Illustrative output. The real run happens in your notebook.
Test selection
Outcome: continuous (reaction time, ms)
Groups: 2, independent
Normality: Shapiro-Wilk p = .31 ✓
Equal variance: Levene p = .04 ✗
→ Welch's t-test (unequal variances)
The decision tree, with each branch stated.
Result
t(43.2) = 2.31, p = .026
Cohen's d = 0.65, 95% CI [0.08, 1.22]
Group A slower than Group B by 48 ms on average.
Reported in APA style with effect size and interval.
Description
Which statistical test to use depends on the question, the measurement scale, the number of groups, whether samples are paired, and whether the assumptions hold. Agents often skip straight to a t-test. This skill walks the decision tree and shows its reasoning.
Then it runs the test in Python with SciPy, statsmodels or pingouin, checks normality and variance assumptions, falls back to a non-parametric alternative when needed, and reports the result with effect size and confidence interval in APA format.
Recommends a test and its non-parametric alternative
Checks assumptions with tests and plots and states which were violated
Runs the analysis in SciPy, statsmodels or pingouin
Reports statistic, p-value, effect size and confidence interval in APA style
Explains the result in plain language, including its limits
Editor's Recommendation
Recommended by
Runcell team
Why we built this skill
Every analyst has watched an agent reach for a t-test without checking anything. This skill makes the decision visible: outcome type, groups, pairing, assumptions, then the test. The APA sentence at the end is small, but it is what people paste into the results section.
”
What you get
A test recommendation with the reasoning
Assumption check results
The test output with effect size and confidence interval
An APA-formatted sentence ready for a results section
Use it when
Comparing groups in an experiment or A/B test
Survey and questionnaire analysis
Anyone who knows the question but not which test answers it
FAQ
How do I know which statistical test to use?
Start from four things: what kind of outcome you measured (continuous, ordinal, categorical), how many groups you compare, whether the measurements are paired, and whether the data meet the test's assumptions. Two independent groups with a continuous outcome usually means a t-test, or Mann-Whitney if normality fails. Three or more groups means ANOVA or Kruskal-Wallis. Two categorical variables means chi-square. The skill walks this tree with you and records the answer.
Does it just pick a test or also run it?
Both. It recommends, checks assumptions, runs the test in Python, and reports the result. In Jupyter every step is a cell you can rerun with new data.
Does it report in APA format?
Yes. It writes the statistic, degrees of freedom, p-value, effect size and confidence interval in APA style, for example t(48) = 2.31, p = .025, d = 0.65, 95% CI [0.08, 1.22].
What if my data violate the assumptions?
The skill tests for it and switches to a non-parametric or robust alternative, then explains the change. It never silently runs the parametric test on data that fail the checks.
Runcell team