About Stats for Psychology

Who Made It

Stats for Psychology was created by Elizabeth Bayardelle, PhD, a behavioral scientist who studies what happens after painful experiences are over and what helps people recover.

• PhD in Business Psychology, The Chicago School of Professional Psychology. Her dissertation, Behavioral Responses to Failure, used structural equation modeling to test how goal orientation, psychological capital and intrinsic motivation shape the way people respond to failure.
• Master's degree, John Jay College of Criminal Justice, with a thesis classifying offender and victim behaviors using Smallest Space Analysis.
• BA in Psychology, Pomona College, where she worked in a research group whose study of reactions to health threats was published in the British Journal of Health Psychology (2011).
• Applied behavioral researcher at FrayaTech, where she designs and runs studies with pre/post instruments and longitudinal data, and author of two books from MSI Press.

More at elizabethbayardelle.com.

Why It Exists

Most statistics resources are written for people who will become statisticians. Psychology students, researchers and clinicians need something different: to know which method fits their question, what a reported result means, and enough of the math to trust it. Every page here starts from that need.

Methodology

This section says exactly what is checked and how, including what isn't.

Which Methods Are Covered

The methods on the site come from Blanca, Alarcón and Bono (2018), a review of the statistical analyses reported in psychology journals. Where the review counted several procedures as one category (for example, several kinds of ANOVA), they're split so each gets its own page. A few pages go beyond the review: methods other pages send readers to when their own method doesn't fit, such as moderation analysis, partial correlation and the exact binomial test. A few procedures the review lists don't have pages yet, such as odds ratios and relative risk. The site covers 72 methods in 15 families.

How Each Page Is Built

Every page has the same six parts in the same order and follows one worked example from the research question to the APA write-up. Every example dataset is simulated, and each page labels it “Simulated data.” Each dataset except the one-sample t-test's was generated by a script with a fixed random seed; re-running every one of those scripts reproduces its stored data exactly.

How the Numbers Are Checked

These checks run as automated tests:

• Results are recomputed from the data. Every Napkin Math result and every numeric cell in the results tables is recomputed from the page's dataset by separate code, and the test fails if a stored value disagrees at the precision shown.
• Reported statistics are checked against those results. Wherever a page makes clear which result a number stands for, the statistics it reports are checked too: on its “Reported As” lines, in its one-line results and in the APA write-up sentence. The test fails if a number doesn't round to the recomputed value. That covers most numbers on the reported lines and about a quarter of the numbers in the write-up sentences.
• Written-out arithmetic is checked where it's plain numbers. On about six in ten Napkin Math steps, the arithmetic as written must come to the result shown, within one unit of the last decimal or 1%, whichever is larger, to allow for rounded intermediate values. Steps that show a p-value lookup or a function the checker can't evaluate are skipped.
• The calculators are tested against independent references. Power, sample-size, confidence-interval and p-value results are compared with SciPy and statsmodels, and with worked examples published by Faul et al. (2007), the G*Power 3.1 manual, Hedges and Rhoads (2010), Bloom (2006) and the PASS documentation. The Result Decoder's effect sizes and their intervals are compared with R's effectsize package, its RMSEA values with lavaan, and its consistency verdicts with the test cases in statcheck's own test suite, apart from a few documented differences.

What isn't checked automatically: numbers in explanatory prose, such as readings and handwritten notes, the write-up numbers that can't be tied to one result, and some approximations the calculators rely on. Power for rank-based tests (Mann-Whitney, Wilcoxon, Kruskal-Wallis) uses a standard, conservative efficiency adjustment, and several calculators use normal or large-sample approximations. The tests confirm these are computed correctly, not that the approximations are exact.

Independent Review

Every page has been through two blind reviews, in September and October 2026. Each time, a reviewer who saw only the published page and its data reproduced the page's numbers in R, and in Python where practical, ran the page's code, and spot-checked its citations. Every finding was then checked before it was fixed or rejected with evidence, and each page's notes record the outcome. Every critical and major finding was fixed or rebutted; a handful of minor items wait on a source or a decision.

Sources and Quotations

References were checked against Crossref or the publisher's record when each page was written; a random audit of 40 in October 2026 found no errors in authors, year, title, journal, volume or pages. Every quotation on the site has been matched word for word against the source's own text, in September and October 2026. Most were matched by script. Twelve were checked by eye against the page image, where a scan's text layer garbled a symbol or a line break, and five were matched against the authors' accepted manuscript rather than the published version. Quotes that couldn't be matched were removed, with their citations kept, and the record of each match is kept with the site's source code.

Software Output

Every page shows how to run the method in R and Python, and almost every page in SPSS (three use SPSS Amos instead). Each output panel says how it was made: some R and Python output was captured from a real run, with the version and date given; the rest, including all SPSS output, was laid out from the page's computed results, so labels and layout may differ slightly in your version.

How Pages Are Written

Pages are drafted with AI assistance, following a detailed written specification for every section, and then go through the checks and review above.

Corrections

If you find an error, from a wrong number to a misleading sentence, please email support@statsforpsychology.com with the page and what you found. Confirmed errors are corrected on the page.

Scope

Stats for Psychology is an educational resource. It explains methods and how to read them; it isn't statistical consulting, and it can't see your data or your design. For analyses that carry real weight, such as a thesis, a submitted paper or a clinical decision, check your choices with an advisor or a statistician. Benchmarks for “small,” “medium” and “large” are conventions, not rules.

How to Cite

For the site as a whole, in APA style:

Bayardelle, E. (2026). Stats for Psychology. https://statsforpsychology.com

For a single page, cite it as a web page with its title and URL, and the date you read it.