The capstone for high-school science: study cards over three tiers, the tricky ones, a practice quiz at three levels — Foundation, On-Level, and Challenge — and a lab where you put a whole inquiry in order, graph a bridge team's data and find the knee in the curve, annotate a methods paragraph for its variables, controls, sample and measurement, and write the claim–evidence–reasoning for the finding, no wider than the data. Work here, or print it and use a pencil — both count.
Two groups, two averages. One is two units higher. That difference looks real.
A difference only counts when the GAP between the averages is big compared to the SCATTER within each group. A 2-unit difference with a ±4 spread is no difference — and reporting it as one is how a study gets retracted.
What people get wrong
⚠️People often think…
One plant in each group is enough to show a difference.
With n = 1 you cannot see the scatter at all, so you have no way to judge whether the gap is bigger than the noise. Two plants can differ by that much for no reason whatsoever.
n = 1 hides the scatter.
⚠️People often think…
Averaging more measurements fixes any measurement error.
Averaging kills RANDOM error, which scatters both ways. It does nothing to SYSTEMATIC error — a scale reading 2 grams heavy reads 2 grams heavy a thousand times in a row. Only calibration fixes that.
Averaging fixes random, calibration fixes bias.
⚠️People often think…
A data point that disagrees with the rest can be dropped as an outlier.
Only for a documented reason decided BEFORE you saw the results — the thermometer fell in, the sample was contaminated. Dropping a point because you do not like it is how you guarantee getting the answer you started with.
Rules for dropping points come first.
⚠️People often think…
A strong correlation shows one thing caused the other.
It shows they move together, which could be cause, reverse cause, or a third thing driving both. Correlation is a reason to RUN an experiment, not a substitute for one — only random assignment earns the word cause.
Correlation starts the experiment. It does not end it.
Ask, control, count, claim
1Watch one
Plants in the sunny window grew taller than plants in the hallway. Can you credit the light?
List every difference between the two groups, not just the one you meant to test.
The window is sunnier — and also warmer, and drier, and gets less foot traffic.
Each of those is a confound: a second difference that could explain the result.
So no — not yet. Control temperature and water, then the claim has one cause left standing. ✓
2Do one with me
Fill in what the study needs.
A second difference between the groups is called a
Averaging only removes error that is
Correlation earns you an experiment, not a claim about
💬One sentence, then you move on
Why can you not judge a difference without knowing the scatter?
3Try one
Your experiment detected no effect. What do you do with that?
I want a hint first
Knowing where the effect is NOT is information. What do you do with information?
💬Last one — then you're done here
Why must the rule for dropping a data point be set in advance?
Where this goes
Where this lives
Reading a headline about a study, judging a product claim, deciding whether a school program actually worked or just happened alongside something else.
What this feeds
That closes the science wing. Every unit before this one was practice for defending a claim you can actually support.
Name one claim you have heard that would need a control group.
One card at a time — tap “Show me” to check yourself, then Next. Start at Foundation; when those feel easy, climb.
Helpful Hints
🪜 Argue from evidence — the ladder
Start with a question a measurement can answer, or a problem with criteria and constraints → change one variable, measure one, hold the rest, with a control group → enough samples to see the spread, and a second run to catch what you missed → measure with known uncertainty, and calibrate against a standard → graph the changed thing on x and the measured thing on y, and compare the gap to the scatter → write claim, evidence and reasoning no wider than the data → iterate one change at a time; then let peer review try to break it.
Every science unit before this one handed you a finding. This one hands you the machine that makes findings — and the machine has one rule: arrange things so that the result can only mean one thing, then say exactly that much and no more.
One cause left standing; one claim, no wider.
⚠️ Tricky ones
One plant per group is an anecdote. You cannot see the spread with n = 1, so you cannot know if the gap is real.
A confound is a SECOND difference between groups. Find it before you credit the first.
Compare the GAP to the SCATTER. A 2-unit difference with ±4 spread is no difference.
Averaging fixes RANDOM error. Only calibration fixes SYSTEMATIC error.
Correlation is a reason to run an experiment, not a substitute for one.
Never drop a data point because it disagrees with you. Outliers are removed for documented reasons decided BEFORE the results.
A negative result is a result. “No effect detected” saves the next person a month.
Inside your data, a fit is a description. Outside it, it is a guess until you measure.
🧪 The parts of a study — the chart
Part
What it is
In the bridge study
The question a reviewer asks
question
testable: change, measure, conditions
does a truss beat a beam at 30 cm with 40 sticks?
could a measurement settle it?
independent variable
the one thing changed
truss vs beam
is it really only one thing?
dependent variable
the thing measured
mass at failure, grams
measured how, with what uncertainty?
controlled variables
everything held the same
sticks, glue, span, loading rate, tester
what else differed? (the confound)
sample and replication
how many, and how many runs
five of each; run twice
enough to see the spread?
result
mean ± spread, on a graph
4,800 ± 400 vs 2,100 ± 300 g
is the gap bigger than the scatter?
claim
no wider than the data
for stick bridges at 30 cm, the truss holds about twice the beam
did you test that, or just say it?
Fill the third column for your own project; the fourth column is your reviewer.
🎯 How the test will ask
A described experiment — “identify the independent, dependent and controlled variables.”
A flawed design — “name the flaw and fix it.” (Usually a confound or n = 1.)
A data table — “graph it, choosing the right kind, and state the trend.”
Two means with spreads — “is the difference meaningful? Explain.”
A conclusion that overreaches — “rewrite the claim to match the evidence.”
An engineering brief — “state the criteria and constraints, propose a design, and describe one iteration.”
✅ Can you…
Turn a vague question into a testable one?
Name all three kinds of variable in any study, and spot a confound?
Say why n matters and what replication catches?
Compare a gap to its scatter and decide honestly?
Write a CER whose claim is exactly as wide as the data?
Review a classmate’s study with six questions and turn every no into a next step?
Pick your level
Look back at anything you missed — the hint that appeared is exactly what to reread tonight.
How sure did you feel?
Lab
Work like a scientist: label the diagrams, turn the table into a graph, then write a claim with evidence and reasoning. Every lab checks itself on the spot, and hints are free.
Your practice never leaves this device. There is no account and no sign-in. Your work is saved in this browser only, and you can erase it whenever you want.
Your practice record — saved on this device
This is your record of the module on screen — it stays here and goes nowhere. Independent means you got it right on the first tap; supported means you got it after the explain-and-retry, or marked ‘I had it’ on a revealed answer. Both count, and neither is a grade. If your teacher asks, copy the row or show them this screen.
Date
—
Topic
—
Level
—
Independent
0 of 0
Supported
0 of 0
Hints used
0
Cards studied
0
Lab
—
Traps met
0
Explained in own words
0
Confidence
—
The answer key is for a teacher: it prints only from here, for the unit on screen. Print the study packet prints the study pages and a blank quiz — never the answers.