Marketing Research and Analytics
Marketing research and analytics tests whether you can turn a decision problem into evidence: define the question, choose a design, collect and analyze data, and communicate an actionable answer. The exam emphasizes validity, sampling logic, and digital measurement.
Core rules and concepts
- Research process: define the problem and decision; specify information needs; choose secondary before primary; select design (exploratory, descriptive, causal); design sampling and instruments; collect; analyze; report.
- Samples: probability methods (simple random, stratified, cluster, systematic) support statistical inference; non-probability methods (convenience, judgment, quota, snowball) do not justify population estimates. Margin of error shrinks with the square root of sample size.
- Measurement: reliability is consistency; validity is measuring the intended construct. Avoid double-barreled, leading, and ambiguous questions; randomize order where order bias is possible.
- Analytics: distinguish correlation from causation; use control groups, pre/post designs, and holdouts. Track conversion rate, CAC, ROAS, retention, and incrementality rather than vanity impressions.
Worked example
An A/B test compares a landing page: control converts 400 of 10,000 visitors (4.0%); variant converts 480 of 10,000 (4.8%). Relative lift is (4.8−4.0)/4.0 = 20%. The absolute difference is 0.8 percentage points. With 10,000 per arm, the pooled rate is 4.4%, and the standard error of the difference is about 0.29 percentage points; the z-score is roughly 2.7, so the result is statistically significant at the 95% level. Practical significance still depends on the value of incremental conversions versus implementation cost. The exam answer should state both the lift and the decision rule.
Traps that cost marks
- Generalizing a convenience sample to the whole market.
- Treating a statistically significant lift as automatically profitable.
- Asking “How satisfied and loyal are you?”-a double-barreled item.
- Reporting last-click attribution as if it were causal incrementality.
- Ignoring nonresponse and coverage error in online panels.