Market Research and Data Interpretation
This domain tests whether you can pick a research design, judge data quality, and draw only the conclusions the data support. Expect tables, sample sizes and contradictory findings.
Design choices
- Exhaust secondary sources before commissioning primary research; primary is justified when existing data cannot answer the decision.
- Qualitative (depth interviews, groups) explains why; quantitative (surveys, panels, experiments) sizes how many. Qualitative findings are not projectable to a population.
- Probability sampling (random, stratified, cluster) supports inference; convenience, quota and snowball samples do not.
- Sample size drives precision: for a simple random sample at 95% confidence, margin of error ≈ 1/√n, so n=400 ≈ ±5% and n=1,000 ≈ ±3%.
Interpreting results
- Separate statistical significance from commercial importance.
- Correlation is not causation; only randomised experiments (A/B or holdout tests) establish causality.
- Watch wording bias: leading, double-barrelled and unstated-frame questions destroy validity.
- Useful tools: cross-tabs, Van Westendorp price sensitivity meter, conjoint for trade-offs, NPS = % promoters − % detractors.
Worked example
A survey of n=1,000 puts Brand A at 32% and Brand B at 29%, margin of error ±3 points. Reasoning: each figure carries ±3, and the standard error of the 3-point gap is roughly 2 points, giving a t-value near 1.5 - well short of 1.96. The gap is not significant at 95%, so report a statistical tie, or raise the sample to about 4,000 per group to detect a 3-point difference.
Traps
- Quoting subgroup results from tiny bases as if they were projectable.
- Confusing margin of error with total survey error (non-response, coverage, wording).
- Reading a recommendation into data that only show an association.