Butterflies, Jazz Bands, and Golf Balls: Why the Traditional HR Measurement Is Missing the Point

How do we better measure human behaviour in HR & Psychology?

May 02, 2025

In the interest of providing multiple means of engaging with this blog, below is the podcast style conversation, using our AI friends Johnny and Joanne who go into each blog post at a deeper level. Or below is the usual written format which is a little more TL;DR

1×

0:00

-18:32

This month's blog emerges from a few events that could not have been predicted. The most interesting was a chance exposure to a new colleague working in our R&D function—a qualified psychologist with a rich background in research. The second was revisiting another chapter from the brilliant Introduction to Ecological Psychology (Blau & Wagman, 2023). Warning: it's another one that looks to shake many of the foundations we in psychology place ourselves on, but one that needs exploring in the mainstream.

The Traditional Approach: A Quick Primer

The traditional approach to measuring behavioural changes is through identifying the dependent variable and independent variable. The dependent variable is the outcome you are looking to measure; the independent variable is the thing you are manipulating to understand how it interacts with the dependent variable. For instance, how much force I put into hitting a ball (independent variable) affects how far it travels in meters (dependent variable).

This works really well in linear systems—the basis of most traditional sciences we're educated in, such as Newtonian physics. In linear systems, the thing we're manipulating will have a proportional effect on the outcome. The harder you hit a ball, the further it goes. The heavier the ball, the shorter the distance it will travel with the same force.

This well-regarded scientific approach is then, understandably, repeated in the study of human behaviour, whether psychology, neuroscience, or cognitive science. You take an outcome (dependent variable) such as wellbeing, control for everything else, and manipulate something like how much of an asshole your boss is (independent variable).

When Linear Thinking Meets Non-Linear Reality

This approach to researching phenomena is normalised. I did an MSc in human systems, and at no point was this questioned. It's culturally embedded, in that, "it's just the way we do things round here."

But through my exploration into Ecological Psychology and Dynamical Systems theory, I discovered this:

"On the other hand, if a systems dynamics are the result of nonlinear (e.g. multiplicative) processes, then it follows that the effects of perturbations will not be localized and will percolate throughout the system due to interaction-dominant dynamics" (Favela, 2024, pg.102)

What this means is that if a system is non-linear, then any manipulations of independent variables (perturbations) will not have a proportional, nor local effect on the dependent behaviour. This is what's commonly referred to as the butterfly effect—a butterfly flapping its wings over one side of a continent can cause a hurricane on the other.

This is not a linear cause. If the voraciousness of a butterfly wing flap was the independent variable and hurricanes were the dependent, you would not find a statistically significant correlation. This is because it is local effects and component-dominant interactions that independent and dependent variables sciences focus on. The butterfly flap is perturbing through the system, and only the interaction with other perturbations creates an effect. This is the essence of chaos theory and complex dynamical systems.

Jazz Bands Don't Play Solo

If you're investigating a complex system, such as one with more than one human and an environment, then controlling for other variables by isolating or freezing them changes the thing you're studying so much it will no longer behave in the same way. This is due to the "interaction-dominant dynamics" part—meaning it's the interaction of components, not the individual components themselves, that dominate the reason for behaviour.

This is a bit like separating a jazz band into separate rooms, asking each one to play a solo, then wondering why this sounds different when they all come together and jam off each other. The improv jazz sound is dominated by the interaction of players (interaction-dominant), not the individual players in isolation (component-dominant).

As a far cleverer fellow David said to me, "independent variable assumes independent components stupid”(I added the stupid, as upon reading what he said it seemed obvious) Another assumption making an ass out of me...

The difference between Newtonian physics and human behaviour is one is linear and one is non-linear. Not taking this into account and using measures for independent components in interaction-dominant systems is a little like asking what weight is the colour yellow...

When Noise Is Actually Signal

The second issue is the timing of measures. Stable systems are that—stable. If I put my keys on the table, they don't move, unless someone moves them (damn that other human in the environment). A dynamic system is never stable. They constantly oscillate to different degrees, so single measures of aggregates potentially removes the value. In linear measures, we want to remove noise. When in dynamic systems, the value can actually be the noise!

A Better Way Forward: Mapping the Contours

Some of the best researchers and practitioners in this space know this. Instead of independent and dependent variables and aggregates of scores at fixed points, they're moving towards using RQA (recurrence quantification analysis). RQA effectively maps out the landscape of attractors in a system.

Think of attractors in a system like contours, valleys, and sand bunkers on a golf green. The sharper the contour, the easier it is for the ball to get stuck in it. The broader the valley or sand bunker, the more likely the golf ball will enter it. Balls rolling from different directions on the same green will all perform differently, each contour changing the direction, influencing how it interacts with the next.

(Vallacher et al., 2010)

You can start to map out the contours, valleys, and sand bunkers by measuring the recurrence of different ball trajectories. If you merely added all the directional changes up and averaged them, you wouldn't understand the environment a ball has to navigate. The value is in the differences, the noisy variable data.

From Golf Balls to Culture: Practical Implications

Those studying the behaviour of humans also understand that the landscape has many different performance attractors. Some shallow and narrow, some very broad and very deep. This may be a great way to measure culture (or econiches)—"why do golf balls keep ending up over there?" may be applied to "why do people more often than not behave in this way?"

Some things are really obvious, like are people not coming into the office? Well, the door is broken. Once you get past the linear, local, component-dominant reasons, you start to see variance of behaviour within a trend—for example, why do some people come in to the office more, but some less? You start to move towards non-linear, non-local, component-interaction-dominant systems. And the most effective way is measuring them using the correct, most valid tools.

Is your organisation still trying to play Newtonian physics in a quantum world?

People Geek's Substack

Discussion about this post