STATS4STEM

NSF-supported research platform

A living laboratory for statistics education.

Real classrooms, real data, and a new frontier: teachers, students and AI agents working together to understand how statistics is taught and learned.

Explore the research Collaborate with us

The platform at a glance

student answers
5 million+
students
13,000+
questions in the item bank
34,000+
in classrooms since
2015

Platform totals as of October 2026.

Teaching, learning and human–AI collaboration

Teaching

Classroom orchestration, pacing and real-time insight into where students struggle.

Learning

Step-by-step hints, tap-through worked examples, mastery tracking and immediate feedback.

Human–AI collaboration

Teachers and AI agents co-designing, checking and improving instruction — with the teacher in charge.

Inside the platform.

What students and teachers actually work with, every school day.

A tap-through worked example on a phone: a probability tree for drawing two marbles without replacement. The red-then-red path is highlighted and the student sees P(red then red) = 3/5 times 2/4 = 6/20, on step 5 of 8 with Back and Next buttons.
A second tap-through example in dark mode: a screening-test probability tree with the disease-and-positive path highlighted, showing P(disease and positive) = 0.02 times 0.95 = 0.019, on step 6 of 8.

Tap-through worked examples

Students step through a solved problem one move at a time. Each branch of a probability tree lights up as it is used, and the arithmetic appears beside it, so the reasoning unfolds at the student’s own pace.

A worked example on a phone: delivery times are approximately normal with mean 30 and standard deviation 5 minutes. Normal curves with labeled axes show the region above 40 shaded; step 1 places 40 at 2 standard deviations above the mean, step 2 fills in the empirical-rule blocks, and step 3 concludes P(X greater than 40) = 0.025.

Worked examples with pictures, and hints that walk to the answer

Worked examples build the picture one step at a time, and every question carries a ladder of hints that use that problem’s own numbers. The last hint shows the final step.

A live map of the class

Every bubble is a student, placed by how many questions they have attempted and how accurate they are, and sized by the hints they used. As the class works, teachers see at a glance who is flying, who is guessing and who needs help.

An assignment report for Topic 2.10, The Binomial Distribution, showing a class bubble chart: each bubble is a student, placed by questions attempted on the horizontal axis and percent correct on the vertical axis, sized by hints used and colored by accuracy band. Summary tiles show 26 students, 30 questions, a 73 percent class average and 2 hints per student.
Shown with demo data.

Learning analytics on written reasoning

For AP-style free-response questions, teachers see which rubric points the class earns on the first try, weakest first, and how much students recover after revising their work. Every point is a window into how students reason with statistics.

Free-response analytics for an AP Statistics question about community gardens. A bar chart ranks ten rubric points by the share of the class that earned each on the first try, from 30 percent for interpreting the confidence level to 78 percent for the interval capturing the parameter. Below, a dumbbell chart shows each point before and after Fix-It revisions, for example 36 to 52 percent for identifying the population.
Shown with demo data.

Experimenting with how students collaborate

Team games put the whole class on one shared board. A team moves forward only when its members answer correctly, so students explain, argue and check each other’s reasoning. We use these games to study how students collaborate on statistics.

The Graveyard Run projector board in mid-game: a 31-space track with climb, haunted, reaper and robber squares, five teams with statistics names such as The Residuals and Standard Deviants on a leaderboard, a 9:38 clock, and a banner announcing that The Residuals rolled a 5.
Shown with demo data.

Areas of research interest.

We pursue many strands at once — each grounded in a platform that runs in real classrooms every day.

  • Classroom orchestration

    Running live, whole-class lessons: pacing, checkpoints and the flow of a class period.

  • Real-time teacher dashboards

    Showing teachers, in the moment, where a class is stuck and who needs help.

  • Applets and video in lessons

    Integrating interactive applets and short videos into the flow of instruction.

  • Teacher decision-making

    How teachers choose what reaches students, and when, with the teacher always in charge.

  • Collaborative and game-based learning

    Whole-class team games and paired problem-solving built on practice questions.

  • Learning analytics and mastery

    Modeling skill growth from every answer, hint and attempt.

  • Formative assessment

    Immediate feedback, tiered hints and carefully engineered answer formats.

  • Simulation

    Simulation-based reasoning about chance, variation and inference.

  • Spaced review and retrieval practice

    Bringing earlier ideas back at the right moment so they stick.

  • AI feedback on written answers

    Rubric-based feedback on free-response statistics answers, with the teacher reviewing.

  • Agentic harnesses

    Multi-agent workflows that draft, review and quality-check instruction.

  • MCP servers

    Model Context Protocol tools that give AI agents safe, structured access to curriculum and data.

  • AI–human collaboration

    Teachers and AI agents co-designing lessons, with the teacher in charge.

  • AI quality assurance

    Independent multi-model review of instructional content before it reaches students.

  • Privacy-preserving AI in education

    Keeping student identities out of AI features by design.

U.S. National Science Foundation

Built with National Science Foundation support.

$3.4 million+ across four NSF awards (PI or Co-PI).

  • NSF 0937989 (opens in a new tab) Oct 2009 – Sep 2013

    STATS4STEM.ORG: Enriching STEM education through real-world data sets, computing, and statistical analysis

    National STEM Digital Library (NSDL)
    Boston Public Schools

    $149,743 Eric Simoneau, Principal Investigator
  • NSF DRL-1418163 (opens in a new tab) Sep 2014 – Feb 2020

    CodeR4STATS – Code R for AP Statistics

    ITEST / Discovery Research K-12
    Tufts University

    $563,835 Eric Simoneau, Co-Principal Investigator
  • NSF DRL-1742083 (opens in a new tab) Sep 2017 – Aug 2022

    Computing with R for Mathematical Modeling

    STEM + Computing (STEM+C)
    Concord Consortium

    $1,850,959 Eric Simoneau, Co-Principal Investigator
  • NSF OAC-2118302 (opens in a new tab) Jan 2022 – Dec 2026

    CyberTraining: Implementation: Medium: Collaborative Research: Computational and Data-Centric Ecology Training

    CyberTraining
    Clarkson University

    $920,482 Eric Simoneau, Co-Principal Investigator

Amazon Web Services previously supported this project with donated server and database resources.

Let’s study this together.

We welcome research collaborators, districts and reviewers interested in statistics education, formative assessment and human–AI collaboration.

Get in touch