Cause, correlation, and chemistry
SteveI am willing to be you have probably read, heard, or even repeated the following admonition in the last few weeks:
Correlation is not causation.
Here's why I think this assertion need to be retired, or at least pushed off to the side and filed away for a while with your Vanilla Ice CDs, He-Man and the Masters of the Universe figures, and Gloria Vanderbilt jeans.
You have the power.
First, on the inability of correlation, i.e. lining up two sets of data that seem to track in the same direction and making a claim that one event or activity 'causes' the other, well sure, I think everyone understands that trap.
After seeing everyone at the State Fair rocking their bad tattoos, and therefore thinking that getting a bad tattoo will make one attend the State Fair is not a conclusion most rational observers would reach.
But the problem with the 'Correlation is not Causation' admonition is that it has the effect or shutting down the debate and stifling the potential discovery of useful information. A strong correlation between two related and relevant data series may not imply or prove causation, but it probably implies something. And that something might just be really important for us to understand - say the correlation between the course of study our last dozen newly promoted employees took, or the relationship between managers that successfully completed the latest leadership development program and the 12-18 month success of their team members.
In HR and Talent Management I am not sure the goal should be to try and 'prove' any one thing can actually cause another thing to happen anyway. We are dealing with people, not robots, (not yet anyway), and attempting to make sense out of interpersonal relationships, motivations, rewards, and capabilities. It is, in many ways, much harder than sitting in a chemistry lab tracking how agents react to one another.
If I remember my high school chem lab accurately, loading up a beaker with a few solids and enough unstable acids and stopping up the top caused it to explode every time. People, while often predicable, are not always that consistent.
The last warning I will raise is that team 'Correlation is not Causation' like to use this conclusion as a fake scientific argument against any proposals or ideas that they disagree with, or did not come up with themselves.
Here's a simple example:
'Hey, I noticed last week when his car was in the shop, Jake got a tremendous amount of work accomplished working from home - maybe we should explore letting some of the other developers do some teleworking?'
Can we know for certain that working from home was the reason for the spike in Jake's productivity?
Of course not.
Might there have been a dozen other factors that might have been more responsible for the increase in output?
Sure.
Does your organization have the time or capacity to set up highly controlled experiments to try and figure it out - assuming that is even possible?
No way.
In science, proving causation might be the goal, the desired end state, but in Talent, we are much better served finding the correlations, using our understanding of work, people, and the world, and seizing on the ones that make sense for our business and our teams.
What do you think? Do you ever drop the 'Correlation' bomb around the office?
Reader Comments (6)
Such an interesting post. Correlation can actually move in the same direction as a positive correlation or in different directions...negative correlation.
I think correlation is just that it is interesting but as an analytic I need causation to understand what has the bigger impact on which variable. In talent this is never more important.
Take the engagement survey with 30 plus questions. Do you want to guess what impacts engagement by looking at correlation or do you want to be more confident by running regressions or perhaps structured equation modeling so that we can sustain or improve what drives engagement and quit what negatively ?impacts engagement.
Causation iis extremely important when an HR professional is trying to choose between several initiatives like wellness, HRIS or perhaps a new rewards and recognition program. At the end of the day you want to know which initiative will "cause" a certain outcome like an increase in productivity, engagement, or revenue.
I understand we are people and not robots but. Also have limited resources so we need to maximize what we have based on historical data and behaviors. We can't afford not to!
Thanks for the comments, Cathy. I suppose I come down just a bit more on the side of it being very difficult to 'prove' one behavior or condition will actually cause another one, and that the value the talent pro can bring to the table is making those judgment calls on what actions to pursue and what ones to let go. Would it be better to have more data, to process that data through more sophisticated systems and analyses to try and make more confident conclusions, and then determine courses of action? Sure, I think so. But I also think that we can over think and over analyze and spend lots of time trying to prove things that either can't really be proven, or don't add much additional value even if they are 'proven.' Maybe I am a bit of a contrarian on this these days, with Moneyball and Nate Silver being so popular - but to borrow a phrase from sports - 'You still need to watch the games.' Thanks again Cathy and have a great holiday.
Steve--thanks for the post. The issue with correlation is that it can be easily discounted by two key constituencies--front-line managers and senior leaders. We use structural equations modeling because you can infer some level of causality and it is just as straightforward as running correlations--plus it is a little more bullet-proof. The key is to make these analytics easy to access for front-line leaders so that they can take action quickly and in the right areas that have the biggest impact. Showing leaders at all levels how talent management processes drive real business outcomes (profits, revenues etc) will increase buy-in and get everyone aligned to the "critical few". Whether it's employee surveys or 360s or whatever, you can link them directly to business outcomes and show the relative impact and make those analytics/reports accessible and actionable for all leaders.
Thanks very much for your comments, Scott. I think I agree with your points, but just wonder a little bit if they really are about showing 'cause' or are just showing a better likelihood of causation. I guess I am splitting hairs a little, but I do agree that to get traction in most organizations, the more 'proof' the better.
Steve--absolutely. Without getting too 'geeky', we look to uncover 'relevant causal variables'. We know that there are numerous things that drive an outcome--what the analysis does is allow you to say with certainty that one area is part of what makes an outcome move--and by how much. This allows for an ROI calculation and prioritization of what you can work on. Where a correlation gets you in trouble is with the classic 'shark attacks and ice cream sales are correlated'. We both know that they both just go up in the summer time. The danger is when the ice cream company invests millions of dollars on fish bait to chum the coastline waters in order to drive up sales--a correlation analysis could be quite costly.
Steve--absolutely. Without getting too 'geeky', we look to uncover 'relevant causal variables'. We know that there are numerous things that drive an outcome--what the analysis does is allow you to say with certainty that one area is part of what makes an outcome move--and by how much. This allows for an ROI calculation and prioritization of what you can work on. Where a correlation gets you in trouble is with the classic 'shark attacks and ice cream sales are correlated'. We both know that they both just go up in the summer time. The danger is when the ice cream company invests millions of dollars on fish bait to chum the coastline waters in order to drive up sales--a correlation analysis could be quite costly.