01 · Learn the idea
Start with paired observations
Use a balance to measure total mass for a cup containing 0,2,4,6,8 identical washers, or use a teacher-provided record of an actual measurement session. Keep the cup and washer type fixed. Record counts and masses with units and instrument precision; do not substitute invented results for measurements. Plot the paired data.
Draw a reasonable line by hand through the trend, not necessarily every point. Choose two points on that line to calculate its equation. The intercept can represent the empty cup, and slope the added mass per washer. Compare this interpretation with the physical setup.
02 · Build understanding
Calculate and inspect the fit
Use a graphing calculator or spreadsheet to fit y=mx+b and calculate Pearson correlation r. In Desmos, enter a table and select linear regression, or enter y_1 ~ m x_1 + b; record m,b,r and residuals. Tool guide: https://help.desmos.com/hc/en-us/articles/4406972958733-Regressions . A residual is observed minus predicted output.
For an explicitly invented arithmetic check, x=0,1,2,3,4 and y=1,3,4,7,10 give a least-squares line y=2.2x+0.6 and r≈0.984. At x=2.5 it predicts 6.1. r near 1 indicates strong positive linear association, not that every point lies exactly on the line.
03 · Build understanding
A strong pattern does not establish every causal story
Correlation lies from −1 to 1 when defined. Values near −1 indicate strong negative linear association; values near zero indicate weak linear association, possibly with a nonlinear pattern. Correlation is undefined if one variable has no variation.
If warmer days coincide with more cold-drink sales and more swimming, sales and swimming may be associated because temperature influences both. Correlation alone does not show that buying a drink causes swimming. The controlled washer setup gives more causal information about added mass than a casual comparison of unrelated observations.
04 · Build understanding
Worked example: a fitted line for a burning candle
These numbers are invented for practice. A large candle is lit and its height is written down every 10 hours. Time in hours, x: 0, 10, 20, 30, 40, 50 and height in centimeters, y: 20.1, 17.8, 16.1, 13.9, 12.2, 9.9.
Step 1: Get the line and r from technology. Enter the two lists in a graphing calculator and run linear regression. It returns y = −0.2x + 20 and r ≈ −0.999. The calculator does this step. Every step after it is your own reasoning.
Step 2: Read the slope and the intercept with their units, because the units tell you what the numbers mean. The slope −0.2 says the candle loses about 0.2 centimeter of height each hour. The intercept 20 says the model puts the candle at 20 centimeters when x = 0.
Step 3: Read r. It is negative because the height goes down as the time goes up. It is very close to −1, so the points sit close to one falling line. The graph shows the six measured points along that line.
Step 4: Predict. At 25 hours the model gives y = −0.2(25) + 20 = 15 centimeters. The time 25 is inside the measured times, 0 to 50, so this is interpolation and the prediction is on firm ground.
Step 5: Find a residual, which is the observed value minus the predicted value. At x = 20 the candle measured 16.1 and the model predicts −0.2(20) + 20 = 16. The residual is 16.1 − 16 = 0.1 centimeter. A positive residual means the measured point sits above the line.
Step 6: Check against the table. The prediction of 15 centimeters for 25 hours falls between the heights measured at 20 and 30 hours, 16.1 and 13.9, so it is reasonable. Now try x = 120. The model gives −0.2(120) + 20 = −4 centimeters, and a candle can't have a negative height. A prediction outside the measured range is called extrapolation, and this one fails. On the graph, the line has dropped below the horizontal axis by x = 120.
- y = −0.2x + 20
05 · Build understanding
Worked example: a line by hand, then by calculator
More invented data: a sapling's age in years, x: 1, 2, 3, 4, 5, 6 and its height in inches, y: 15, 20, 27, 32, 37, 44. You can write a reasonable line without technology and then compare it with the calculator's line.
Step 1: Draw a line through the middle of the trend and pick two points on your line. A line through (2, 20) and (6, 44) works well here. Three of the six data points are on it, two are 1 inch above it and one is 1 inch below it. The graph shows the six points and this line.
Step 2: Find the slope from those two points, because slope is the change in y divided by the change in x: (44 − 20)/(6 − 2) = 24/4 = 6 inches per year.
Step 3: Find the intercept by putting one of the points into y = 6x + b. With (2, 20) you get 20 = 12 + b, so b = 8. The hand-fit line is y = 6x + 8.
Step 4: Compare with technology. Linear regression on the same lists gives y ≈ 5.74x + 9.07 and r ≈ 0.999, with the slope and intercept rounded to the nearest hundredth. The two lines are close. The calculator line is the least-squares line: it makes the sum of the squared residuals as small as possible.
Step 5: Check with a prediction inside the data. At 4.5 years the hand-fit line gives 6(4.5) + 8 = 35 inches and the calculator line gives 5.74(4.5) + 9.07 = 34.9 inches. They differ by 0.1 inch, and both land between the heights measured at 4 and 5 years, 32 and 37.
- y = 6x + 8
06 · Build understanding
What r tells you, and what it cannot
The correlation coefficient r measures how closely the points follow a line. The sign gives the direction, and the distance from 0 gives the strength. So r = −0.93 shows a stronger linear association than r = 0.78, because 0.93 is greater than 0.78.
Here is an invented set with a looser pattern. x: 1, 2, 3, 4, 5, 6 and y: 5, 9, 4, 10, 7, 12. A calculator gives r ≈ 0.611. The trend is upward, but the points scatter widely around any line, so a prediction from a line fitted to these points deserves much less trust than one from the candle line. The graph shows the six points and their least-squares line, with several points far above or below it.
A strong r still describes only how two lists of numbers move together. Picture an invented survey of students from kindergarten through fifth grade that records each student's shoe size and the number of words the student can spell. The two numbers rise together, and r would be high. Bigger shoes don't teach anyone to spell. Age drives both: older children have larger feet and have had more years of school. Age is called a lurking variable.
To say that one thing causes another you need more than a scatterplot. You need a controlled experiment: change one variable yourself, keep everything else the same, and watch the other variable. The candle comes close to that. Only the burning time changed, so it is fair to say that burning made the candle shorter.
- least-squares line
07 · Build understanding
Three mistakes to watch for
Judging strength by the sign. A student sees r = −0.93 and r = 0.78 and calls the positive one stronger. Compare the distances from 0 instead: 0.93 against 0.78. The negative sign only says the line falls.
Subtracting a residual backward. For the candle at x = 20, observed minus predicted is 16.1 − 16 = 0.1. Predicted minus observed gives −0.1, which puts the point on the wrong side of the line.
Treating a strong association as proof of a cause. Shoe size and spelling move together, yet neither causes the other. Before you use the word causes, ask what else could be moving both variables, and ask whether anyone ran a controlled experiment.
Apply and explain
Your turn
Collect or use recorded real mass data, calculate a hand-fit line and technology regression, and report r and residuals. Predict the mass at an intermediate count and explain why a prediction far outside the measured range needs caution. Run the invented arithmetic check and critique a causal claim about sales and swimming.
Use paper or explain your thinking aloud before opening the solution.
Check your reasoning
Measured results vary; a sound fit has units, sensible slope/intercept, and residual checks. The invented check gives m=2.2,b=0.6,r≈0.984 and prediction 6.1 at 2.5. A high correlation does not identify a cause; confounding and study design matter.
- Real paired data documented
- Hand and technology fits compared
- Correlation interpreted as linear association
- Prediction limits and causation distinguished
Ready to practice?
Try the questions, get feedback, and review your reasoning.
Linear fit, correlation, and causation practice →