Teach with StatsCapybara

StatsCapybara has more than a hundred interactive lessons and a set of tools, all free and open to anyone. If you teach statistics or research methods, you can use any of it in a course: link it, embed it, print it, and assign from it. This page shows how.

What your students see

There are no accounts, no login and no email sign-up. A student opens a lesson and can start reading it and using its interactive straight away.

Progress is stored only in the student's own browser. Finished lessons show fainter in the lists, each course has a completion ring, and the homepage remembers where the student stopped, all from localStorage on their machine. None of it is sent anywhere, so there is nothing for you to administer, and you cannot use it for grading. Clearing browser data resets it, and a different device starts fresh.

The site does count page views anonymously with Google Analytics, set up so that it stores no cookie, and it records an anonymous event when someone uses the search, the quiz or an embed. I use those counts to see which lessons get read and which tools are worth maintaining. No student is identified and there is no coursework data to collect, so assigning StatsCapybara needs no data-processing paperwork. The privacy page explains all of it.

The site works offline once visited, prints cleanly, and reads well on a phone.

Link lessons from your syllabus or LMS

Every lesson has a clean, permanent URL, so you can link the right section for each week:

• A lesson: statscapybara.com/stats-1/central-limit-theorem/
• A calculator: statscapybara.com/power.html for sample-size questions.
• The test chooser: statscapybara.com/which-test.html when students ask "which test do I run?"

Paste those into Canvas, Moodle, Blackboard, a slide or a printed handout, and they work the same in each.

Printable handouts and posters

Every lesson prints as a clean handout: the text, the visualization as it was on screen, the worked APA example and the code, without the navigation or the sidebar. A student who presses Ctrl/⌘ P on any lesson gets a set of study notes. The formula sheet prints the same way.

Three one-page posters are built to pin up or hand out, and each fits a single sheet at A4 and US Letter:

• Which test should I use? lays the decision map out as four grids.
• APA reporting covers how to write up t, F, χ², r, and regression, with the italics and p-value rules.
• Assumption checks gives what to check per test, how to check it, and the fix when it breaks.

If your exam is on paper with printed tables, the tables page has them: a standard normal table, a t table with confidence levels along the bottom, and a χ² table. Print all three prints each one on its own sheet. The tables are computed by the same functions as the site's calculators, so a handout and an answer key made from the site give the same values. The same page has a ten-question lookup drill that accepts the table's rounded value as well as the exact one. Students are often unsure whether the rounded value counts.

Assignments on reproducible datasets

The practice datasets are CSV files, each with a short backstory, exercises and a worked solution. Each file is fixed, so every student who downloads it gets the same numbers. A whole class can run the same analysis and get the same answer, which makes marking simple. When a student's results differ, the difference is in their analysis, and you can look for it together.

Set a dataset for homework, then have students check their output against the lesson's SPSS/JASP walkthrough and APA example. All the numbers were checked against statistical software.

A ready-made semester project

If your course ends in a mini-project rather than an exam, One Study, Start to Finish is a worked model answer that you can hand out or keep back. It takes memory-2x2.csv through every stage a project report has to cover: the research question and its operationalization, the design and a power analysis, the data checks, the assumption checks, a two-way ANOVA with simple effects, effect sizes with confidence intervals, the APA results paragraph, and the limitations. Every number in it is reproducible from the CSV.

Two problems that a clean example cannot teach are left in on purpose. The study recruited 25 per cell when the plan asked for 32, so the guide states the shortfall and lists it among the limitations. And a boxplot rule flags four points in one cell that should stay in the data, even though deleting them would make the hypothesis look stronger. Either one makes a good seminar discussion on its own. If you assign a different file from the datasets page, students can copy the guide's structure but not its numbers.

An intermediate course usually sets a different project. Students collect their own data, answer one question with an ANOVA of three or more conditions, answer it again with a regression that includes control variables, and write it up. One Question, an ANOVA and a Regression works that project end to end on commute-mood.csv. Its seventh step, which student projects often leave out, makes the two analyses agree. The walking coefficient falls from 9.12 to 5.09 once distance from campus is in the model. The guide shows where the 4.03 points went, so the ANOVA and the regression turn out to be one model written two ways. Hand it out as a worked example, or keep it back and grade against its final checklist.

The guide also says once that it cannot do a student’s project, and that each course sets its own rules on help with a student’s own data, including help from AI tools.

An advanced course often ends with a journal club instead, where the class reads published studies that claim a causal effect and takes their designs apart. Reading a Causal-Inference Paper is written for that. It gives the seven questions to ask of a study built on an instrument, a threshold or the date a policy arrived, and the standard objections to each design, phrased as questions a presenter can answer. Then it works through one study from its first table to its last. That study is invented, numbers and all, so no real author’s paper is being critiqued on your behalf. The best moment for discussion is a placebo test that needed a gap of 1.14 points to flag anything, while the violation that mattered was 0.34, so the test passed but could not have failed. The last section is a discussion checklist that prints on one page from its own print button, so you can hand copies round a seminar table. Like the project guide, it says once that students should follow their course’s rules on help with set readings, including AI tools.

Problem sheets you can hand out

The practice problems page is for work on paper with a calculator. It has exam-style scenarios with numbers small enough to work by hand, each followed by a full solution that shows the formula, the substitution, the arithmetic and the reporting sentence. Every value in every step has been checked.

Two print buttons at the top of that page turn it into a handout. Print problems hides every solution, even those left open on screen, so you get a clean problem sheet. Print with solutions opens all of them for a marking key. Problems are tagged warm-up, exam-level or stretch, so you can choose a seminar's worth from one course without reading the whole library. The last set is different: six studies, figures and results sentences, each with something wrong in it, and no arithmetic to do. They work well for a seminar discussion or a take-home critique. Two sets are written as exam papers. Exam-style, a Stats 1 midterm covers §1.1 to §1.12 and Exam-style, a Stats 2 midterm covers §2.1 to §2.5. Each has eight multi-part questions to work with a calculator and the printed tables, and either one makes a revision sheet for the week before an exam. The Stats 3 set on latent variables and causal inference does the same for the advanced course, from reading a reverse-coded item through Cronbach’s α to working an encouragement design as an instrument.

Revision: quiz, mock exams and flashcards

For review weeks, the quiz has two modes. Practice mode gives instant feedback one question at a time. Exam mode sets a mock paper of 10, 20 or 40 questions over the courses you choose. It marks the paper at the end and lists every wrong answer with an explanation and a link to the course to revisit. There is an optional timer, which counts up from zero. The glossary flashcards drill every glossary term with spaced repetition. All of it is self-paced, and like the rest of the site it keeps its state only in the student's browser.

The exam results screen prints cleanly, so a student can give you a marked mock paper or keep it as a revision list.

Set a mock exam over the sections you have taught

A midterm rarely covers a whole course, so an exam can cover a course and a range of its sections. Each course in the scope picker has a from and to pair filled from its own section list, and narrowing the range narrows the paper. Every question in Stats 1, Stats 2 and Stats 3 is tagged with its section. The other courses have no tags, so they have no range control and are examined whole, as the note under the picker says.

You can also put the scope in a link. Post one URL and every student gets an exam over the same sections at the same length, with the questions drawn at random from that range. The link opens exam mode with the scope and length already set, and the student presses Start.

https://statscapybara.com/quiz.html?exam=stats-1&from=1.1&to=1.12&n=20

https://statscapybara.com/quiz.html?exam=stats-2&from=2.1&to=2.5&n=20

https://statscapybara.com/quiz.html?exam=stats-3&from=3.1&to=3.5&n=20

exam= takes one course slug, several separated by commas, or all. from= and to= take section numbers and work only when the link names a single course. You can give either one alone, and if the two ends are reversed the page swaps them. n= is 10, 20 or 40. Anything the page does not recognize is ignored and it opens on its ordinary start screen, so a typo in front of a class does no harm.

Embed a live visualization

Any lesson can render as a bare interactive widget by adding ?embed=1 to its URL. That hides the navigation, sidebar, text and footer, and leaves the section label, the title, and the interactive with its controls. It is meant for an <iframe> in your slides, notes, or LMS page.

Copy this and change the lesson path to whatever you want to show:

<iframe
  src="https://statscapybara.com/stats-1/central-limit-theorem/?embed=1"
  width="100%" height="640" loading="lazy"
  style="border:1px solid #e2e8f0;border-radius:14px"
  title="Central Limit Theorem — StatsCapybara"></iframe>

The widget has a small footer that links back to the full lesson. That link opens in the main window, not inside your frame.

Embed a specific setting, not just a lesson

A plain embed opens at the lesson's default settings, which are rarely the ones your slide needs. Several lessons accept extra parameters in the URL that set the widget before it opens: the central limit theorem at n = 50, a target correlation of .30, or a signal-detection observer with a conservative criterion. Add them next to ?embed=1, in any order.

<iframe
  src="https://statscapybara.com/stats-1/central-limit-theorem/?embed=1&n=50&pop=skewed"
  width="100%" height="640" loading="lazy"
  style="border:1px solid #e2e8f0;border-radius:14px"
  title="Central Limit Theorem at n = 50 — StatsCapybara"></iframe>

The same parameters work on the full lesson page, so you can put a configured link in your slides or notes without embedding anything. The lesson ignores anything it does not recognize, so a typo loads the ordinary default and nothing breaks in front of a class.

A value outside a listed range is moved to the nearest value the control allows, and a value between two slider steps moves to the nearer step. So a link always opens on a setting a student could have reached by hand.

In §2.9 one quantity the lecture needs is not a parameter. The Power of “Controlling For” prints δ, the slope of the confounder on the predictor, because the omitted-variable bias is b₂ × δ. With a two-group confounder, though, δ depends on both the separation and the spread of the predictor, so the page shows it as a readout and you cannot set it. A ?delta= could disagree with the number on the screen, so there is none. The link sets sep, and the page reports the δ that results. All four sign combinations can be reached by moving either slider through zero.

On a page with more than one interactive, the interactives share parameter names: n means sample size wherever it appears, so one link sets up the whole page. Each control still keeps to its own range, so ?n=4 sets the coverage playground to its minimum of 5 and the Interval Builder beside it to 4. On §1.12, mu0, sigma, n and xbar set the numbers-to-z panel, which passes its statistic to the p-value explorer below it. The parameters mu1, n and alpha set the Two Curves power picture at the bottom of the page. §1.14 is the one exception. Its two interactives both have a gap slider, but on different scales, so the Tug-of-War takes gap and the overlapping-intervals block takes sep.

Put your own numbers in the link

Four interactives read a ?data= parameter, so your own numbers can go into a live widget without anyone typing them in front of a class. Two of them take a single column. In the Mean vs. Median Playground in §1.3 the values become the draggable dots and the axis rescales to fit them, and the Descriptives calculator fills its box with them and calculates. Separate the numbers with commas, up to 200 of them.

The other two take pairs, written x:y and separated by commas, up to 100 of them. The Correlation Explorer in §1.17 replaces its simulated cloud with your points and rescales both axes to fit them. The Least-Squares Playground in §1.18 does the same and also rescales its two slope sliders so that the least-squares line is within their range. Add &resid=1 there to open the residual plot underneath. In §1.17 your points stay until someone moves the target-correlation slider or asks for a new cloud, and either one returns the widget to simulated data. In §1.18 the "New data" button does the same.

All four handle bad input the same way. If any value is not a plain number, the whole parameter is ignored, so a broken paste loads the ordinary page and never half a dataset. Pairs also need at least three points, and both columns have to vary, because otherwise the correlation and the slope are undefined.

Four more interactives take groups, under their own parameter name. Build the ANOVA Table in §2.2 reads ?g=, with commas between the values in a group and semicolons between groups, up to four groups of ten numbers. The contingency builder in §1.16 uses the same format for ?t=, and so does every widget on the site that takes group data, so your class's own three or four conditions can go straight into the table. Add ?step= to open at an earlier step, such as the sums of squares, or switch the widget to exam mode and it blanks four cells for the class to fill in.

Where the Variance Goes in §2.4 reads ?g= as a 2 × 2 design: exactly four cells, in the order a1b1, a1b2, a2b1, a2b2, all the same size. An unbalanced factorial design has no clean four-way split of the total sum of squares to draw. Loading data sets the effect-of-B slider to zero, so the widget shows your data unchanged. The interaction plot above it takes ?cells=, which is the four cell means, not four columns of data. If your slide has only the means, you can still show them live.

Contrast Builder in §2.3 reads the same ?g=, up to five groups of twelve, and adds ?c= for the contrast: one plain number per group, separated by commas. The widget ignores ?c= unless there is one number for every group and they sum to zero. So one link loads the week’s dataset and its planned comparison together, and ?mode=pairs switches the same numbers to the all-pairs panel, which shows the plain and the Bonferroni verdicts side by side. Fractions have to be written as decimals, and the sum-to-zero check allows for rounding, so -0.333333 is accepted as minus a third.

Removing the Individual Differences in §2.5 reads ?g= as within-subjects data. Each group is one condition, and the values are paired by position, so the first number in every condition belongs to the same person. Every condition must therefore have the same count, up to six conditions of twenty people, and a parameter with unequal counts is ignored completely. Your data replaces the built-in eight people. Moving either slider afterwards rebuilds the built-in study, and so does the "Back to the built-in eight" button. The partition bars and the two ANOVA tables underneath follow whatever is loaded, so you can analyze your own repeated-measures dataset both ways on the screen.

https://statscapybara.com/descriptives.html?quart=textbook&data=22,22,26,26,26,27,27,28.5,28.5,29,34,34,36,39,46,47,50,71

https://statscapybara.com/stats-1/describing-data/?data=22,22,26,26,26,27,27,28.5,28.5,29,34,34,36,39,46,47,50,71

https://statscapybara.com/stats-1/correlation/?data=2:45,4:34,6:40,8:30,10:31

https://statscapybara.com/stats-1/simple-linear-regression/?data=1:12,2:20,3:26,4:30,5:32,6:32,7:30,8:26&resid=1

The eighteen numbers are the hot-dog contest file from the practice datasets. They are few enough to work by hand, and they are clearly skewed. The calculator also takes ?quart= set to textbook, spss or r. Switching it shows a class that a paper exam and a statistics package do not compute Q1 the same way. The two paired links are the datasets behind problem 23 and problem 25: five wards with a negative correlation, and eight fertilizer doses whose residual plot is curved even though R² is .500.

Put it on the projector

Adding ?present=1 to a lesson URL gives you the same bare widget sized for a projector: larger readouts, larger control labels and tabs, and a wider column that fills a 1280 × 720 screen. The PNG download button is hidden. The link back to the full lesson stays, so one click opens the lesson text when a question needs it.

It works with everything above. Presets still apply, and you can add ?embed=1 as well if you want the larger widget inside a frame. On a narrow screen the scaling turns itself off, so if you check the link on your phone the night before, you see the ordinary lesson widget.

https://statscapybara.com/stats-1/hypothesis-testing-logic/?present=1&mu0=227&sigma=5&n=4&xbar=222&tail=two

A preset fixes the settings but not the sample. On lessons that simulate data, every visit draws a new random sample. So a scatter set to r = .30 looks a little different each time you open the slide, and the observed correlation varies around .30. That variation is the sampling variability the lesson teaches, so you can point it out to the class.

A one-semester map

The table below is one way to spread the Stats 1 and Stats 2 material over a typical fourteen-week intro course. It borrows from Stats 3 three times: effect size and power in week 8, and in weeks 10 and 13 a causal design built from a model the class has just learned. Treat it as a starting point. Merge weeks, skip the ANOVA if your syllabus leaves it for a second course, or spend two weeks on the topic your students find hardest. The five Stats 2 links carry presets, so they open on the settings that week is about. The two causal links have no preset, so each lesson opens on its default, where the design works as intended. An intro course should see that case first.

Each causal lesson builds on a model from the same week. A difference-in-differences estimate is the interaction in a two-by-two design whose factors are time and whether a unit got the policy. That makes it the same difference of differences the factorial lesson draws in week 10. A regression discontinuity is two of week 13’s regression lines, fitted on either side of a cutoff, with the gap between them read as the effect. Both lessons begin with the assumption the design depends on, so the course’s usual warning about correlation and causation gets a worked example.

Teaching research methods too? The Research Toolkit track adds full courses on Methods, Data, Writing, and Ethics, which fit a dissertation or capstone module. For a quantitative-methods sequel, ML & AI is the last course of the Statistics Core and starts where regression ends.

Use it freely

StatsCapybara is free and will stay free. There is no institutional license to sign and no seat count. You can link it, embed it, print it, and assign from it in any class, at any institution, at no cost.

It is also openly licensed, so you can do more than link to it. The teaching material is CC BY 4.0 and the code is MIT. You may translate a course, cut a lesson down to a single handout, copy a practice problem into your own exam, or fork the site and run your own version, without asking me first. The license page has the attribution line to copy.

I have one small request. If the site helps your students, a visible link back from your syllabus, course page or notes helps other teachers find it and helps keep the site running. If you build something with it, spot an error, or wish a certain tool existed, say hi. I read every message.