Personality and Individual Differences

State versus Trait Tracker

Fifty-six moments from each of four fictional lives. Judge them from one of those moments first — then see the fortnight it came from.

Simulated data — fictional experience sampling, generated from a seed

Learning objective

By the end you should be able to separate a person's typical level from where they are today, and treat variability as a characteristic rather than noise.

About 25 minutes. Nothing you do here is saved or sent anywhere.

How to use this

  1. Rank four people by their typical level, using one moment each.
  2. See the full fortnight and find out how you did.
  3. Then take the controls: within-person variability, measurement error, life events, and the random seed.
  4. Watch the running average settle, and see how many observations each person needs.
  5. Answer the two challenges.

One moment

Four people were asked how they were feeling, on a 0–100 scale, at a single moment. Rank them by what you think their typical level is — 1 for the highest.

That moment, and your ranking

Two challenges

Challenge 1 — same average, different people

A questionnaire asks both of them "in general, how do you feel?" and gives them the same score. What has it missed?

Challenge 2 — one reading, three ingredients

Someone scores 45 on a single administration. How much of that is their typical level, how much is today, and how much is measurement error?

Your answer

What this demonstrates

A trait is a distribution, not a setting

The tempting picture of a trait is a dial fixed at some value, with everything else counting as interference. The picture that survives contact with experience-sampling data is a distribution: a person has a typical level and a characteristic amount of movement around it, and both are stable properties of them. Asking whether someone "is" anxious is asking about the centre of a distribution and ignoring its width.

Within-person variability is a characteristic, not noise

Ada and Bo share a mean. A trait questionnaire describes them identically, and they are not living the same fortnight. How much a person fluctuates is measurable, reasonably stable across months, and predicts things the mean does not — which makes treating it as error a substantive loss rather than a tidy simplification.

Error and fluctuation look the same once and differ across many

In a single observation, measurement error and genuine within-person movement are indistinguishable: both push the number away from the person's typical level. Across repeated observations they separate, because error averages towards zero and the person's own variability does not. This is the argument for repeated measurement, and it is why the running-average chart settles at different rates for different people.

How many observations is a question about the person

The stability curve reaches a usable estimate quickly for Ada and slowly for Bo. There is no general answer to "how many observations do I need" — it depends on how much the person varies and how precisely you need to know their level. A study design that samples everyone equally is implicitly measuring some people much better than others.