Statistics for marketing and consumer research: survey work, end-to-end

What survey-driven marketing statistics actually looks like

The typical marketing study is a cross-sectional survey (or a wave of one) that answers one of three questions: who our customers are (segmentation, brand awareness), what drives their behaviour (purchase intent, satisfaction, loyalty), and whether concept A beats concept B (product tests, pricing studies, message tests). Nearly every analysis is a group comparison, a correlation, or a regression — used in disciplined combinations.

Your typical study, mapped

• Brand-awareness or preference by segment — chi-square with Cramér's V, or Fisher's exact for niche categories.
• Satisfaction / NPS / loyalty across brands — one-way ANOVA / Kruskal-Wallis with post-hocs.
• Driver analysis (what predicts purchase intent) — multiple linear regression, or logistic if the outcome is binary (bought / didn't).
• Concept test (three product variants) — ANOVA on the rating scale; chi-square on the "which do you prefer" question.
• Scale reliability (brand-personality, satisfaction, loyalty scales) — Cronbach's α with the item-drop table; McDonald's ω when the scale isn't strictly tau-equivalent.
• Dimensionality discovery (a battery of attitude items) — exploratory factor analysis with KMO, Bartlett, and rotated loadings.
• Pre/post campaign lift — paired t-test / Wilcoxon if you have the same panel; independent-samples if the pre and post are different respondents.
• Correlations between attribute ratings and overall satisfaction — Pearson (or Spearman for skewed items).

The workflow, cell by cell

1. Export your survey data — Qualtrics, SurveyMonkey, Typeform, or hand-cleaned Excel; CSV/XLSX/SPSS are all accepted.
2. Enter the study statement and your specific objectives (e.g. "does purchase intent differ by age band?", "what predicts loyalty score?").
3. thericerca reads variable types, picks the right test per objective, and checks assumptions on your data.
4. Post-hocs fire only when their omnibus was significant, so the report doesn't over-claim.
5. Download the report as HTML, DOCX (for the deck), or PDF, plus a Jupyter notebook so the analytics team can rerun everything from the raw file.

For quick one-off checks, the free tools cover the most common needs: use the test-picker when you're not sure about paired vs independent, the sample-size calculator for planning a wave, and the reliability tool for a fast α on a scale before you commit to using it.

When you'd need something else

A concrete example

A new-product concept test with 300 respondents rating three variants on purchase intent, price fairness, and a 6-item brand-fit scale.

• Sample profile — descriptives (age, income) and frequencies (region, past-brand-user) as Table 1.
• Brand-fit scale reliability — Cronbach's α on the 6 items, with an item-drop table to catch any item that hurts reliability.
• Purchase intent by concept — one-way ANOVA with Tukey / Games-Howell post-hoc.
• Price fairness by concept — same shape.
• Preferred concept overall — chi-square on the "which would you buy" question, with Cramér's V.
• Drivers of purchase intent — multiple regression with brand-fit score, price-fairness rating, and past-brand-usage as predictors.

Every one of those analyses is in the single thericerca project. The report is an APA-formatted write-up plus the Jupyter notebook for the client's data team.