When the sample is small, noise looks like culture. Statistics help HR distinguish the difference.
Message for CEOs and managers: relax, it’s probably statistics, not incompetence.
When N is small, noise doesn’t whisper: it screams.
In HR we love percentages: “We have a 33% turnover risk”.
Translation: 2 out of 6 people are thinking about leaving.
And this is where the problem begins.
The small-N illusion
With very small samples:
-
Variance is huge.
-
Confidence intervals are extremely wide.
-
Any change looks dramatic.
-
Any conclusion is fragile.
With N=6, a single person changes the picture by ±16.6%.
In People Analytics this leads to:
-
Over-intervening.
-
Stigmatizing teams.
-
Overreacting to a survey.
-
Making decisions based on noise.
And noise, in HR, costs trust.
So should we stop measuring?
On the contrary. We measure better.
When N is small, you need more context.
-
Look at trends over time, not a single snapshot.
-
Use deviations, not only the mean.
-
Combine data with real conversations.
-
Communicate uncertainty without losing leadership.
The normal distribution is fantastic… but with N=6, assuming normality is statistical optimism.
The real problem is not statistical. It’s cognitive.
With small samples we activate several biases:
-
Availability bias: a recent case weighs too much.
-
Gambler’s fallacy: we believe “it’s due” for someone to leave.
-
Trend illusion: we see patterns where there is only randomness.
And as leaders, we are programmed to act.
But acting on noise is expensive.
The uncomfortable question
When you look at a team’s data and decide to intervene…
Are you considering the sample size, or reacting to a real pattern instead of a statistical fluctuation?
Because if N is small, the most strategic move may not be to act more.
It may be to model better.