Measurement

Defining a variable by the exact way it will be measured.

e.g. the number of words correctly recalled in two minutes.

Quantitative: numbers to compare. Qualitative: words and meanings, for depth.

Asking people directly: questionnaires, rating scales, interviews.

Watching and recording what people actually do.

People may be dishonest or inaccurate about their own behaviour or feelings.

You see behaviour but not the thoughts or feelings behind it.

If different measures agree, the finding is more trustworthy.

No. It is a measure of stress, not stress itself.

An idea, such as stress or memory, that cannot be seen directly and has to be measured through something else.

Direct: the behaviour itself is recorded. Indirect: a sign of something hidden, such as heart rate for stress.

Retrospective: people look back and report. Prospective: things are recorded as they happen.

Type I: finding an effect that is not real, a false alarm. Type II: missing an effect that is real.

The consistency of a measure: the same result under the same conditions.

Giving the same test to the same people twice; stable scores mean it is reliable over time.

Different observers rating the same behaviour and closely agreeing.

Questionnaire items meant to measure one thing giving similar answers.

A scale that always reads 5 kg too heavy. It is consistent, and consistently wrong.

A measure can be consistent and still measure the wrong thing.

Train the observers and use a clear, agreed coding scheme.

That the thing being measured has not really changed between the two tests.

Reliable means consistent. Whether the measure is accurate is a separate question.

Vague categories, changing conditions, and too little to go on.

A written definition of exactly what behaviour counts, so that every observer records the same thing.

Whether a measure or study actually captures what it claims to measure.

Whether the measure captures the actual concept, not something else.

Whether an effect is really due to what the researcher changed, and not to something else.

Whether findings apply to real-life settings, not only to the laboratory.

Reliability: consistent. Validity: measuring the right thing.

Measuring "intelligence" by how fast people click a mouse. That captures reaction time, not intelligence.

Tight control protects internal validity, but it can make the setting artificial and lower ecological validity.

A confound — another variable that could explain the effect (e.g. no control group).

Yes — it can give consistent results while measuring the wrong thing.

It is not a general word for "good". Say which validity you mean and what threatens it.

Face = looks right on the surface; content = covers every part of the concept.

Whether results apply beyond the study: other people, settings and times.