SAT Math lesson · Scatterplots & models
SAT scatterplots and models: lines of best fit, residuals, linear vs. exponential
How to answer SAT two-variable data questions: predictions from a line of best fit, what its slope and intercept mean, actual vs. predicted values, and telling linear from exponential models in tables and graphs.
The core idea
A line of best fit turns a scatterplot into predictions: go up from x to the line, then across to y. Its slope is the predicted change in y for each 1-unit increase in x. A data point's actual value differs from the prediction by its vertical gap to the line. Linear data change by a constant amount for equal steps in x; exponential data change by a constant factor.
Core ideas
Words used here
- Scatterplot: a graph with one dot for each person or thing, placed by two measurements (x across, y up).
- Line of best fit: a line drawn through the middle of the dots that shows the trend. It is used to predict y from x.
- Predicted value: the y-value on the line of best fit at a given x.
- Actual (observed) value: the y-value of a real data point.
- Residual: actual value − predicted value. Positive for a point above the line, negative below.
- Linear model: y = mx + b; y changes by the same amount for each 1-unit step in x.
- Exponential model: y = a(b)^[x]; y is multiplied by the same factor b for each 1-unit step in x.
- Prediction: go up from x to the line, then across to the y-axis; or substitute x into the line's equation.
- Slope: the predicted change in y for each 1-unit increase in x, in y-units per x-unit.
- y-intercept: the predicted y when x = 0 (it may have no real meaning if x = 0 is far from the data).
- Above or below: a point above the line has an actual value greater than predicted.
- Linear or exponential: for equal steps in x, constant differences mean linear; constant ratios mean exponential.
On the SAT.
- Reading graphs: SAT scatterplots are drawn to scale, and the values a question needs sit on gridlines. Read the line, not a nearby dot.
- "Best interpretation" questions: the right choice names the predicted change in y, with y's unit, for each 1-unit increase in x.
- Calculator: you can type the data and the line into the graphing calculator, but most questions take one reading from the graph or one substitution into the equation.
- Typed answers: a count of points or a difference between actual and predicted values goes in the box as a number, without units.
Try it. A line of best fit for study time x (hours) and quiz score y (points) is y = 2.5x + 40. What score does it predict for 6 hours of study?
Show the answer
y = 2.5 × 6 + 40 = 15 + 40 = 55 points.
Predictions from a line of best fit
A line of best fit predicts y for any x: read the line's height at that x, or substitute x into its equation. To go backward (which x gives a predicted y?), set the equation equal to that y and solve for x.
Why it works. The line summarizes the trend of all the dots, so its height at x is the typical y for that x. A single dot at the same x is one real case, which can be above or below the trend.
How to spot it: "based on the line of best fit, what is the predicted …", "according to the line of best fit, … corresponds to what value of x".
Example. A line of best fit passes through (0, 40) and (8, 60). What is the predicted y at x = 4, and which x gives a predicted y of 70?
- Slope: (60 − 40) ÷ (8 − 0) = 20 ÷ 8 = 2.5, so the line is y = 2.5x + 40.
- At x = 4: 2.5 × 4 + 40 = 50.
- Predicted y of 70: 70 = 2.5x + 40, so 2.5x = 30 and x = 12.
Trap: answering with the y-value of a dot at x = 4. That is an actual value, not the prediction.
Edge cases.
- Outside the data (extrapolation): a prediction far beyond the plotted x-values is less reliable, even though the arithmetic is the same.
- A point on the line has actual value equal to its prediction.
- Reading off the graph: use x-values where the line crosses a grid intersection, so the reading is exact.
Check yourself. A line of best fit is y = −100x + 1,700, where y is monthly rent in dollars and x is miles from downtown. What distance gives a predicted rent of $1,100?
Check your answer
1,100 = −100x + 1,700, so −100x = −600 and x = 6 miles.
What the slope and intercept mean
In a line of best fit y = mx + b, the slope m is the predicted change in y for each 1-unit increase in x, and b is the predicted y when x = 0. Both carry units: m is in y-units per x-unit. From a graph, find the slope from two grid points on the line (rise ÷ run) and b where the line crosses the y-axis.
Why it works. Moving 1 unit right along a line always moves it m units up (or down when m is negative), so m is the change per unit. At x = 0 the mx term is 0, which leaves b.
How to spot it: "what is the best interpretation of 2.5 in this context?", "which equation could represent the line of best fit?", "by how much does the predicted … increase when … increases by 5?"
Example. For adults, the line of best fit y = −0.5x + 95 relates daily exercise x (minutes) to resting heart rate y (beats per minute).
- Slope −0.5: for each additional minute of daily exercise, the predicted resting heart rate decreases by 0.5 beats per minute.
- Intercept 95: the predicted resting heart rate for someone who doesn't exercise is 95 beats per minute.
- Over 20 more minutes of exercise: the predicted rate falls by 0.5 × 20 = 10 beats per minute.
Trap: "every adult's heart rate drops by exactly 0.5 per minute of exercise". The line gives predictions; the dots scatter around it.
Edge cases.
- A negative slope is a predicted decrease; say "decreases by 0.5", not "increases by −0.5".
- Equation from a graph: a line through (0, 5) and (8, 45) has slope (45 − 5) ÷ 8 = 5, so y = 5x + 5. Run over rise (8 ÷ 40 = 0.2) is a common slip.
- An intercept outside the data (x = 0 when every x is between 50 and 100) is part of the equation but not a sensible prediction.
Harder example. A line of best fit for hand span x (cm) and height y (cm) passes through (16, 160) and (26, 185). By how much does the predicted height increase when hand span increases by 4 cm?
- Slope: (185 − 160) ÷ (26 − 16) = 25 ÷ 10 = 2.5 cm per cm.
- Over 4 cm: 2.5 × 4 = 10 cm.
Check yourself. A line of best fit is y = 4x − 180, where y is iced drinks sold and x is the day's high temperature in °F. What does 4 mean?
Check your answer
For each additional degree Fahrenheit of high temperature, the predicted number of iced drinks sold increases by 4.
Actual vs. predicted values
A data point's actual value is its own y; its predicted value is the line's y at the same x. The difference, actual − predicted, is the residual: positive for points above the line, negative for points below. "How much greater than predicted" is the vertical gap between the dot and the line.
Why it works. The prediction and the data point share the same x, so the only difference between them is vertical: how far the dot sits above or below the line.
How to spot it: "how many data points lie above the line of best fit?", "the actual … is how much less than predicted?", "the actual value is p% greater than the predicted value".
Example. A line of best fit for car age x (years) and value y (thousands of dollars) passes through (0, 27) and (10, 12). Point M is (8, 18).
- Slope: (12 − 27) ÷ 10 = −1.5, so at x = 8 the line is at 27 − 1.5 × 8 = 15.
- Actual − predicted: 18 − 15 = 3 thousand dollars; M is above the line.
- As a percent of the prediction: 3 ÷ 15 × 100 = 20% greater than predicted.
Trap: dividing by the actual value (3 ÷ 18) for "percent greater than predicted". The base is the predicted value.
Edge cases.
- Counting above and below: compare each dot with the line at that dot's own x, not with a horizontal level.
- A residual of 0 means the point lies on the line.
- Big residuals mark points the line predicts poorly; they don't make the line wrong.
Check yourself. A line of best fit predicts a rent of $900 at 8 miles from downtown. An apartment 8 miles out rents for $1,000. How much greater is its actual rent than predicted, and by what percent?
Check your answer
1,000 − 900 = $100 greater, which is 100 ÷ 900 ≈ 11.1% of the prediction.
Linear or exponential?
For equal steps in x, a linear relationship adds the same amount to y each step (constant differences), and an exponential one multiplies y by the same factor each step (constant ratios). An exponential model y = a(b)^[x] starts at a when x = 0 and is multiplied by b for each 1 added to x: growth when b > 1, decay when 0 < b < 1.
Why it works. In y = mx + b, adding 1 to x adds m to y. In y = a(b)^[x], adding 1 to x multiplies y by b. So the step-to-step pattern tells the two apart.
How to spot it: a table of values, "which of the following best describes the relationship", a curved line of best fit, "increases by 10% each year".
Example. Which model fits each table?
| x | 0 | 1 | 2 | 3 |
|---|---|---|---|---|
| Table P | 605 | 580 | 555 | 530 |
| Table Q | 500 | 600 | 720 | 864 |
- P: differences −25, −25, −25. Constant, so decreasing linear.
- Q: differences 100, 120, 144 aren't constant; ratios 600 ÷ 500 = 720 ÷ 600 = 864 ÷ 720 = 1.2. Constant, so increasing exponential.
Trap: calling Q linear because the differences are "close". A constant ratio is exponential, however slowly it grows.
Edge cases.
- Uneven steps in x: compare changes per unit of x, or use only equally spaced rows.
- Reading a curve: an exponential curve through (0, 20) and (1, 30) has a = 20 and b = 30 ÷ 20 = 1.5, so y = 20(1.5)^[x]; check that x = 2 gives 45.
- Percent language: "decreases by 20% each year" is the factor 0.8, an exponential decay.
Harder example. An exponential curve of best fit passes through (0, 80) and (1, 40). Which equation could it be: y = 80(0.5)^[x], y = 0.5(80)^[x], or y = −40x + 80?
- a is the value at x = 0: 80. b = 40 ÷ 80 = 0.5.
- The model is y = 80(0.5)^[x]. The line −40x + 80 matches only the first two points: at x = 2 it gives 0, while the curve gives 20.
Check yourself. Does the table x: 0, 2, 4, 6 and y: 3, 12, 48, 192 show a linear or an exponential relationship?
Check your answer
Each step of 2 in x multiplies y by 4 (12 ÷ 3 = 48 ÷ 12 = 192 ÷ 48 = 4): increasing exponential.
How hard questions disguise it
For harder questions (L4–L5)
At L4, the slope must be read from the graph and multiplied over several units ("by how much does the predicted height increase over 5 weeks?"), or a prediction is run backward to find x. At L5, residuals combine with percents ("the actual value is p% greater than predicted"), and exponential curves combine with exponential models (reading a and b from the curve). Read only values where the line or curve meets a grid intersection, and name the base before you compute a percent.
Practice
Practice these patterns in timed sets on the scatterplots practice page.
Common traps
Reading a data point instead of the line
A prediction is the height of the line of best fit at that x, not the height of a dot there. Go up from x to the line, then across to the vertical axis.
Misreading what the slope says
In y = 2.5x + 40, the slope 2.5 is the predicted change in y for each 1-unit increase in x; 40 is the predicted y at x = 0. The line describes the trend, not an exact change for every data point.
Stopping at the change for one unit
Over a change of 5 in x, the predicted y changes by 5 times the slope. A value read from the line is a prediction, not a change.
Confusing constant differences with constant ratios
Check equal steps in x. Constant differences in y mean linear; constant ratios mean exponential. Values like 1,000, 1,100, 1,210 grow by 10% each step, so they are exponential, though the differences look close.
Mixing up the parts of y = a(b)^x
a is the value at x = 0 and b is the factor for each 1 added to x. b greater than 1 is growth; b between 0 and 1 is decay.