Sampling Distribution Slider: A 40-Minute Lesson

Objective and materials

By the end of the period, students can state which of three controls on the interactive, population shape, sample size, or number of samples drawn, actually changes the spread of a sampling distribution, and can explain why drawing more samples smooths a histogram without narrowing it.

• A shared screen or projector open to the sampling distribution and CLT visualizer, left untouched at its default settings until Move 1
• One notecard or half sheet per student for written predictions
• A whiteboard, or one shared document, where the class prediction gets recorded before every reveal
• Optional for the no-tech variant: one six-sided die per student or pair, and a number line taped to the board from 1 to 6

Nothing needs to be built ahead of time. The tool already opens with the right-skewed population selected, sample size n set to 10, and no samples drawn yet, so the empty bottom plot is itself the hook: a blank histogram waiting to be filled in.

The hook, 4 minutes

Point at the two stacked plots before touching anything. The top one shows a right-skewed population that piles up near zero and thins into a long right tail, with a dashed line at its mean. The bottom one is empty. Ask the class: if you draw one random sample of 10 values from that population and average them, where will that single average land, close to the population's dashed mean, or could it land almost anywhere?

Take a quick show of hands for "close to the mean" versus "could be almost anywhere," then click draw 1 sample. Ten dots rain onto the population plot one at a time, landing exactly where the theory says they should since they were drawn from that population's own shape, and a single larger dot marks their mean, labeled with its value. Ask which side of the show of hands got it right, and move straight into Move 1 without resolving the general case yet.

Move 1: one sample is not the story, 4 minutes

Before clicking anything else, ask students to write a private prediction: if you clicked draw 1 sample nine more times, would the ten resulting dots in the bottom plot land in a neat pile, or scatter widely with no visible pattern yet?

Click draw 1 sample nine more times, one click at a time, pausing on each so the class can watch a single new mean rain in. Have them watch above the plot too: an observed SD of those means appears as soon as the second mean lands, since the tool needs at least two accumulated means to compute one, and by the time all ten have landed it has had time to settle. After ten total clicks, the bottom plot holds ten dots with some shape starting to suggest itself, but not yet a clean pile. Ask the class to compare their prediction against what actually happened: the sample count now reads 10 samples drawn. Name the real lesson out loud: a single sample mean tells you almost nothing on its own, which is exactly why the next two moves are about drawing thousands of them, not one.

Move 2: shrink the sample size, 8 minutes

Click reset to clear the ten dots. State the change coming next before making it: sample size n is about to drop from 10 down to 2, staying on the same right-skewed population. Ask every student to write a prediction: after drawing 2,000 samples of size 2, will the histogram of their means look symmetric like a bell, or will it still lean to one side the way the population does?

Drag the n slider down to 2, which silently clears the plot again since changing n always resets the count to zero. Click draw 2,000. The histogram fills in fast, and it visibly leans right, the same direction as the population above it. Ask students to read the two numbers printed above the plot: predicted SE, sigma over the square root of n, now shows about 1.288, next to whatever the observed SD of the 2,000 means actually came out to. Push on the gap between the two readouts and the shape: the SE formula already predicts this spread correctly at n equals 2, but nothing in that formula promises a bell curve, and the histogram is the proof that a small n does not deliver one from a skewed population.

Move 3: quadruple the sample size, 10 minutes

Before touching the slider, tell the class exactly what is coming: n is about to go from 10 up to 40, four times as large, same right-skewed population. Ask for a written prediction of what happens to the printed predicted SE: does it get four times smaller, twice as small, or does it barely move?

Reset, then drag n to 10 and click draw 2,000 to reestablish a baseline: predicted SE reads about 0.576. Reset again, drag n to 40, and click draw 2,000 once more. Predicted SE now reads about 0.288, exactly half of 0.576, not a quarter of it. Also point at the histogram's shape, which now sits close to the indigo curve overlaid on it rather than leaning to one side. Have the class state the rule in their own words before you say it: quadrupling n cuts the spread in half, because the spread scales by the square root of n, and a bigger n also gives a skewed population's sampling distribution enough room to round into something close to normal.

Move 4: change how many samples you draw, 6 minutes

Keep n fixed at 40 for this move so only one thing changes at a time. Ask students to predict: will clicking draw 2,000 a second time, on top of the samples already there, change the predicted SE reading at all?

Without resetting, click draw 2,000 again. The printed sample count jumps by 2,000, the histogram gets visibly smoother and fuller, and the observed SD of means creeps even closer to the predicted 0.288, but the predicted SE line itself does not move, because it was never a function of how many samples you have drawn. Name the distinction plainly: sample size n changes the actual spread of the sampling distribution, while the number of samples drawn only changes how clearly that fixed spread shows up in your histogram. A researcher who runs one more copy of the same study learns more about the truth without changing what the truth's variability actually is.

Check for understanding, 4 minutes

Ask the class, in writing, then cold-call: according to the formula printed on screen, sigma over the square root of n, which single change would cut the predicted SE in half, doubling n, quadrupling n, or doubling the number of samples drawn? The answer is quadrupling n, since the square root of 4 is 2.

Follow with a quick verbal check: does clicking draw 100 a second time ever change the predicted SE number on screen? No, and the reason is worth hearing a student say aloud: predicted SE depends only on the population's spread and on n, never on how many times you have already drawn a sample. For a class that finishes early, point at the built-in challenge chip labeled make 2,000+ skewed samples come out normal, which only checks off for the right-skewed population at n of 30 or higher, and ask a pair to find the smallest n where it lights up.

Exit ticket, 4 minutes

Hand out or project these three prompts and collect written answers before the bell.

1. Quadrupling the sample size divides the standard error of a sampling distribution by about what number? Show the square root that gets you there.
2. True or false, with one sentence of justification: a sampling distribution built from a skewed population will still look skewed no matter how large the sample size gets.
3. In your own words, what is the difference between the distribution of the ten values inside one sample and the sampling distribution of the mean built from thousands of samples?

The first answer is 2, from the square root of 4. The second is false: the population itself always stays skewed, but the sampling distribution of its mean straightens toward a bell shape as n grows, which is the Central Limit Theorem's actual claim. The third answer should separate the ten raw values a single sample happened to contain from the pile of means built by repeating the sampling process thousands of times, since those are two different distributions with two different jobs.

No-tech variant: dice and a class survey

Give every student, or every pair, one six-sided die. A single die roll is a draw from a flat population running 1 through 6, with a true mean of 3.5. Have each student roll once and call out their number: tallied on a number line taped to the board, thirty rolls make a rough, flat-looking population histogram, which stands in for the right-skewed population's top plot.

Now have each student roll the die three times and record the average of their three rolls, rounded to the nearest tenth, then come place a sticky note above that value on a second number line. Even though a single die's outcomes are flat, the class's pile of thirty averages of three rolls will mound up visibly near 3.5 and thin out toward the extremes, the same rounding-toward-the-middle effect the slider shows for n equals 2 versus n equals 40. Repeat with an average of five rolls if time allows, and the class's sticky notes should mound up more tightly around 3.5 than the average-of-three round did.

For a right-skewed no-tech population, run a quick anonymous class survey instead of dice: minutes of screen time yesterday, or number of text messages sent this morning, both of which tend to pile up low with a long tail of high values. Split the survey responses into groups of three or four, compute each group's average, and plot those group averages on a number line the same way as the dice averages. The two physical activities and the on-screen tool are teaching the identical fact: averaging shrinks spread and, given enough values per average, straightens out a lopsided population.

Misconceptions this lesson targets

A sampling distribution is not the distribution inside one sample. The distribution of the ten raw values a single sample happened to contain describes that one sample. The sampling distribution is a different object entirely: it is built by repeating the whole sampling process thousands of times and plotting the resulting means, one dot per sample, never one dot per raw value. Students who mix these up will describe the top population plot, the one sample's rained-in dots, and the bottom histogram of means as though they were three versions of the same thing, when they are three distinct distributions doing three distinct jobs.

A bigger sample size shrinks spread by a square root, not by the same factor. Doubling n does not halve the spread, and quadrupling n does. The predicted SE readout on screen makes this checkable in real time: sigma over the square root of n, so spread only falls as fast as n's square root grows. A student who expects doubling n to halve the spread will be consistently surprised by the numbers the tool prints, which is exactly the moment to have them recompute the formula by hand.

The Central Limit Theorem describes the sample mean, never the raw data. No matter how many samples get drawn, the individual values inside any one sample still follow the parent population's own shape, skewed or flat or bimodal, and that never changes. What straightens toward a bell curve is the distribution of the means, and only as n grows large enough. A student who says drawing more samples makes each sample's own data look more normal has confused the two plots on screen; the fix is pointing back at the top plot, which never changes shape no matter what n or the draw count are doing.