In workforce analytics just pretend that correlation is the same as causality

06/01/2014 at 22:19 8 comments

causality

If you read any article, book or blog post about workforce analytics, HR data or evidence based HR you will eventually read the following sentence “…but remember that correlation is not the same as causality”. And this is factually correct. Just because two items correlate does not mean that they are connected or that there is causality.

The problem is that it does not help you with your communication with top management or your recommendations for action. Take this example;

Imagine that you have worked really hard with your analytics project, and you have build a model which shows connections between various training programs and profits. You identify that one particular training program correlate very strongly with profits. You then produce your fine PowerPoint slides and present your findings to the management board. You say “I have identified this strong correlation between this particular program and bottom line profits BUT as you all know, correlation does not mean causality”. The managers look at each other and one say “So what do you say, should we roll this program out to all of our sales people or not?”. You look confused down in your papers and then respond “Well, I can’t actually say for sure that attending this training program will make your sales people produce higher profits or if it is the sales people who are very profitable who happens to attend this particular training program for unknown reasons. The only thing I can say is that the two variables appear to correlate”. The meeting ends and you wonder why your next analytics project don’t get funded.

You see the problem? On one hand you are right. You cannot for sure say which way the arrow turns between the two variables. But on the other hand the management board does not care about correlations (nor should they), they only care about solid and value-added recommendations.

What to do? My advise is; pretend that correlation is the same as causality. Test your hypothesis with qualitative data and present a recommendation based on your findings. So in the above example, interview five sales managers and ask them about their view on the causality; do they believe that the particular training program has led to significant increase in profits from the sales people attending the program? If it has they that’s it. I know it is not ‘correct’ or ‘true’ but it is good enough. And most likely you will be right.

 I know there are a lot of fancy and even simple ways to test the hypothesis with time-lagging-tests or blind-tests or something similar. I have just experience a lot of places where even that kind of tests are not possible, too expensive or not available due to poor data. Make it easy. Do a qualitative test and present your findings as if correlation is the same as causality

 (just don’t tell anybody I told you so – this approach is not looked well upon in analytics circles. Leaders don’t mind – trust me)

Entry filed under: Analytics. Tags: , , , .

Workforce Analytics can add value in two ways (only) How you create a Superhero analytics team

8 Comments Add your own

  • 1. Frank DiBernardino  |  07/01/2014 at 02:56

    Morten,

    Loved this post and couldn’t agree more. Too many academic types get their underwear in knots trying to prove causality. While credible hard data is always nice to have, qualitative insights add valuable context and credible insights. While the business world is moving at light speed, the “causality” activists move at a snail’s pace.

    When I was in the Army we had a saying “the quick and the dead.” I’m still walking.

    Thanks for sharing the post. Hope that our paths cross.

    All the best in 2014,

    Frank
    Frank J. DiBernardino
    Managing Principal
    Vienna Human Capital Advisors
    (C) 610.505.5674

    http://www.viennaindex.com

    Reply
  • […] If you read any article, book or blog post about workforce analytics, HR data or evidence based HR you will eventually read the following sentence "…but remember that correlation is not the same …  […]

    Reply
  • 3. David Wilson  |  22/01/2014 at 10:41

    Morten. Think we need to see a lot more of this discussion in the HR function where there still seems to be an impression than analytics conclusions are the dark arts that must either be bullet proof to be useful on one hand or never be questioned on the other.

    Outside the analysts,the conflict between correlation and causality can go either way – do “best practice” companies invest more because they perform better or the other way around. Deconstructing this is important but realistically often needs a balance of probabilities to determine whether there is underlying causality.

    Thanks and continue to find your content interesing.
    DAVID

    Reply
  • 4. Joe Frank, Ph.D.  |  22/01/2014 at 18:37

    This makes sense in the corporate environment, but when the senior leadership to whom you speak are themselves academics and/or physician-scientists, one must have more charts, R-squared, coefficients, etc. up your sleeve in order to be ready for the inevitable questions and challenges. However, sometimes all that is necessary is a good graphical illustration, like a bar chart or a regression line showing the relationship, while acknowledging that outliers may exist.
    Thanks,
    Joe Frank, PhD
    HR data analyst
    Washington University School of Medicine

    Reply
  • 5. Archer Blog  |  20/04/2014 at 10:16

    Pretend You Have The Same

    […] o test the hypothesis with time-lagging-tests or blind-tests or something simila […]

    Reply
  • 6. David Pethick (@davidpethick)  |  28/09/2014 at 02:06

    Great post Morten.

    The “correlation is not the same as causality” caveat overlooks the obvious fact that the person who is presenting the work is the expert in the room and should be comfortable making a judgement about causality. I would go so far as to say that any expert who presents strong a strong correlation but doesn’t back it up with a strong recommendation has misjudged the needs of the audience.

    I’d be interested in your thoughts about the use of proxy data (with lower correlations) as a basis for decision making. So often I see good recommendations held back because of a lack of perfect data, when there is a clear story to be told that is supported by evidence.

    Cheers.

    David Pethick
    Founder, http://leading.io

    Reply
  • 7. Audrey Ciccone (@humanperspectiv)  |  17/11/2014 at 10:31

    Arriving at this party (blog post) a little late. I have to agree with Frank and the David’s comments. Statistical significance looks for causality but one would be hard pressed to find a finance guy having to prove causality to get their initiatives backed either. If you have strong correlations and have backed them up with additional data and investigation, take a stand and present recommendations. No prediction in predictive analytics is going to be 100% right either but we do the best we can to narrow the playing field to a small number of options which can produce the desired ROI.

    Reply
  • […] influences these specific factors. The model is shown below. This is important because as you know, correlation is not the same as causality. We have linked other published studies to the model as a way to validate the conclusion of our […]

    Reply

Leave a Reply

Fill in your details below or click an icon to log in:

WordPress.com Logo

You are commenting using your WordPress.com account. Log Out / Change )

Twitter picture

You are commenting using your Twitter account. Log Out / Change )

Facebook photo

You are commenting using your Facebook account. Log Out / Change )

Google+ photo

You are commenting using your Google+ account. Log Out / Change )

Connecting to %s

Trackback this post  |  Subscribe to the comments via RSS Feed


Enter your email address to subscribe to this blog and receive notifications of new posts by email.

Join 1,186 other followers

Latest Tweets

Feeds


%d bloggers like this: