01 · Learn the idea
Measure repeated transmission
Using a fixed light source, light sensor, and identical translucent sheets, measure transmitted intensity with 0,1,2,3,4 sheets. Keep distances, alignment, ambient light, and sensor settings fixed. Use room-temperature materials and follow the teacher’s equipment directions. A teacher-recorded actual data set is an alternative. Record units, repeats, and sensor limits.
A model y=ab^n describes repeated multiplication: a estimates the no-sheet value and b the retained fraction per layer. This model is only a candidate; reflections, sheet variation, ambient light, or sensor response can make a simple exponential inadequate.
02 · Build understanding
Fit rather than assume
Enter the measured pairs in a graphing tool and choose exponential regression or y_1 ~ a b^(x_1), with positive b. Record coefficients and residuals. Use ordinary least squares consistently; changing to a logarithmic fitting mode changes the error measure and can change results. Tool explanation: https://help.desmos.com/hc/en-us/articles/204349605-Log-Mode .
For an invented exact check, inputs 0,1,2,3 with outputs 80,60,45,33.75 fit y=80(3/4)^x. At x=4 the prediction is 25.3125. The model approaches zero but does not reach it at a finite input; an instrument’s zero reading may instead reflect its detection limit or background correction.
03 · Build understanding
Worked example: a growing chess club
These numbers are invented for practice. A school chess club counts its members at the start of each month. Month, x: 0, 1, 2, 3, 4, 5 and members, y: 40, 51, 62, 79, 97, 123.
Step 1: Test the pattern with ratios, because an exponential model multiplies by the same number at every step. Divide each count by the one before it: 51/40 = 1.275, 62/51 ≈ 1.216, 79/62 ≈ 1.274, 97/79 ≈ 1.228 and 123/97 ≈ 1.268. The ratios all sit near 1.25, so an exponential model is reasonable. The differences are 11, 11, 17, 18 and 26. They grow, so a line would fit poorly.
Step 2: Get the model from technology. Enter the two lists and run exponential regression. The calculator returns y ≈ 40(1.25)^x, with a rounded to the nearest whole number and b rounded to the nearest hundredth.
Step 3: Read a and b with their units. The 40 is the value of the model at month 0, so the model starts the club at 40 members. The 1.25 is the growth factor: each month the model multiplies the membership by 1.25, which is growth of 25% per month.
Step 4: Predict month 6. The model gives 40(1.25)^6. Do the power first: (1.25)^6 ≈ 3.8147, and 40(3.8147) ≈ 152.59, so about 153 members. Month 6 is one step past the data, so this is a short extrapolation.
Step 5: Check against the table. The last count was 123, and 123(1.25) = 153.75, which agrees with about 153. Inside the data, the model gives 40(1.25)^4 ≈ 97.66 for month 4, and the count was 97. Now try month 24. The model gives 40(1.25)^24, which is about 8,470 members, more than most schools have students. A model fitted to six months says very little about two years.
The graph shows the six counts as points and the model as a curve. The points sit close to the curve, and the curve bends upward, getting steeper each month. The extra point at month 6 is the prediction, which lies on the curve itself.
- y = 40(1.25)^x
04 · Build understanding
Worked example: a bouncing ball
More invented data: a ball is dropped and the height of each bounce is read from a video. Bounce number, x: 0, 1, 2, 3, 4, where 0 stands for the drop itself, and height in centimeters, y: 150, 106, 74, 51, 37.
Step 1: Compare differences with ratios. The differences are −44, −32, −23 and −14, so the ball does not lose the same height each time and a line is a poor model. The ratios are about 0.707, 0.698, 0.689 and 0.725. They stay near 0.7, so an exponential model is reasonable.
Step 2: Get the model from technology. Exponential regression gives y ≈ 150(0.70)^x, with a rounded to the nearest whole number and b rounded to the nearest hundredth.
Step 3: Read a and b. The 150 is the drop height in centimeters. The 0.70 is less than 1, so the heights shrink: each bounce keeps about 70% of the height before it and loses about 30%.
Step 4: Predict bounce 5. The model gives 150(0.7)^5 = 150(0.16807) ≈ 25.2 centimeters.
Step 5: Check. The last measured bounce was 37 centimeters, and 37(0.7) = 25.9, close to the 25.2 the model gives. A residual is the measured value minus the value the model predicts. Take bounce 3: the model predicts 150(0.7)³ = 150(0.343) = 51.45 and the measured height was 51, so the residual is 51 − 51.45 = −0.45 centimeter. The measured point is just below the curve.
The graph shows the five heights as points and the model as a curve. The curve falls quickly at first and then flattens, because each bounce loses 30% of a smaller height. It gets close to the x-axis but stays above it.
- y = 150(0.70)^x
05 · Build understanding
Judging the fit and its limits
A calculator will fit an exponential curve to any list of positive numbers. Checking that the ratios are roughly constant is how you decide whether the curve belongs there.
Residuals tell you how well it worked. For the bouncing ball the residuals are 0, 1, 0.5, −0.45 and about 0.99 centimeter. Next to heights of 37 to 150 centimeters those are small, so the model describes the five measurements well.
Tools can differ a little. Some calculators fit a line to the logarithms of the y-values, and some graphing tools fit the curve to the y-values directly. For the chess club and the bouncing ball, both methods give the same a and b after rounding. With rougher data the two can disagree in the last digit, so write down which tool you used.
Be careful far from the data. For bounce 20 the ball model gives 150(0.7)^20, about 0.12 centimeter. The model never reaches 0, but a real ball stops bouncing well before that.
06 · Build understanding
Three mistakes to watch for
Reading b as the percent change. A growth factor of 1.25 means 25% growth, because 1.25 − 1 = 0.25. A factor of 0.70 means a 30% loss, because 1 − 0.70 = 0.30. Writing 125% growth or a 70% loss misreads the model.
Multiplying before taking the power. In 40(1.25)^6 the exponent belongs to 1.25 only. Multiplying first gives 50^6, which is 15,625,000,000. The correct value is about 152.59.
Using the factor as if it were a slope. Some students compute 40 + 1.25(6) = 47.5 for month 6. That adds 1.25 members each month. The model multiplies by 1.25 each month, which gives about 153.
Apply and explain
Your turn
Fit an exponential to the measured or recorded light data. Interpret a and b, graph residuals, predict the reading with an additional sheet, and compare with a held-out measurement. Run the invented check and explain why a high fit statistic does not guarantee accuracy for 100 sheets.
Use paper or explain your thinking aloud before opening the solution.
Check your reasoning
Measured coefficients vary; b should be interpreted as a multiplicative retained fraction only if the data support that model. The exact check uses a=80,b=0.75 and predicts 25.3125 at four layers. Extreme extrapolation can fail because the measuring system and physical assumptions have limits.
- Actual data and controls recorded
- Exponential fit computed with technology
- Parameters and residuals interpreted
- Prediction checked and extrapolation qualified
Ready to practice?
Try the questions, get feedback, and review your reasoning.
Exponential regression and prediction practice →