5 min de lectura
Correlation is not diagnosis: the costliest mistake in people analytics
Two lines moving together explain nothing on their own. David Botella explains why mistaking correlation for diagnosis is the costliest error in people analytics.
David Botella
CTO · 16 Nov 2026
Índice de contenidos
The chart that seems to explain everything
In people analytics it is easy to find two lines moving together: tool usage and team satisfaction, training hours and retention, manager tenure and declared engagement. That coincidence is often mistaken for an explanation.
- Two variables moving together says nothing about which one causes the other.
- A convincing visual pattern can be pure coincidence in the available data.
- Acting on a correlation without checking it can make the original problem worse.
When two lines rising together get mistaken for a cause
We see people analytics teams presenting a finding with enthusiasm, "teams with more training have less turnover", and jumping straight from there to a recommendation: invest more in training everywhere.
Training might reduce turnover. It is also possible that teams with less turnover, being more stable, simply have more time to train. The causal arrow could point either way, or depend on a third factor nobody measured.
- Recommendations based on a single chart with no proof of causation.
- Large investments justified by a correlation never checked with an experiment.
- Conclusions that flip direction as soon as one more variable is added to the analysis.
Correlation describes, diagnosis demands another question
Our reading is that a correlation answers "what moves together", and a diagnosis answers "what would happen if we changed this and everything else stayed the same". These are different questions and demand different evidence.
“A correlation tells a coincidence. A diagnosis tells a proven cause. Confusing them is expensive.”
— David Botella
CTO at People & Play
That jump from one to the other is exactly where most money gets lost in data driven people decisions.
Three steps before turning a pattern into a diagnosis
You don't need a team of data scientists. You need discipline before acting on an eye catching finding.
- Look for alternative explanations before accepting the first one
- Ask what third variable could explain both things at once.
- If a plausible alternative exists, the finding needs more evidence before acting.
- Test the change in a small slice before scaling it
- A pilot in one team or site reveals more than a retrospective chart.
- If the cause is real, the pilot should show the same effect.
- Separate who finds the pattern from who decides to act
- The data team reports the finding with its limits, not with closed conclusions.
- The decision maker asks for causal evidence before approving the investment.
Questions before acting on a correlation
Before turning a data finding into a people decision, it is worth asking:
- What third variable could explain this pattern without one thing causing the other?
- Did we test the change on a small slice before scaling it to the whole organisation?
- Did the data team present the finding with its limits, or as a closed truth?
- What evidence would we have if the cause ran exactly opposite to what we assumed?
- How much money are we about to invest based on a single chart?
The right data doesn't replace the right question
People analytics doesn't fail from lack of data. It fails when it mistakes a well visualised coincidence for a proven cause.
If you want to review how your team moves from a data pattern to a decision, let's talk. One more question before investing is usually enough.