? And so on, and see kind of what seems to have the strongest effect on who stays and who leaves. Even better than this, if you're kind of doing this professionally. Is to use a slightly different class of models called survival models to really help you understand what's going on. I just want to spend a minute or two talking about what those are. So survival analysis take the name I think in part from the fact that they're used heavily in epidemiology. Where your question is the worst case scenario. We assign a treatment to somebody. How long does it take them to die? And who survives? And for how long? So, what you see is something like the curve that I've got on the slide. Where over time, you want to understand what proportion of the sample stay at the firm. And you're probably going to see something like an exponential decay. Where so say, people earlier on the job decide whether they like it or not. They're more likely to leave early. And then the longer they stay, the more likely they are to remain there. So turnover drops over time. So what survival analysis do is they first try and fit this kind of time function. So we can see over kind a range of people who've been there for different amounts of time. Basically, what's the likelihood of their leaving? And how does that change? And then they say, okay. And now how the various factors shift that kind of curve. And so we could imagine some sort of variable, I don't know. Let's say, experience level at the point of hiring, which makes them either more, or less likely to stay. So in this case, if it reduces the general propensity to leave. What this does is it shifts the curve up with it. And so every point in time people more likely to still be at the firm. The nice thing about this kind of analysis is that you're not just looking three months, six months, one year chunk separately. We look at them all together. Account for that kind of time dependence of turn over. And then say, what are the characteristics of individuals, the groups they're in, the managers. All those things that we talked about before. Which of them ultimately make it more likely that people are going to leave at some point? And so what I've tried to emphasize over the course of this module is the different ways in which bringing data to the question could inform how we make some of the most important staffing decisions. That not only shape how people perform in their jobs, but shape their careers as well. Ever since there've been organizations, we've been doing this based on instinct, hunch, and so on. Now that we have so much data about what people are doing, strong analytical tools. There really is the opportunity to bring more rigor to the process. Make it more data driven. And certainly all of the early evidence suggested doing so improves the quality of decisions that we make. Helping us get the right people into the right jobs. And helping them stay there. In ways that again, are going to help the organizations. And also going to make for better careers for the people within them.
