Research statisticsMethod & computationVersion of record

Your panel will ask why you used that test. Have the answer ready.

Problem
A wrong test still returns a clean-looking p-value. Nothing in the output flags the mistake, so it survives every check until somebody asks.
Method
A deterministic decision table maps measurement levels, group counts and design to a test, and states the reason. Reviewed Python computes every number in an isolated sandbox. A language model, when enabled, only reads your objectives — it never produces a statistic.
Checks
Assumptions are tested on your data. When one fails the analysis is re-run with the appropriate alternative and the report says which test was used instead, and why[1]. Every figure in the prose is traced back to the computation before the report ships.
Availability
The feasibility assessment — the recommended tests and an itemised estimate — is free and precedes payment. Twenty complete analyses are published in full[2], and six tools need no account at all[4].
Responses, one item
Response distribution for the item “I review lessons daily”
11.1
210.6
343.9
435.6
58.9

“I review lessons daily”, five-point scale, n = 180.

Result

Students with stronger study habits earned higher averages.

rs .68df 178p < .001significant

Pearson correlationSpearman rank correlationnormality of study habits rejected, p = .005

rs(178) = .68, p < .001

Fig. 1. Study habits and academic performance, from a published sample study[3]. The distribution, every statistic and the superseded test are what the engine produced; nothing here is illustrative.

§1 What a report is made of

Selecting a part shows what produces it. Nothing in the list is written by a language model; the prose is generated from the computed results and then checked against them.

  1. 01

    The chosen test

    Measurement level, number of groups and design decide it. Asked to “correlate gender and civil status”, the engine refuses a correlation — both are categorical — and runs chi-square instead, saying so.

    Rules engine · deterministic
  2. 02

    The assumption check

    Shapiro-Wilk and Levene run on your data. A failure changes the test rather than being noted and ignored, and the substitution is printed with the p-value that caused it.

    Checked per analysis
  3. 03

    The statistic and its size

    The test statistic, degrees of freedom, exact p-value, and an effect size with a plain magnitude word, because significance alone does not tell a reader whether the finding matters.

    scipy · statsmodels
  4. 04

    The plain-language finding

    What the result means for someone who does not read statistics, set beside the exact APA statement for the record.

    Generated, then validated
  5. 05

    The traced figures

    Every number in the prose is matched back to a computed result. One that cannot be traced is flagged rather than published.

    Numeric-claim validator
  6. 06

    The reproduction

    A Jupyter notebook or plain Python script that reruns the analysis from the raw file, so a reviewer can check the arithmetic themselves.

    Exported with the report

§2 Scope

Stating the limits is what makes the rest credible. Forty-two tests are implemented.

Covered

Descriptives and frequencies. Weighted means with verbal interpretation. Correlation. t-tests and their rank equivalents. One-way, two-way, repeated-measures and mixed ANOVA, with post-hocs gated on a significant omnibus. ANCOVA. Chi-square, Fisher’s exact and McNemar. Linear, logistic, ordinal and Poisson regression. Mixed models. Reliability and inter-rater agreement. Exploratory factor analysis. Survival.

Not covered

Structural equation modelling. Confirmatory factor analysis. Item response theory and Rasch. Meta-analysis. Propensity-score matching. Time series. If your design needs one of these, this is not the tool, and the feasibility assessment will say so before you pay.

§3 Rates

The feasibility assessment is free. Credits are spent only when an analysis runs. Across the published studies the estimates run from roughly 25 to 100 credits, and a four-objective survey lands near 50[2].

PackCreditsPricePer creditStudiesAction
Starter100$5$0.0502Choose
Researcher500$20$0.04010Choose
Lab1,500$50$0.03330Choose

You see the estimate before you spend, and credits do not expire.

§4 References

The claims above, and where to check them. No account required.

  1. 1
    Assumption-driven substitution, worked through end to end. Parametric vs non-parametric: when your data breaks the rules. How normality is checked, why n > 30 is not the licence it is taken for, and what a rank test actually tests.
  2. 2
    Twenty complete analyses with their reports, estimates and assumption checks. The sample gallery. Every figure quoted on this page is drawn from them.
  3. 3
    The study in Fig. 1. Descriptive-Correlational Survey: Study Habits and Academic Performance. Includes the full report, the assumption diagnostics and the reproduction script.
  4. 4
    Free tools, no account: the test chooser, the sample-size calculator, the reliability calculator, the APA formatter, the data cleaner and the chart builder. All six are listed at /tools.

Start with the free assessment. Spend nothing until you agree with the plan.