Chapter 6. Designing variables and outcomes — choosing your ruler

Key points

• Understanding the type of your data makes the later analysis precise
• Use three rulers appropriately: continuous, ordinal, nominal
• How you choose what to measure decides how persuasive your research and your practice become

Data has types

Once the design is settled, the next question is what to measure, and in what numbers. This is where understanding variable types becomes essential. Unglamorous, but hold it and you stop stumbling later — because the type of variable changes which analyses are available.

Three main types.

Continuous variables. Measurable as a continuous quantity. Examples: memory score, reaction time, age, gait speed, grip strength. Averages and differences are meaningful.

Ordinal variables. Ordered, but the intervals are not equal. Examples: a five-point rating from "low" to "very high" for concentration; a pain NRS; a Likert satisfaction scale. The order means something, but "4 is twice 2" does not hold.

Nominal variables. Category labels with no order. Examples: sex, region of residence, occupation, diagnosis. Counting proportions is what makes sense.

The idea of choosing a ruler

I think of these three as rulers. Measuring the same "concentration":

• by the number of correct answers on a test → continuous (fine-grained, but effortful to measure)
• by a five-point self-rating of how well you concentrated → ordinal (easy, but coarse)
• by "concentrated / did not" → nominal (easiest, least information)

Which ruler you choose changes the resolution of what you get. Measure finely and the information is richer, but the effort grows. This design judgement is exactly the feeling of a therapist choosing an assessment measure. MMSE, or a quicker screen? You pick the optimum between purpose and the constraints of the setting.

Outcome design decides persuasiveness

Among the variables, the outcome matters most: what you finally want to say got better.

There is a trap here. Measuring what is easy to measure. In a cognitive service it is simple to show that a "brain age" score improved. But is a score what the user actually wanted? Or is it

"I can do the shopping on my own again", "there is more conversation with my family", "my confidence about going out has come back"

— change in living? The same holds in business. In the discussion behind this text, the hypothesis was that customers find more value in "recovered confidence about social participation" than in "a visualised brain age".

Measure the meaningful indicator, not the measurable one.

For those of us dealing with wellbeing this is a decisive design principle. Functional ability, quality of life, meaning for the individual — how do you get the hard-to-measure but important things onto a ruler? It is precisely here that a therapist's expertise shows.

Designing the ruler with AI

Variables and outcomes can be discussed with AI too. Ask "what should be measured, and in what type, to test this question?" or "is there an existing measure for this concept (say, social participation)?" and it will lay out candidates.

But beware hallucination with assessment scales — fabricated instruments that do not exist. Always confirm the existence and validity of any scale the AI names against primary sources: the original paper, the official site. This is the distance repeated since Foundations.

A prompt to try

You are a researcher who knows measurement and scales well. I want to measure (the concept — e.g. older adults' confidence about social participation) in research and in practice.

1. Lay out candidate variables for measuring this concept, grouped by type: continuous, ordinal, nominal.
2. Tabulate the trade-off between resolution and practical effort for each.
3. If there are existing scales that measure this concept, list them (only ones you are certain exist; mark anything uncertain as "needs checking").

Chapter 6 summary

• The variable type decides which analyses are available later
• Choosing a ruler is a trade-off between resolution and effort — the same as choosing an assessment measure
• Measuring the meaningful indicator rather than the measurable one is what persuades