MKTG90011 · Marketing Research
Measurement and Scaling
A construct — loneliness, satisfaction, loyalty — can't be observed directly, so we operationalise it into survey items and assign numbers by a fixed rule. The scale type of each variable is the hinge of the whole subject: it fixes both the summary statistic you may report and, more importantly, which inferential test is legal. The four types form a ladder — nominal (labels, e.g. gender), ordinal (ranked, e.g. finishing place), interval (equal gaps, no true zero, e.g. temperature), ratio (true zero, e.g. spend) — and each rung unlocks more statistics. A key MKTG90011 convention: Likert items are treated as interval, so means and parametric tests are allowed on them. A measure must also be reliable (consistent — checked with Cronbach's α ≥ 0.7) and valid (actually captures the construct). Get the scale wrong and every test downstream is wrong, which is exactly why “name the scale of each variable” is step two of the exam's signature which-test drill.
What this chapter covers
- 014.1 From construct to measure — operationalisation
- 02The four scale types: nominal, ordinal, interval, ratio (N / O / I / R)
- 03What statistic each scale permits (mode → median → mean)
- 04The Likert-as-interval convention
- 05Reliability — consistency and Cronbach's α ≥ 0.7
- 06Validity — content, construct and criterion
- 07Reliable vs valid: why you need both
Worked example: classify the scale and pick the legal statistic
- +1(a) Postcode — nominal. The numbers are labels with no order or arithmetic meaning, so the only valid summary is the mode (or counts/percentages).
- +1(b) Likert item — interval (by this unit's convention). Equal-appearing gaps, no true zero, so you may report the mean and run parametric tests.
- +1(c) Finishing position — ordinal. Ranked but gaps are unequal (the gap 1st–2nd need not equal 2nd–3rd), so the valid central tendency is the median.
- +1(d) Monthly spend — ratio. True zero and equal intervals, so the full toolkit applies, including the mean and ratios (twice as much spending is meaningful).
Key terms
- Construct
- A variable that cannot be observed directly (trust, loyalty, loneliness, willingness-to-pay). It must be operationalised into one or more measurable items before it can enter any analysis.
- Scale of measurement
- The rule by which a variable's numbers carry meaning — nominal, ordinal, interval or ratio. It fixes both the permitted summary statistic and which inferential test is legal.
- Likert-as-interval
- The MKTG90011 convention that multi-point agreement scales are treated as interval data, so means, standard deviations and parametric tests (t-tests, ANOVA, regression) may be applied to them.
- Reliability
- The consistency of a measure — it gives the same answer on repetition. For multi-item scales it is checked with Cronbach's α, with α ≥ 0.7 the usual threshold.
- Validity
- Whether a measure actually captures the construct it claims to. A measure can be reliable but not valid (consistently wrong), but it cannot be valid without first being reliable.
Measurement and Scaling FAQ
Why does the scale type matter so much?
Because it decides which statistics are legal. You can take the mode of a nominal variable but not its mean; you can rank an ordinal variable but the gaps are unequal; only interval and ratio variables support means and parametric tests. Misclassify the scale and every downstream test — t-test, ANOVA, regression — is invalid.
Is a Likert scale ordinal or interval?
Strictly it is ordinal, but MKTG90011 follows the common applied convention of treating multi-point Likert items as interval. That is what lets you report a mean rating and run parametric tests (t-tests, ANOVA, correlation, regression) on survey ratings — know this convention for the exam.
What is the difference between reliability and validity?
Reliability is consistency (the same result on repetition); validity is accuracy (it measures the right thing). A bathroom scale that always reads 2 kg high is reliable but not valid. You need both: a measure can be reliable without being valid, but never valid without being reliable.
What Cronbach's alpha is good enough?
The standard rule of thumb is α ≥ 0.7 for a scale to be considered internally consistent. Higher is better up to a point; very high values (> 0.95) can signal redundant items. Report it for each multi-item construct in the project.
Exam move
Drill the N/O/I/R ladder until you can classify any survey item in seconds, then pair each scale with its legal central-tendency statistic (mode/median/mean) — this is step two of the which-test drill and a steady source of MCQ marks. Lock in two unit-specific facts: Likert is treated as interval (so means and parametric tests are allowed) and Cronbach's α ≥ 0.7 for reliability. Be able to explain in one sentence why a measure can be reliable yet not valid — a classic short-answer item — and remember that getting the scale wrong invalidates every test that follows.
