Measurement is everything that happens in the gap between an idea and a number, and what gets lost crossing it.
Psychology studies things nobody can see. Depression, executive function, belonging, theory of mind. Before any of them can be studied, something observable has to stand in for them. Measurement is that substitution, and every conclusion in the study rests on it.
Three things that are always true of it
It is a bridge, not a window. Constructs cannot be observed. Measurement builds something observable that stands in a defensible relation to them. You never see the construct. You see the bridge.
The number is not the thing. A score of 18 is a score on that questionnaire. It is not a quantity of depression.
It partly redefines. Operationalising narrows a construct, and over time the construct drifts toward whatever the standard task measures. Executive function becomes what executive function tasks measure. That drift is invisible from inside the literature and it is worth noticing.
The question that catches most overclaiming
Here it is, and it is the easiest analytical move in the whole course.
Is the conclusion written at the level of the construct, when the evidence only reaches the indicator?
Compare the conclusion sentence with the method section. That comparison is usually the whole answer, and it takes twenty seconds.
If the conclusion says wellbeing and the method says one item on a ten-point scale, those are not the same size. If the conclusion says executive function and the method says one computerised inhibition task, same problem.
Weak and strong, on the same study
Take a study reporting that bilingual children show better executive function than monolingual children, measured by performance on one computerised inhibition task.
Weak. The study used one task to measure executive function, which may not be a valid measure. Executive function is a complex concept and one task may not capture it fully.
That is true, general, and worth very little. It could be written without reading the study.
Strong. Executive function is a family of processes, and inhibition is one of them, so a single inhibition task indicates one component rather than the construct named in the conclusion. Any such task also requires processes beyond the one it targets, including comprehension of the instructions and sustained attention, so performance differences need not reflect inhibition at all. The claim the design supports is a difference in performance on this task. The claim about executive function would require convergent evidence from tasks tapping different components.
The second one names the component, names what else the task requires, and rewrites the conclusion at the level the evidence reaches.
Reliability is not validity
This trips people constantly, and the distinction is worth being precise about.
Reliable means consistent. Same result across time, across items, across raters.
Valid means it measures what it claims to.
Reliability is necessary but not sufficient. A bathroom scale that reads four kilograms heavy every time is perfectly reliable and entirely wrong. Two observers agreeing closely on a badly defined category are reliably coding the wrong thing.
So a measure can be highly consistent and still capture only part of what it claims. Saying that is much stronger than saying a measure was unreliable, which is usually not even true.
Does it work the same way across the groups being compared?
This is called measurement invariance, and it is the most valuable idea in this concept for any study comparing groups.
If an instrument was developed and normed on one population and then used on another, equivalent scores may not indicate equivalent underlying states. The items may be understood differently. The norms were built on somebody else.
Which produces a genuinely three-way conclusion rather than a criticism: it has not been shown that the instrument works the same way in both groups, so the difference could be real, could be the instrument, or could be both.
The same applies across time. Comparing scores at two points assumes the measure means the same thing at both, and developmental shifts and practice effects break that assumption.
When measurement gets used on people
Once a measure is used to make decisions, a different set of questions applies.
Where does the cut-off come from, and what balance of errors does it encode? What is the base rate in this population, and what does that do to predictive value? What are the consequences of a false positive and of a false negative here, and are they equal? Who bears each?
A screening tool that performs identically in two settings can be useful in one and misleading in the other, purely because the base rate differs. The tool did not change. Its usefulness did.
That is a sophisticated point, it is entirely concrete, and it is available whenever a source involves a test, a threshold or an algorithm.
How to write about it
Six moves.
Name the construct and how it is defined in words. Name the indicator, specifically rather than generally. Say what is missed. Ask whether the conclusion would change under a different reasonable operationalisation. Check it works the same across the groups being compared. Rewrite the conclusion at the level the evidence reaches.
The last one is the one that earns the mark. Everything before it is setup.
Where measurement meets the other concepts
Measurement and causality. A clean experiment cannot rescue a badly measured outcome. You have precisely manipulated one thing and precisely measured something else. Randomisation supports a causal claim about scores on this task, but the task is a narrow indicator, so the claim about the construct is weaker.
Measurement and bias. A tool can be perfectly reliable and still systematically misrepresent a group, if its definition, items or norms were built elsewhere.
Measurement and perspective. What counts as the outcome depends on who you ask. Symptom scores, ability to function, and the person's own account are three different definitions of recovery, not three attempts at one. So the gap between clinician-rated and self-rated improvement is not a reliability failure. It reflects two legitimate definitions.
That third one is the most sophisticated sentence available in this concept, and almost nobody writes it.
Where this gets examined
Paper 1 Section C can name measurement as the concept for any Area of Study, and it bites hardest wherever the content involves a construct that has to be inferred.
Paper 2 Section B rewards it on almost any unseen study, usually as the conclusion-versus-method comparison.
Paper 3 questions 1 and 2 turn on it constantly, and question 4 uses it as one of the four ways your own knowledge does interpretive work.
In your IA, Criterion C is measurement applied to something you built yourself.
For the practical side of building a measure, see the guide to operationalisation. This page is about evaluating one.
Can you do this yet?
- Explain why measurement is a bridge rather than a window
- Explain the difference between reliable and valid, with an example of one without the other
- Explain how operationalisation can quietly redefine a construct
- Explain why a task requires more than the process it targets
- Explain what measurement invariance is and which claims depend on it
- Explain why a cut-off's usefulness changes with the population, without the test changing
- Explain why disagreement between two measures may be informative rather than error
- Rewrite an overstated conclusion at the level the indicator actually supports
If you want the concept worked through with labs and worked answers, that is what the PsychExcel Six Concepts Workbook is built for.