Expert Experiments

Measure, doubt, explain: how we design learning experiences

Why every experiment starts with a hunch, why we show students the messy data, and what a book about data visualisation taught us about earning a learner's trust.

Three students lean over a table, studying a sensor wire and a phone graph

How thick is one sheet of paper? Most students have never asked. Hand them a ruler and the question suddenly gets interesting: the smallest tick is a millimetre, and the paper is far thinner than that. So what now?

That moment, when the obvious method fails and you have to think, is what we design for. This post explains the ideas behind every experiment on Expert Experiments, and where some of them come from. A surprising number came from a book about data visualisation: R. J. Andrews' Info We Trust.1 It isn't a book about physics teaching, but it describes, better than most teaching guides, how people turn measurements into understanding.

The loop we build around

Andrews organises his book around a loop. The world is measured and becomes data. Data turns into information through repeated back-and-forth between what the numbers say and how they are shown. That information then changes how we see the world, and the loop starts again.1 A good school experiment follows the same path, so we use our own version of it:

The learning loop Four steps in a row: Predict, Measure, Explore, Explain. An arrow returns from Explain to Predict, labelled "a better question". 01Predict 02Measure 03Explore 04Explain a better question
Our learning loop, adapted from the measure-and-inform cycle in Info We Trust.1 The last step feeds the first: a good explanation always raises a new question.

Each step below is one idea we hold ourselves to, and what it looks like on the platform.

1. Start with a hunch

A number on its own means very little. Andrews keeps returning to one question: compared to what? We notice differences from what we expected, not values in isolation.1 A student who measures a pendulum's period as 2.01 seconds has learned a number. A student who first guessed it would be 4 seconds, because the string is four times longer, has learned something about the world, and will remember it.

That's why our experiments ask for an estimate before the measurement, and why the first thing on our home page is a question rather than a feature list. We ask students to put their reasoning into words, not just a number into a box. A wrong prediction isn't a failure. It's what makes the result worth having.

The interesting part of an experiment is the gap between what you expected and what you measured.

2. A measurement is a recording, not the truth

Andrews compares a dataset to a live recording of a concert: a faithful trace of something real, but compressed. Some things were kept, others were lost, and someone made choices along the way.1 The same is true of every reading a student takes, whether it comes from a ruler, a phone's accelerometer or a sensor pod on the Hepek.

So we don't hide how a number was made. Students pick the sensor, can see and change how many readings it takes per second, and every column carries its unit. Back to the sheet of paper: once a student realises a single sheet is below the ruler's resolution, the natural next step is to measure a stack of a hundred and divide. That is a real experimental idea, and they came up with it because they understood the limits of the instrument.

3. Look at the data before the average

The statistician John Tukey argued that understanding data begins with many quick, rough plots, each one prompting the next question.2 Andrews builds on that idea and adds a warning: summaries hide things. A mean can conceal an outlier, a wide spread or a sample that was simply too small.1 The classic demonstration is Anscombe's quartet: four small datasets with almost identical averages and correlations that look completely different once you plot them.3

On our platform, every recording becomes a graph straight away. Students can drag across part of it to measure just that section, overlay several runs, and try different curves to see which one fits. The raw points stay visible underneath the fitted line. When a free-fall graph isn't quite a parabola, the interesting physics, like air resistance or a clumsy release, is usually hiding in that gap.

Students around a table with calculators, phones and a hand pump, working out their results
Students at one of our STEM camps in Bosnia and Herzegovina comparing measurements. Different groups, different methods, plenty to argue about.

4. Uncertainty is part of the answer

Every measurement is really a range. When a reading falls between two ruler ticks, the honest answer includes how far between. Andrews points out a quirk of how we read charts: a precise result has a small error bar, so it attracts less attention than a sloppy one with a big bar.1 He cites the statistician Howard Wainer's three duties for any display of uncertainty: remind people it exists, show how large it is compared with the claim, and help them avoid drawing the wrong conclusion.1

We take that seriously. When students fit a line on the platform, they get the slope with its uncertainty, not a bare number. Answers to numerical questions are checked against a tolerance, because "9.7" and "9.81" can both be good measurements of gravity. And we encourage students to state predictions that could turn out wrong. Science advances by surviving attempts to disprove it, not by being declared correct.

5. Explain it to someone else

Some of the most useful ideas in Info We Trust come from outside data visualisation. From museums: visitors look at the object, then read the label, then look at the object again, and see more the second time. From storytelling: a good explanation is a chain of questions, each one answered just as the audience starts to ask it.1

Our guides follow that pattern. Students play with the real signal first, read the theory second, then return to their data with new eyes. Each step poses a question the next measurement can answer. Hints point to the next step without doing the thinking for them. At the end, students explain what they found in their own words, and teachers respond to the reasoning, not just the score.

6. Trust is earned

The last part of Andrews' book is about trust. People aren't fooled by bad design so much as put off by it: one wrong number, one confusing chart, and they stop believing the rest. He lists what earns trust: getting things right, being open about sources and limits, following conventions, standing behind your work, and respecting the audience's intelligence.1

For us, the audience is a fourteen-year-old with a phone and a pendulum. Respecting their intelligence means not hiding the messy data, not pretending a measurement is exact, and not giving them the answer before they've had a chance to find it. When we get that right, students stop asking "is this correct?" and start asking "how do I know?" That second question is what science is.

References

  1. R. J. Andrews, Info We Trust: How to Inspire the World with Data. Hoboken, NJ: Wiley, 2019.
  2. J. W. Tukey, Exploratory Data Analysis. Reading, MA: Addison-Wesley, 1977.
  3. F. J. Anscombe, "Graphs in Statistical Analysis," The American Statistician, vol. 27, no. 1, pp. 17–21, 1973.

Try it yourself

Every guide on the platform is free to read. Pick one, make a prediction, and see what your measurements say.

Explore experiments

More from the blog

From “wow” to evidence: when a demonstration becomes an experiment

A striking effect can spark curiosity. Measurement, analysis and explanation turn that moment into a scientific experience students can own.

← All posts