01 · Learn the idea
Obtain data with a meaningful scale
Photograph a teacher-approved curved arch or use a supplied straight-on image with a visible scale. Set an origin and record at least five horizontal-position/height pairs along the same edge. Avoid perspective distortion where possible. Record the image source, scale, and measurement uncertainty. Not every arch is a parabola; whether a quadratic is useful must be checked.
In a graphing tool, enter the pairs and fit y=ax²+bx+c using quadratic regression. Record a,b,c, then graph the fitted curve with the points. Do not report a coefficient of determination as the linear correlation r. Inspect residuals for systematic patterns.
02 · Build understanding
Use a known check before the measured case
Invented exact coordinates (0,2),(1,8),(2,10),(3,8),(4,2) come from y=−2x²+8x+2. A quadratic fit should reproduce that rule within rounding. The vertex is (2,10), and the predicted height at x=1.5 is 9.5. Exact fit in constructed data does not guarantee exact fit in a photograph.
Use the measured fit to estimate a height between sampled points. Compare one prediction with a withheld measured point. Large or patterned residuals can indicate measurement problems or a poor model family. Do not extend the arch formula to distant positions where the physical structure does not exist.
03 · Build understanding
Worked example: a tossed ball
These numbers are invented for practice. A ball is tossed straight up in a gym and its height is read from a video. Time in seconds, t: 0, 0.5, 1, 1.5, 2 and height in feet, h: 5.1, 15.7, 19.3, 13.9, 1. The graph plots these five points with the model found in Step 2.
Step 1: Look at the shape before you fit anything. The heights rise, reach a high point near 1 second and then fall. A line only rises or only falls, so a line is a poor choice. A quadratic has one turning point, so it is a candidate.
Step 2: Get the model from technology. Enter the two lists and run quadratic regression. The calculator returns h = −16t² + 30t + 5.
Step 3: Read the coefficients. The −16 is negative, so the graph opens downward, which matches a ball that goes up and comes back down. The 5 is the value of the model at t = 0, so the model releases the ball at a height of 5 feet. The measured height at 0 seconds was 5.1 feet, which is close.
Step 4: Predict inside the data. A prediction between measured inputs is called interpolation. At t = 0.75 the model gives −16(0.75)² + 30(0.75) + 5. Square first: 0.75² = 0.5625, and −16(0.5625) = −9. Then −9 + 22.5 + 5 = 18.5 feet.
Step 5: Find a residual, which is the observed value minus the predicted value. At t = 1 the ball measured 19.3 feet and the model predicts −16 + 30 + 5 = 19 feet. The residual is 19.3 − 19 = 0.3 foot.
Step 6: Check against the table. The time 0.75 is between 0.5 and 1, and 18.5 is between the heights measured there, 15.7 and 19.3. Now try t = 3, which is outside the measured times. A prediction outside the measured range is called extrapolation. The model gives −16(9) + 30(3) + 5 = −49 feet. The ball was on the floor long before that, so this extrapolation means nothing.
04 · Build understanding
Worked example: ticket price and revenue
More invented data, this time for a school play that ran several nights at each ticket price. Ticket price in dollars, x: 4, 5, 6, 7, 8, 9 and average revenue per night in dollars, y: 802, 874, 901, 873, 801, 676. The graph plots these six points with the model found in Step 2.
Step 1: Test the pattern with differences, because the x-values are evenly spaced. The first differences are 72, 27, −28, −72 and −125. They are not constant, so the data are not linear. The second differences are −45, −55, −44 and −53. They stay near −50, so a quadratic is a reasonable model. Real measurements give second differences that are close to constant, not exactly constant.
Step 2: Get the model from technology. Quadratic regression gives y ≈ −24.66x² + 295.53x + 14.05, with each coefficient rounded to the nearest hundredth.
Step 3: Estimate the price with the greatest revenue. The vertex of y = ax² + bx + c is at x = −b/(2a). Here that is −295.53/(2(−24.66)) = 295.53/49.32 ≈ 5.99, so the model points to a price of about $6.
Step 4: Predict the average nightly revenue for a price of $6.50. The model gives −24.66(6.5)² + 295.53(6.5) + 14.05 = −24.66(42.25) + 1920.945 + 14.05 = −1041.885 + 1920.945 + 14.05 = 893.11 dollars.
Step 5: Check. $6.50 is between $6 and $7, and 893.11 is between the revenues measured there, 901 and 873. At x = 6 the model predicts −24.66(36) + 295.53(6) + 14.05 = 899.47, so the residual is 901 − 899.47 = 1.53 dollars, tiny next to 900. One caution: the 14.05 is the value of the model at x = 0, a price far below the $4 to $9 that was tried. Free tickets bring in $0, so do not read 14.05 as real revenue.
05 · Build understanding
Deciding whether the fit is reasonable
A calculator will fit a quadratic to any list you give it, so the calculator cannot tell you whether a quadratic was a sensible choice. You decide that.
First check the shape: one turning point, with values rising on one side and falling on the other. With evenly spaced x-values, also check that the second differences are roughly constant.
Then look at the residuals. For the tossed ball they are 0.1, −0.3, 0.3, −0.1 and 0 foot. They are small next to heights of up to 19.3 feet, and they switch between positive and negative with no pattern. That is what a good fit looks like. Residuals that are all positive in the middle and all negative at the ends would mean the curve has the wrong shape.
Last, keep predictions inside the measured range. The ball model is trustworthy from 0 to 2 seconds because that is where the measurements are.
06 · Build understanding
Three mistakes to watch for
Multiplying before squaring. In −16(1.5)², square the 1.5 first: 1.5² = 2.25, and −16(2.25) = −36. Multiplying first gives (−24)² = 576, a positive number where the correct term is negative.
Stopping at the first differences. For x-values 1, 2, 3, 4, 5 and y-values 3, 5, 9, 15, 23 the first differences are 2, 4, 6, 8. They are not constant, and some students stop there and say no simple model fits. Go one more row: the second differences are 2, 2, 2, so the data are quadratic.
Trusting the model outside the data. The ball model gives −49 feet at 3 seconds, and the ticket model gives $14.05 of revenue for free tickets. Both inputs are outside the measured range, and both outputs are nonsense.
Apply and explain
Your turn
Create the measured coordinate table, quadratic regression, graph, and residual record. Estimate an intermediate height and compare a withheld point. Run the invented exact-data check, then explain whether your real arch is adequately modeled for a stated purpose.
Use paper or explain your thinking aloud before opening the solution.
Check your reasoning
The check model is −2x²+8x+2 with prediction 9.5 at 1.5. The measured model depends on the image and coordinates; accept a reasoned fit only if residuals are small enough for its stated use and scale. Poor fit is a valid finding, not a reason to alter measurements.
- Source, scale, and measured pairs retained
- Technology actually computes quadratic fit
- Prediction tested against withheld evidence
- Model scope and residual pattern evaluated
Ready to practice?
Try the questions, get feedback, and review your reasoning.
Quadratic regression from a measured curve practice →