Investigación de mercados y análisis de la información
Design before data
Market research converts a decision problem into a research question and then into a design. Exploratory designs (secondary data, focus groups, depth interviews) clarify a fuzzy problem; descriptive designs (surveys, panels, observation, audits) quantify who, how much and how often; causal designs (experiments, test markets, split tests) test whether a change in one variable moves another. Candidates must separate primary data (collected for this problem) from secondary data (already exists, cheaper, often stale or mismatched to the question).
Sampling: probability methods - simple random, systematic, stratified (sample inside every stratum), cluster (sample whole groups) - allow inference with a known margin of error; nonprobability methods - convenience, judgment, quota, snowball - do not.
Measurement: nominal (labels), ordinal (order, unequal gaps), interval (equal gaps, arbitrary zero), ratio (true zero, so ratios are meaningful). Reliability is consistency; validity is measuring the right construct. Questionnaires fail through leading wording, double-barreled items, ambiguity and order effects.
Analysis: cross-tabulation with chi-square, t-tests and ANOVA for means, regression for association and prediction, factor and cluster analysis for data reduction and segmentation, conjoint for attribute trade-offs.
Traps. A large sample never repairs a biased sampling frame. Correlation is not causation; only a controlled design supports "causes". Stratified is not cluster. Interval scales forbid statements such as "twice as favourable". Ranking items are ordinal, not interval.
Rule of thumb. Let the question's verb pick the design: "how many / how often" → descriptive survey; "does X cause Y" → experiment; "we don't know the options yet" → exploratory qualitative work.
Key points
- Exploratory → understanding, descriptive → measurement, causal → proof.
- Probability sampling enables inference; nonprobability does not.
- Identify the zero to name the scale.
- Reliability ≠ validity.
- Match the statistical test to scale and design.