Data: the world's best ventriloquist

In the early 1980s, two psychologists ran a small experiment that still explains a lot of what happens in modern risk meetings.

People were given a simple scenario: a disease is expected to kill 600 people, and a decision has to be made about which programme to choose. In the first version, the choice was described in terms of lives saved. One programme would save 200 people for sure. The other offered a one-in-three chance of saving all 600, with a two-in-three chance of saving none. Most people went for the guaranteed 200.

Researchers then ran the same decision again, but described it in terms of lives lost. One programme meant 400 people would die for sure. The other offered a 1/3 chance that nobody would die, with a 2/3 chance that everyone would. This time, most people preferred the gamble.

Nothing about the underlying stats changed. The only thing that changed was the frame. The story the numbers were allowed to tell shifted, and so did the decision that felt sensible.

It’s tempting to file that away as ‘a bias’ and move on, but it’s more useful than that. It’s a reminder that numbers don’t arrive with meaning attached. Meaning gets built on top of them, and it’s built out of perspective: what you count, what you compare to, what you treat as normal, and what you decide the number is actually a proxy for.

That’s why the same dataset can genuinely support different decisions without anyone lying. One person sees a warning sign. Another sees noise. One sees a situation that needs immediate intervention. Another sees a system behaving exactly as expected. The disagreement isn’t always about intelligence or honesty. It’s often about which view of the same reality they’re looking at, and which assumptions they’ve smuggled in without anyone noticing.

This is where ‘data-driven decision making’ can be misunderstood — people talk as if the data will settle the argument, but data rarely settles anything on its own. It needs to be observed from more than one angle before it becomes useful. If you only look at one frame, you’re not being rigorous — you’re being persuasive. You’re choosing the lens that makes a particular action feel justified.

Most risk decisions get better when you rotate the view on purpose. Look at the same numbers as gains and as losses. Look at trends over different windows. Look at the organisation and then zoom in until the individuals appear. Look at the average and then look at the spread, because the spread is usually where the risk is hiding. Look at what went up, and then ask what changed in the environment that might make ‘up’ either good or bad.

It’s a bit like a kaleidoscope. The pieces don’t change, but the pattern does, and the pattern is what your brain responds to. The mistake isn’t that different patterns exist; the mistake is believing the first pattern you see is the truth.

Good analytics isn’t about producing a single definitive view that ends the conversation. It’s about giving people enough honest perspectives that they can see what’s stable across frames and what only exists because of the way something was presented. That’s the difference between ‘numbers on a screen’ and something you can actually trust when the decision is uncomfortable.

When I hear someone say “the data is clear”, I don’t assume certainty. I assume a frame. The better question is: what else could these numbers reasonably mean, and have we looked from enough angles to know the difference?