Using the Scientific Method to improve #PeopleAnalytics: A quick primer

In the comments of their latest excellent Directionally Correct podcast with Emily Pelosi, PhD , I asked Cole Napper & Scott Hines, PhD , PhD whether #PeopleAnalytics is scientific. Perhaps I should have asked:

Should #PeopleAnalytics be scientific in the same way that, e.g. pharmaceutical research is scientific because peoples' well-being depends on it?

Since #PeopleAnalytics can substantially contribute to employee well-being, it should also be scientific.

But what do we mean by "scientific"? For something like PA to be scientific, it must follow the scientific method, which means much more than data-led data mining and experimental research design.

Here is a quick primer on the scientific method as applied to #PeopleAnalytics :

1. Ask a research question

This should preferably relate to a business outcome that is interesting to your ELT (Executive Leadership Team) who creates the budget for your continued salary, rather than some workforce characteristic which relates to some area of your fun research:

a) Correct: Business outcome question: Why is profit low?

b) Incorrect: Workforce question: Why are employees unhappy? Wrong, because disgruntled employees may not mediate the relationship between people processes & business outcomes (this excellent Directionally Correct podcast with Marcus Crede refers to mediation).

?

2. Generate hypotheses based on a model linking people processes & business outcomes, for example, model:

People Processes -> Workforce Capabilities -> Organizational Capabilities -> Business Outcomes

?

You can generate hypotheses using the following:

a) Ideal: Qualitative research like interviewing business outcome owners

Or?

b) Less ideal: Data mining for correlations because you're likely to find spurious correlations , not to mention that this method will not engage the ELT, who will then know less about you and your excellent work and may label you as "some techie" and #PeopleAnalytics as a "some techie discipline" (unless this is what you want of course).

Remember that qualitative research tells you what; quantitative analysis tells you why.

?

3. Create an experimental research design

Experimental research design ensures that the business outcomes resulting from your analytics are caused by the factors you're testing in your hypotheses. So, for example, you might claim that improving employee experience improved productivity, whereas it was because your company started using more robots that year. Again, a good research design would address this.

So research design is a necessary but insufficient condition for labelling a process as scientific.

?

4. Generate data for testing constructs (variables) in the hypotheses

Chances are your systems won't already contain the data you need to test your hypotheses, such as psychological constructs (as noted elsewhere by Igor Menezes ), so you'll need to generate a lot of the data for any given project yourself.

5. Select an appropriate statistical model and test the hypotheses

Selecting a model means choosing one whose assumptions match the nature of the data you're analyzing. e.g. if you're analyzing hierarchical data, you can't use multiple regression .

And statistical model means statistical model because machine learning models are not designed for hypothesis testing. I sometimes break this rule because #PeopleAnalytics operates in the real world, not in an ivory tower. Still, I'm aware that I've deviated from the scientific method and list it as a risk factor in my project report.

Here endeth the formal scientific method

#PeopleAnalytics is an applied science, and therefore its process must continue beyond the scientific method for a bit:

6. Use hypotheses supported by the model to build new people processes

As Emily Pelosi, PhD notes, analytical results must be applied. Where do you apply them? To your people processes because they'll then give rise to the workforce capabilities you need to enable the organizational capabilities which give rise to your ELT's desired business outcomes:

People Processes -> Workforce Capabilities -> Organizational Capabilities -> Business Outcomes

?

7. Measure business outcomes resulting from the new people processes and keep adjusting the processes?

Your analysis is unlikely to be perfect, and in any case, organizations are not static beasts: life changes, and so your processes must change along with them, e.g. many of the people processes you implemented for those long ago days when employees mainly worked in offices are probably not as effective for generating desired business outcomes in the post-covid era as they once were.

And there it is. Use it in good health.