Normal distribution classroom activity: ruler drop
By Jude Wallis · Published
Students catch a dropped ruler five times each and record the catch distance in centimeters. The class plots all the distances, finds the mean and standard deviation, checks the counts within 1, 2 and 3 standard deviations against 68%, 95% and 99.7%, and uses z-scores to locate single catches.
AP Statistics: Unit 2 (topics 1.9 Comparisons of the Distributions for One Quantitative Variable, 2.11 The Normal Distribution). Students describe a distribution of class measurements and use z-scores to locate individual values (topic 1.9), then judge whether a normal model fits before using it to find proportions (topic 2.11).
What the activity shows
A normal distribution is a model. It earns its place when a real set of measurements is close enough to it that the model's percentages can be trusted. This activity has students collect a measurement, judge that closeness with a specific check, and then use the model.
The measurement is how far a ruler falls before a student catches it. Thirty students making five catches each give 150 values in about ten minutes, enough for a histogram with a visible shape. Whether that shape is close to normal is a question the class answers from its own data. This guide does not promise that it will be.
Run it after students know the 68-95-99.7 rule and the z-score formula.
Materials and timing
- One 30 cm ruler per pair.
- A recording sheet with five blanks per student.
- A number line on the board from 0 to 30 cm for a class dotplot, or axes for a histogram with bins 2 cm wide.
- Calculators with one-variable statistics, or a screen showing the standard deviation calculator.
Plan on 45 minutes: 5 to demonstrate, 12 to collect, 8 to plot, 10 to compute and count, 5 for z-scores, and 5 to discuss.
Steps
- Demonstrate with one pair. The dropper holds the ruler vertically by the 30 cm end. The catcher rests a forearm on a desk with the hand off the edge, thumb and forefinger about 2 cm apart, level with the 0 cm mark and not touching the ruler.
- The dropper lets go without warning. The catcher pinches the ruler. Read the centimeter mark at the top edge of the thumb, to the nearest half centimeter. That is the catch distance.
- Each student takes one practice catch, then five recorded catches, and then the partners switch. If the ruler falls through uncaught, do not record a number. Repeat the catch.
- Students add all five of their distances to the class dotplot or histogram.
- The class describes the shape, center and spread of the distribution, and notes any gaps or unusual values.
- Compute the mean and standard deviation of all the class distances. Enter them once on a shared screen as pairs read their values aloud. Do not have every student key in 150 numbers.
- Mark on the plot, then the values 1, 2 and 3 standard deviations on each side of it.
- Count the catches within 1, within 2 and within 3 standard deviations of the mean. Convert each count to a percent of the total.
- Compare the three percents with 68, 95 and 99.7 and decide whether a normal model is reasonable for these data.
- Each student picks one of their own catches and computes its z-score, , and says in a sentence where that catch sits in the class distribution.
What students should see
What a normal model predicts. To two decimals, a normal distribution has 68.27% of its values within 1 standard deviation of the mean, 95.45% within 2, and 99.73% within 3. For 150 catches those percentages come to about 102, 143 and 150 catches. For 120 catches they come to about 82, 115 and 120.
How close is close enough. Real counts will not match exactly. A count within a handful of the predicted one is what random variation looks like. Reasons to doubt the normal model are specific: a histogram with a long tail on one side, two separate peaks, or a count within 1 standard deviation that is far from 68%, such as 80%.
What often goes wrong, and why it is useful. A catcher who loses focus produces a long catch, and nobody can catch at a distance below zero, so a tail on the high side is possible. If the class sees one, the right conclusion is that the normal model fits poorly on that side and its percentages should not be trusted there. That is a successful outcome for the activity.
What z-scores do. A z-score states how many standard deviations a catch is from the class mean, and in which direction. A shorter distance is a faster catch, so here a negative z-score is the faster side. If every catch in the class is converted, the z-scores have mean 0 and standard deviation 1, and the shape of the distribution is unchanged. Standardizing does not make data normal.
Using the model. If the normal model is reasonable, it turns a z-score into a percentile. A catch with is shorter than all but about 6.7% of catches under the model. Students can check a model percentage against the actual count in their own data, as in the worked example below.
Discussion questions
- Describe the shape of our distribution. Is it roughly symmetric? Is there one peak?
- Our counts within 1, 2 and 3 standard deviations were not exactly 68%, 95% and 99.7%. Does that mean the normal model is wrong for our data? Small differences are expected in any real sample. Look at the size of the gaps and at the shape together.
- Your catch has a z-score of -1.5. Say what that means without using the word z-score. The catch was 1.5 standard deviations shorter than the class mean distance, so faster than most.
- A z-score of 2.5 would be unusual. What might have happened on that catch?
- Each student contributed five catches. Are the 150 values 150 independent pieces of information about the class? Not entirely. One student's five catches tend to resemble each other. That does not stop us from describing the distribution, and it would matter if we tried to do inference with these values.
- If we measured in inches, would any z-score change? No. The mean and standard deviation convert along with every value, so each catch stays the same number of standard deviations from the mean.
Extension
From distance to reaction time. A dropped object falls a distance , so the time for a catch at centimeters is seconds, using cm per second squared. A 17 cm catch corresponds to about 0.186 seconds. Catches of 13 cm and 21 cm correspond to about 0.163 and 0.207 seconds.
Have students notice that 13 and 21 are equally far from 17 in distance, but 0.163 and 0.207 are not equally far from 0.186 in time: the gaps are 0.023 and 0.021 seconds. Converting centimeters to inches multiplies every value by the same number and leaves z-scores alone. Converting distance to time uses a square root, which changes the shape of the distribution and therefore the z-scores.
Compare two conditions. Repeat with the other hand and compute z-scores for the new catches using the first distribution's mean and standard deviation. Students can then say how the other hand compares in standard deviations.
Tools. Enter the class mean and standard deviation into the normal distribution calculator to get the model's proportion below any cutoff. The z-score calculator converts a single catch to a z-score and percentile. The normal curve explorer shows why the area between two z-scores stays the same when the mean and standard deviation change.
Checking example data against the 68-95-99.7 rule
Example data, not from a real class: 150 catch distances have mean 17.0 cm and standard deviation 4.0 cm. There are 104 catches from 13 to 21 cm, 143 from 9 to 25 cm, and all 150 from 5 to 29 cm. Decide whether a normal model is reasonable.
Find the three intervals: is 13 to 21 cm, is 9 to 25 cm, and is 5 to 29 cm.
Convert the counts to percents: , , and .
Find what a normal model predicts for 150 values: , , and .
Compare: 104 against 102, 143 against 143, and 150 against 150. The largest gap is 2 catches.
Check the shape as well. Counts alone can match for a distribution that is visibly lopsided, so look at the histogram before deciding.
The percents within 1, 2 and 3 standard deviations are 69.3%, 95.3% and 100%, close to 68%, 95% and 99.7%. If the histogram is also single-peaked and roughly symmetric, a normal model with mean 17.0 cm and standard deviation 4.0 cm is reasonable for these example data.
A z-score and a model percentage
Using the same example distribution, with mean 17.0 cm and standard deviation 4.0 cm, find the z-score of an 11 cm catch and interpret it. Then use the normal model to predict how many of the 150 catches are shorter than 12 cm.
z-score of the 11 cm catch: .
Interpret: the catch is 1.5 standard deviations shorter than the mean distance.
Under the normal model, the proportion below is about 0.0668. About 6.7% of catches are shorter, so this catch is faster than about 93% of them.
For the 12 cm cutoff: , and the proportion below is about 0.1056.
Predicted count: , so about 16 catches.
Compare with the data. Suppose the example class counted 14 catches shorter than 12 cm. That is close to the predicted 16.
The 11 cm catch has a z-score of -1.5 and sits at about the 7th percentile of distance, among the fastest catches. The model predicts about 16 of 150 catches shorter than 12 cm, and a count of 14 in the example data is consistent with that.
Frequently asked questions
What if our histogram does not look normal?
Then the class has learned what the check is for. Have students say what feature breaks the model, such as a long tail on the high side, and where the model's percentages would be wrong because of it. They can still compute z-scores, which describe position in any distribution, but they should not convert them to normal percentiles.
We have no rulers. Is there a version with known values?
Yes. Each student rolls five dice and records the total, ten times. The total has mean 17.5 and standard deviation about 3.82, and its distribution is symmetric and mound-shaped. The exact chance that a total is within 1 standard deviation of the mean, from 14 to 21, is 69.6%, and within 2, from 10 to 25, it is 96.8%.
Why does each student make five catches?
Thirty values make a ragged histogram whose shape is hard to judge. Five catches each give about 150 values in the same class time. The cost is that one student's catches are related to each other, which the fifth discussion question addresses.
Should the standard deviation divide by n or by n - 1?
Use the sample standard deviation your calculator labels Sx, which divides by n - 1. With 150 values the two versions differ by about a third of one percent, far too little to change any count in the activity.
What do we do with catches where the ruler fell through?
Repeat them, as in step 3, and tell students that this choice removes the slowest reactions from the data. A 30 cm ruler takes about a quarter of a second to fall its full length, so the data can only describe catches faster than that.