Linear regression classroom activity: arm span

By Jude Wallis · Published

Students measure height and arm span in centimeters, build a class scatterplot, fit the least-squares regression line, and interpret its slope, intercept, residuals and r-squared in context. Arm span is close to height for many people, so expect a strong, positive, roughly linear pattern.

AP Statistics: Unit 5 (topics 5.1 Graphical Representations Between Two Quantitative Variables, 5.2 Correlation, 5.3 Linear Regression Models, 5.4 Residuals, 5.5 Least-Squares Regression). The activity runs through Unit 5 in order: a scatterplot of class data (5.1), correlation (5.2), predictions from a linear model (5.3), residuals and a residual plot (5.4), and the least-squares line with its slope, intercept and r-squared interpreted in context (5.5).

What the activity shows

Regression vocabulary sticks when the points are the students. Each student is one dot on the scatterplot, has a predicted arm span from the class line, and has a residual that says whether their arms are longer or shorter than the line predicts for their height.

The activity covers the full sequence in one period: describe a scatterplot, fit the least-squares line, interpret the slope and intercept in context, compute and plot residuals, and interpret r-squared. It works as an introduction to the unit or as a review of it.

Body measurements are personal. Collect them on unsigned slips, let any student opt out, and consider the alternative pair in the questions at the end of this page: hand span and forearm length.

Materials and timing

  • Meter sticks or measuring tapes, at least one per group of four.
  • Masking tape to fix a tape vertically on a wall for height and horizontally at shoulder level for arm span.
  • One blank slip of paper per student.
  • A large set of axes on the board or chart paper: height in centimeters across the bottom, arm span in centimeters up the side, both running from about 140 to 200.
  • Calculators with two-variable statistics, or a screen showing the regression calculator.

Plan on 50 minutes: 12 to measure, 8 to plot, 5 to describe the scatterplot, 8 to fit the line, 12 to interpret the line and residuals, and 5 to discuss. In a shorter period, measure the day before.

Steps

  1. In groups, students measure each other. Height: shoes off, back to the wall, read to the nearest centimeter. Arm span: arms straight out to the sides at shoulder level against the wall, fingertip to fingertip, to the nearest centimeter.
  2. Each student writes the pair (height, arm span) on a slip with no name and hands it in.
  3. Shuffle the slips and hand them back out so each student plots someone else's point on the class axes.
  4. Students describe the scatterplot in writing: direction, form, strength, and any unusual points, in context.
  5. Before computing anything, a volunteer holds a meter stick against the plot where a line fits best. Mark it lightly.
  6. Enter the data and find the least-squares regression line, the correlation r, and r-squared. Write the equation on the board with the variables named, in the form predicted arm span = a + b(height).
  7. Compute the mean height xˉ\bar{x} and mean arm span yˉ\bar{y}, plot that point, plot one more predicted point from the equation, and draw the least-squares line through both.
  8. Interpret the slope and the intercept in context as a class.
  9. Each student takes the slip they were handed, computes the predicted arm span and the residual, actual minus predicted, and states what the sign means.
  10. Plot residuals against height on a second set of axes with a horizontal line at 0, and decide whether a line is an appropriate model.
  11. Interpret r-squared in context.

What students should see

The exact line depends on your class, so there is no theoretical slope to match. Arm span is close to height for many people, so the usual picture is a strong, positive, roughly linear association with points near the line where arm span equals height. Several facts hold for every data set and are worth checking with yours:

  • The least-squares line passes through the point (xˉ,yˉ)(\bar{x}, \bar{y}).
  • The residuals add to 0, apart from rounding.
  • The slope equals r⋅sy/sxr \cdot s_y / s_x, where sxs_x and sys_y are the standard deviations of height and arm span.
  • The correlation r has no units and does not change if both measurements are converted to inches.

The table below is example data, invented for this guide and not measured from a real class. Use it to rehearse the analysis or to give absent students something to work with.

StudentHeight (cm)Arm span (cm)Predicted arm span (cm)Residual (cm)
1152150151.25-1.25
2157159156.672.33
3160158159.92-1.92
4163166163.172.83
5165163165.34-2.34
6168171168.592.41
7170168170.76-2.76
8173176174.011.99
9178177179.43-2.43
10183186184.851.15

For these ten points, xˉ=166.9\bar{x} = 166.9 cm, yˉ=167.4\bar{y} = 167.4 cm, sx≈9.53s_x \approx 9.53 cm, sy≈10.59s_y \approx 10.59 cm, and r≈0.976r \approx 0.976. The least-squares line is y^=−13.50+1.084x\hat{y} = -13.50 + 1.084x, with r2≈0.952r^2 \approx 0.952 and a standard deviation of the residuals s≈2.47s \approx 2.47 cm.

Slope. For each additional centimeter of height, the predicted arm span increases by about 1.08 cm.

Intercept. The line predicts an arm span of -13.5 cm for a height of 0 cm. That has no meaning. The heights in the data run from 152 to 183 cm, and 0 is far outside that range.

r-squared. About 95.2% of the variation in arm span is accounted for by the linear relationship with height.

Residual plot. The ten residuals are scattered above and below 0 with no curve and no change in spread, so a line is a reasonable model for these example data.

Discussion questions

  1. What does it mean, in terms of a person's body, to have a positive residual? Their arm span is longer than the line predicts for someone of their height.
  2. We used height to predict arm span. Would the line be the same if we used arm span to predict height? No. Least squares minimizes vertical distances, so switching the variables gives a different line.
  3. Would you use our line to predict the arm span of a two-year-old who is 85 cm tall? No. That is extrapolation far below the heights we measured, and nothing in our data says the pattern continues there.
  4. Does the strong association mean that growing taller causes arms to get longer? The data are observational. Height and arm span both reflect overall body size and growth, so the association does not show that one causes the other.
  5. Compare the volunteer's line with the least-squares line. What exactly makes the least-squares line the best one? It has the smallest possible sum of squared residuals.
  6. How would the correlation change if we had measured in inches? It would not change.

Extension

One unusual point. Add an invented eleventh person to the example data: 190 cm tall with an arm span of 172 cm. The line changes from y^=−13.50+1.084x\hat{y} = -13.50 + 1.084x to y^=40.35+0.754x\hat{y} = 40.35 + 0.754x, the correlation falls from 0.976 to 0.849, and the standard deviation of the residuals rises from 2.47 to 5.65 cm. The new point's own residual is −11.66-11.66 cm. Its height is farther from the mean height than any other, which is why it can pull the slope so much. Have students make the same test on the class data by adding one made-up point at the far right and refitting.

See it move. The influential point explorer lets students drag one point and watch the line, r and r-squared respond.

A second pair of variables. Collect hand span and forearm length, or height and shoe length, and compare the strength of the two associations using r-squared. For help with the plot from step 10, see how to interpret residual plots.

Fitting the line from summary statistics

Example data, not from a real class. Ten students have mean height 166.9 cm with standard deviation 9.527 cm, mean arm span 167.4 cm with standard deviation 10.585 cm, and correlation 0.97553. Find the least-squares line for predicting arm span from height.

  1. Slope: b=r⋅sy/sx=0.97553(10.585)/9.527≈1.0839b = r \cdot s_y / s_x = 0.97553(10.585)/9.527 \approx 1.0839.

  2. The line passes through (xˉ,yˉ)(\bar{x}, \bar{y}), so the intercept is a=yˉ−bxˉ=167.4−1.0839(166.9)a = \bar{y} - b\bar{x} = 167.4 - 1.0839(166.9).

  3. Compute: 1.0839(166.9)≈180.901.0839(166.9) \approx 180.90, so a≈167.4−180.90=−13.50a \approx 167.4 - 180.90 = -13.50.

  4. Write the line with the variables defined: y^=−13.50+1.084x\hat{y} = -13.50 + 1.084x, where xx is height in cm and y^\hat{y} is predicted arm span in cm.

Predicted arm span = -13.50 + 1.084(height), both in centimeters. Each additional centimeter of height goes with about 1.08 more centimeters of predicted arm span.

A prediction and a residual

Using the example line, find the predicted arm span and residual for student 4, who is 163 cm tall with an arm span of 166 cm. Then predict the arm span of a student who is 175 cm tall.

  1. Predicted arm span for student 4, using unrounded coefficients: −13.495+1.08385(163)≈163.17-13.495 + 1.08385(163) \approx 163.17 cm.

  2. Residual: actual minus predicted, 166−163.17=2.83166 - 163.17 = 2.83 cm.

  3. Interpret the residual: this student's arm span is 2.83 cm longer than the line predicts for a height of 163 cm.

  4. Prediction at 175 cm: −13.495+1.08385(175)≈176.18-13.495 + 1.08385(175) \approx 176.18 cm. The value 175 is inside the range of heights in the data, 152 to 183 cm, so this is not extrapolation.

  5. Compare with the typical prediction error: the standard deviation of the residuals is 2.47 cm, so a residual of 2.83 cm is ordinary for these data.

Student 4 has a predicted arm span of about 163.2 cm and a residual of about 2.8 cm. The predicted arm span at a height of 175 cm is about 176.2 cm. Rounding the coefficients first to -13.50 and 1.084 shifts these answers by a few hundredths of a centimeter.

Frequently asked questions

Some students do not want to be measured. What then?

Let them opt out without explanation and give them a row from the example table to work through at their desk, keeping invented points off the class plot. Unsigned slips and the shuffle in step 3 keep any point from being tied to a person. You can also switch the whole class to hand span and forearm length, which most students find less personal than height.

Should the line be forced through the origin?

No. The least-squares line is fitted to the data you have, and its intercept is whatever minimizes the squared residuals. An intercept that makes no physical sense is a reminder that the line describes the range of heights you measured.

Does it matter which variable goes on the horizontal axis?

Yes for the line, no for the correlation. Put the variable you want to predict from on the horizontal axis. Either choice is defensible here, so have the class decide and stick with it. Swapping them gives the same r and a different line.

Can we measure in inches?

Yes. The correlation and r-squared are the same in either unit. If both variables are converted the same way, the slope is also unchanged, and the intercept and the residuals are converted along with the data.

My class is small. Is ten points enough?

Ten points are enough to fit and interpret a line, as the example shows. With few points, one unusual student can move the line a great deal, so do the extension and combine sections if you can.