Choose suitable statistical methods, check assumptions and prepare reproducible results.
Statistics with discipline: hypotheses, comparisons and decision thresholds are written down before anything runs, then executed from the prespecified plan — two-group and paired tests, analysis of variance with post-hoc comparisons, nonparametric tests, contingency tables and correlations, standard regression, and pre-experiment sample-size solving. Every result carries its effect size and confidence interval, reported with descriptives in academic style; non-significant findings stay in.
Assumption checks (normality, variance equality) accompany the report as information, never as an automatic switch of tests; missing cells follow stated rules with counts; the same data and plan reproduce the same numbers for review.
Out of scope, stated plainly: Bayesian analysis, multilevel models, survival analysis, meta-analysis and time-series fitting — declined rather than approximated.
Examples describe the request and the shape of the expected deliverable; they are not completed run results.
Run the group comparison per this plan: descriptives, Welch t-test, effect size, assumption checks — keep non-significant results in the report.
Full test results with effect sizes, confidence intervals and assumption checks; odd screens are reported, never used to auto-switch tests.
How many paired samples for d=0.5, power 0.8, two-sided α=0.05? Also solve the d=0.4 case.
Both solved sample sizes with stated assumptions, labeled a-priori power estimates rather than outcome guarantees.
iOS · Android
iOS
Android
MIT
MIT (root LICENSE.md)
MIT at the pinned upstream revision; full license and per-file adaptation/asset audit are preserved in references/licenses/. Floe additions and synthetic fixtures are identified separately.
Publication commit 25e07dbeeb, audited source 84c524b6fb; catalog digest iOS 46196ba57ed7… · Android 16de89d019e2….
Version records ship with the signed catalog; the target and released versions are shown below.
