stat test to use
Got it 👍 — here’s the same “Which statistical test should I use?” guide, but rewritten in clean plain-text formatting, so it displays perfectly anywhere (no tables needed):
---
## 🧪 WHICH STATISTICAL TEST SHOULD I USE?
---
### STEP 1️⃣ — What kind of data do you have?
- Continuous / numeric → e.g. time on site, temperature, revenue
- Categorical / binary → e.g. converted vs not, device type
- Ordinal / ranked → e.g. survey ratings 1–5, rank order
---
## 🧮 IF YOUR DATA IS CONTINUOUS
A. One group (vs a known value) → Test if your sample mean differs from a fixed number (like μ₀ = 10).
- Small sample, σ unknown → One-sample t-test
- Large sample or σ known → One-sample z-test
---
B. Two independent groups → Compare two different groups (e.g. time on site for A vs B).
- Equal variances → Independent t-test
- Unequal variances → Welch’s t-test
- Data not normal → Mann–Whitney U test (non-parametric)
---
C. Two paired groups → Same users measured twice (e.g. before vs after).
- Data normal → Paired t-test
- Data not normal → Wilcoxon signed-rank test
---
D. Three or more groups → Compare multiple versions (A, B, C).
- Data normal → One-way ANOVA
- Data not normal → Kruskal–Wallis test
---
## 🔢 IF YOUR DATA IS CATEGORICAL (YES/NO, 0/1)
A. One group (single proportion) → Test if a single proportion differs from expectation.
- Small sample (n < 30) → Binomial exact test
- Large sample → One-sample z-test for proportion
---
B. Two groups (two proportions) → Compare conversion rates between A and B.
- Large samples → Two-proportion z-test
- Small samples → Fisher’s exact test
---
C. Three or more categories → Test independence (e.g. device type vs conversion).
- Large counts → Chi-square test of independence
- Small counts → Fisher’s exact test (extended)
---
## 📈 IF YOUR DATA IS ORDINAL / RANKED
- Two groups → Mann–Whitney U test
- Paired samples → Wilcoxon signed-rank test
- Three or more groups → Kruskal–Wallis test
- Correlation between ranks → Spearman’s rank correlation
---
## 🧠 IF YOU’RE TESTING RELATIONSHIPS / PREDICTIONS
- Predicting continuous outcome → Linear regression (t-tests on coefficients)
- Predicting binary outcome → Logistic regression
- Comparing nested models → Likelihood ratio test / Chi-square
- Time-to-event outcomes → Cox proportional hazards model
---
## 🧾 ASSUMPTION CHECKS (for parametric tests)
- Normality → Shapiro–Wilk test
- Equal variances → Levene’s test or F-test
---
## 🧭 QUICK REFERENCE SUMMARY
- Continuous data → use t-tests / ANOVA / regression
- Categorical (binary) → use z-tests / Fisher / Chi-square
- Ordinal / non-normal → use Mann–Whitney / Wilcoxon / Kruskal–Wallis
- Small samples (proportions) → use Exact / Fisher’s test