Skip to content
craze_Platform

Groundedness

Groundedness is how much real evidence stands behind the people you are asking. It sits beside alignment, which is how closely those people have matched real human answers when tested. One tells you how solid the inputs are; the other tells you how the outputs have performed.

Every population shows both. You want them before you run something, not after you have been handed a plausible number.

Why two measures

They fail in different directions, and each hides the other's failure.

A population can be well grounded and badly aligned. The people are built from strong evidence, and they still do not reproduce how real humans answer. That is a real signal: something in the method is wrong, and it is worth chasing.

A population can be thinly grounded and accidentally aligned. Very little real evidence stands behind the people, and the answers happened to come out close on the questions tested. You got lucky. There is no reason to expect it to hold on the next question, and alignment alone would never tell you that.

Groundedness is the number that separates those two cases.

What it measures

Three things, combined:

How much we know. The number of distinct things established about each person.

How broadly. Whether that knowledge spans circumstances, behaviour, stated views and their own words — or piles up in one area. Twelve facts about someone's demographics is thinner than four facts spread across what they do, what they say and how they live.

How directly. Observed and reported evidence counts for more than combined or estimated. See Identity for what those labels mean.

It is reported as a distribution rather than a single number, because a population is rarely uniform. A typical one looks something like:

Lapsed subscribers, mid-size metros — version 3
   1,840  built from individual records and interviews
   2,160  drawn from measured distributions
   alignment 0.89  ·  38% individually grounded

That is a usable population. It is also, plainly, a different thing from four thousand interviews, and the number is there so nobody has to guess which one they are holding.

How to read it

High. Most people in the group are built from records about individuals, with behaviour and, ideally, their own words. Trust the twins as individuals — read them, quote them, ask them follow-ups.

Mixed. A grounded core with the rest filled out from measured distributions. This is the normal case for any population large enough to represent a market. Trust the aggregate; treat any single twin as illustrative rather than as evidence.

Low. Little individual evidence. The group's composition may still be right, but you are closer to asking a well-calibrated model of a demographic than a set of people. Treat results as a hypothesis and look at the source mix before looking at anything else.

Raising it

In rough order of effort:

  1. Tell us the shape of your customers. Reweighting does not raise groundedness by itself, but it aims the evidence you do have at the right people, which usually matters more.
  2. Bring behavioural data. Your purchase or usage records attach observed behaviour to matched twins, and observed behaviour is the most valuable thing an identity can hold.
  3. Bring your customers. Where you have rich records and a lawful basis, they become the twins directly. Highest available.
  4. Commission interviews into the group you care about. Slowest, and the only route to a population with real voice at depth.

See Population for how each of these works.

What it does not tell you

Groundedness says nothing about whether your population is the right group. A perfectly grounded population of the wrong people answers the wrong question very convincingly. That check is the ten sample twins on the population page, and a person reading them.