Behind the scenes of the data age, a battle is brewing
On one side, we have data scientists, engineers of modern statistics.
They operate under a clear paradigm: good data is abundant, and the right models will extract meaning even from the mess.
On the other hand, demographer statisticians, heirs to the census tradition, trained in scarcity.
People who can spend hours debating the concept of “domicile”.
For them, reliable data is not what is ready, it is what was well thought out before being processed.
Both use methods, but not the same
Data scientists rely on the emerging pattern, while the demographer pauses in the face of the recurring exception.
- Automated methods extract regularities but ignore ill-defined contexts
- Traditional methods detect biases but can be difficult to apply
There is something, however, that the data scientist is still learning and that the experienced demographer already knows:
We won't know everything, but we can know enough with the right method.
It is a discipline that requires human judgment, shaped by well-trained heuristics, the fruit of experience, not just technology.
A Drop of Blood is Enough
The case of Elizabeth Holmes and Theranos is an almost archetypal example of how the seduction of innovation without method can lead to a resounding collapse.
It's also a modern fable about what happens when you abandon experience for pure technology.
At the center of the Theranos scandal was a promise that seemed as seductive as it was unfeasible: to perform hundreds of laboratory tests with just one drop of blood.
The idea was revolutionary, not only because of its convenience, but because it reconfigured the entire clinical inference model.
Instead of syringes, tubes, waiting and pain, a small hole in the finger would be enough to reveal your entire body.
The droplet as a mirror of the entire body
But there's a crucial difference between ingenious idea and reliable method.
And that's exactly where Holmes Castle collapsed.
What she wanted to do, after all, was what demographers do every day: infer the whole from a part.
But they learned, through centuries of error and revision, that this requires very strict rules.
And that, contrary to what narratives suggest, more people working does not lead to better results.
The difference between sample and divination
The problem was not the idea of using a drop of blood, this is already done for specific tests.
The problem was wanting to extract all the information from a single tiny sample, ignoring:
- The biological variability between capillaries and veins
- The Limits of Laboratory Technology
- The dilution effects
- The Quality Controls That Validate Any Inference
In other words, Holmes made a basic sampling error: thinking that any part represents the whole.
It's like taking the data as released by the IBGE and saying that you can do Geomarketing.
The Census is a universal survey, the opposite of a sample survey, but its information reflects that moment.
Using data from 2022 onward requires a method to define:
- What's being included
- What is being left out
- With what weight does each unit represent the others
The Fascination of the Part by the Whole
What Holmes tried to do was a high-compression sample, capturing a lot with next to nothing.
In demography, this is done with extreme care, mathematical modeling, measurement of uncertainty, and a clear warning that the data is the result of such a method.
For example, we can use small samples when:
- Selected with known odds
- Accompanied by auxiliary variables
- Calibrated for the total population
- Presented with error estimates
Ethics in Statistics
Holmes went beyond the technical error, she made the ethical error of inference.
I declared results as certain when there was neither reliable data, nor valid methods, nor replicability.
She disregarded the fundamental principle that drives all serious sampling: admitting what you don't know.
Demography offers no certainty, it offers estimates with a margin of error.
And it is precisely this methodological humility that makes it robust.
The moral of the story
Elizabeth Holmes reminds us that there will always be someone trying to replace method with charisma and a well-told story.
Turns out, the party can never speak for the whole without clear selection rules.
Trust in data depends, first of all, on trust in the process that generated them.
That's what sampling is: an art of choice, weighting, and inference.
It's not enough to have a drop, not all the blood, you have to know what it represents.
