MKT5612 Chap.7 Experiments and Causal Validity
Experiments and Causal Validity
A causal claim requires covariation, correct time order and a credible account that rules out rival explanations. Experiments strengthen that account by manipulating a treatment, observing an outcome and creating a comparison.
Random assignment balances groups in expectation and addresses selection differences; it is distinct from random sampling, which concerns how units enter the study and the population to which results may generalise.
Pre-experimental designs provide weak controls, true experiments use random assignment, and quasi-experiments construct comparisons when randomisation is unavailable.
Internal validity concerns whether the treatment produced the observed effect in the study, while external validity concerns transfer to other people, settings, treatments and times. Good design anticipates history, maturation, testing, instrumentation, attrition, spillover and noncompliance, then documents assignment, exposure and outcome measurement before looking at results.
What this chapter covers
- 01
Covariation, time order and rival explanations
- 02
Manipulation, treatment and comparison groups
- 03
Random assignment versus random sampling
- 04
Pre-, true and quasi-experimental designs
- 05
Internal-validity threats and design controls
- 06
External validity, spillover and implementation
Worked example · free
Diagnose an email subject-line test
- 3State why the observed association is not yet causal.
- 4Specify assignment, treatment and outcome for a stronger design.
- 3Name one implementation or validity check.
Key terms
- Causal Inference
- Reasoning about whether changing a treatment produces a change in an outcome rather than merely accompanying it.
- Random Assignment
- Allocation of study units to conditions by chance to balance observed and unobserved characteristics in expectation.
- Control Group
- A comparison condition used to represent what would have happened without the focal treatment.
- Internal Validity
- The credibility of the claim that the treatment caused the observed outcome within the study.
- External Validity
- The extent to which a causal result can transfer to other populations, settings, treatments or times.
- Quasi Experiment
- A causal design using a constructed comparison without full random assignment, requiring explicit assumptions about group comparability.
- Spillover
- Exposure of one unit to another unit's treatment, which can blur the intended contrast between conditions.
Experiments and Causal Validity FAQ
Why is random assignment different from random sampling?
Random assignment creates comparable treatment groups and supports an internal causal contrast. Random sampling gives population units known selection chances and supports generalisation. A study can have one without the other, so the two claims must be justified separately.
Can a quasi-experiment support a causal conclusion?
It can when the comparison and timing make rival explanations implausible under clearly defended assumptions. Because assignment is not random, the analysis must show why the groups or trends provide a credible counterfactual and test vulnerabilities.
Does high internal validity guarantee external validity?
No. Tight control can identify an effect in a particular study while the audience, treatment implementation or setting differs from the target deployment. Generalisation requires evidence about those similarities and relevant effect variation.
Exam move
Use a five-box causal audit: unit, treatment, comparison, outcome and assignment. Then draw arrows from every factor that changed with treatment to the outcome; these are rival explanations the design must block or measure. Practise distinguishing design labels only after explaining their control structure. In calculations or result interpretation, identify the estimand before discussing significance.
Strong answers separate internal validity from external validity and random assignment from random sampling. End with an implementation check, because contamination, attrition or noncompliance can weaken an otherwise elegant design after allocation has occurred. Take a simple before-and-after claim and add one control at a time: contemporaneous comparison, random assignment, stable measurement and exposure separation.
After each addition, state which rival path was blocked and which one remains. For quasi-experiments, write the identifying assumption as a sentence about the untreated outcome, then list an observable diagnostic such as pretreatment trends. Practise separating the effect of assignment from the effect among those who received treatment when compliance is imperfect.
Finally, write one internal-validity conclusion and one external-validity conclusion for the same result. Keeping these verdicts separate prevents a strong comparison from being mistaken for universal generalisability.
Working through Experiments and Causal Validity in MKT5612? Sia is AskSia’s AI Marketing tutor — ask any MKT5612 Experiments and Causal Validity question and get a clear, step-by-step explanation grounded in how MKT5612 is taught and assessed. Read this chapter free, then take your hardest questions to Sia.