Skip to content
SAT essentials · Math

SAT scatterplots and lines of best fit

A scatterplot shows how two quantities move together. A line of best fit turns that pattern into predictions, and its slope and intercept have real meanings.

These are the concepts this skill rests on, the same pages as in the lesson.

Reading Two-Variable Data

Each Point Is One Subject Measured Twice

A scatterplot shows two measurements for each subject, whether that's a person, a day, or a tree. Each point is one subject: how far across it sits gives one measurement, and how far up gives the other.

A scatterplot of hours of practice per week and free throws made out of 20 for nine players. One point, at 6 hours and 14 free throws, is highlighted, with dashed guides down to 6 on the horizontal axis and across to 14 on the vertical axis. 0 1 2 3 4 5 6 7 8 9 10 0 4 8 12 16 20 Hours of practice per week Free throws made (out of 20) (6, 14) one player: 6 hours across, then 14 made up

Read across first, then up, and name each number with its axis title. The blue point is a player who practiced 6 hours a week and made 14 free throws: the ordered pair (6,14)(6, 14). Swapping the two numbers is a favorite wrong answer, so read the axis titles before you read the point.

A line graph is the same picture with the points joined in order, usually over time. Only the points are real readings; the segments just connect one reading to the next.

Direction: Positive, Negative, or None

The direction of an association is the way the points run from left to right.

Three small scatterplots. In the first the points rise from lower left to upper right: a positive association. In the second they fall from upper left to lower right: a negative association. In the third they show no trend: no association. positive rises to the right negative falls to the right none no trend
  • Positive association: the points rise to the right. Bigger values across go with bigger values up.
  • Negative association: the points fall to the right. Bigger values across go with smaller values up.
  • No association: the points show no trend up or down.

To check, compare the two ends: in a positive association, points on the left tend to sit lower than points on the right. A few points can break the pattern; direction describes the overall trend.

Form and Strength

Direction is one way to describe a cloud of points. Form and strength are two more, and a question usually asks about just one of them.

Three small scatterplots. The first two rise at about the same rate. In the first, strong, the points lie very close to a straight path. In the second, weak, the points spread far from it. The third is curved: the points rise more and more steeply. strong close to the path weak far from the path curved bends upward
  • Form: linear (the points follow a straight path) or nonlinear (they follow a curve).
  • Strength: how closely the points follow that path. Points close to the path make a strong association; points spread far from it make a weak one.

The first two plots rise at about the same rate, yet one is strong and one is weak. The only difference is the spread. So strength is not steepness: a nearly flat trend can be strong if the points hug it, and a steep one can be weak. Judge strength by how tightly the points cluster around the trend, never by how steeply it rises.

Average Rate of Change

The average rate of change between two points is the change in the up-value divided by the change in the across-value:

A line graph of a club’s members over 6 years, through (0, 10), (4, 30), (5, 40) and (6, 45). From year 0 to year 4, a blue guide shows a change of 20 over 4 years. From year 4 to year 5, an orange guide shows a change of 10 over 1 year. 0 1 2 3 4 5 6 0 10 20 30 40 50 Years since the club began Members +20 over 4 years: 5 per year +10 over 1 year: 10 per year

average rate of change=change in ychange in x\text{average rate of change} = \frac{\text{change in } y}{\text{change in } x}

From year 0 to year 4 the club gained 20 members, a rate of 20÷4=520 \div 4 = 5 members per year. From year 4 to year 5 it gained only 10, but in just 1 year: 10 members per year. The bigger change is not always the bigger rate. When the intervals are different lengths, divide each change by its own length before you compare. On a single graph, the steeper segment has the greater rate.

The Line of Best Fit

A Line of Best Fit Is a Model

A line of best fit is a line drawn through the middle of a scatterplot to model the trend. It usually misses most of the points: it runs through the middle, with the points above it balancing the points below.

A scatterplot of nine points with a line of best fit, y = 1.5x + 3. Five points, in blue, lie above the line, and four, in orange, lie below it, close to it on both sides. 0 1 2 3 4 5 6 7 8 9 10 0 4 8 12 16 20 Hours studied Quiz score (points) line of best fit blue: above the line orange: below the line

The line is the model; the points are the data. So a question about "the line of best fit" or "the model" is asking about the line, and one about "the data" or "the student who…" is asking about a point.

If no line is drawn, sketch one yourself. Follow the trend, and keep points on both sides of it along its whole length, not just at the ends.

The Slope From Two Points on the Line

To find the slope of a drawn line, pick two points on the line where it passes exactly through a gridline crossing, and pick them far apart. Don't use data points; they're usually off the line.

A line of best fit through (0, 20) and (8, 60), with gray data points around it. Two points on the line, (2, 30) and (6, 50), are marked in green, with a run of 4 and a rise of 20 between them. 0 1 2 3 4 5 6 7 8 9 10 0 10 20 30 40 50 60 70 80 x y (2, 30) (6, 50) run 4 rise 20 slope = rise ÷ run = 20 ÷ 4 = 5

slope=riserun=50−306−2=204=5\text{slope} = \frac{\text{rise}}{\text{run}} = \frac{50 - 30}{6 - 2} = \frac{20}{4} = 5

Read the values off the axis numbers instead of counting boxes. Here one box is 1 across but 10 up, so counting boxes would give 2÷4=0.52 \div 4 = 0.5, which is wrong.

Choosing an Equation: Sign, Intercept, Size

To match an equation to a scatterplot with no line drawn, check three things in order. Each check usually knocks out at least one choice.

A scatterplot of eight points falling from (2, 44) to (9, 18). A dashed green line follows the trend and meets the vertical axis near 50. 0 1 2 3 4 5 6 7 8 9 10 0 10 20 30 40 50 60 x y 1. falls to the right: the slope is negative 2. intercept near 50 3. falls about 35 over 10: slope about −3.5
  1. Sign. Points that fall to the right need a negative slope; points that rise need a positive one.
  2. Intercept. Sketch the trend back to the vertical axis (where x=0x = 0). Here it meets the axis near 50.
  3. Size. Read the rise over a long run. Here the trend falls about 35 over 10, a slope near −3.5-3.5.

So y=−3.5x+50y = -3.5x + 50 fits. A choice like y=50x−3.5y = 50x - 3.5 swaps the slope and the intercept; y=3.5x+50y = 3.5x + 50 has the wrong sign.

Changing the Units Changes the Equation

If the data are recorded in different units, the scatterplot looks the same, but the numbers on the axis change, and so does the equation.

Two copies of one scatterplot of age and height with the same line of best fit, one above the other. On top, height is in feet and the line is y = 0.5x + 2. Below, height is in inches, every number on the vertical axis is 12 times as large, and the line is y = 6x + 24. The point on the line at age 4 is marked in green on both: (4, 4) on top and (4, 48) below. 0 1 2 3 4 5 6 7 8 0 2 4 6 8 Age (years) Height (feet) 0 1 2 3 4 5 6 7 8 0 24 48 72 96 Age (years) Height (inches) (4, 4) (4, 48) y = 0.5x + 2 y = 6x + 24
  • Multiply every yy-value by cc: the slope and the intercept are both multiplied by cc. Feet to inches multiplies y=0.5x+2y = 0.5x + 2 by 12: y=6x+24y = 6x + 24.
  • Multiply every xx-value by cc: the slope is divided by cc, and the intercept stays the same. If xx changes from hours to minutes, every xx is 60 times as large, so y=30x+5y = 30x + 5 becomes y=0.5x+5y = 0.5x + 5.

To check either rule, convert one point on the old line and make sure it lands on the new line. The green point (4,4)(4, 4) in feet is (4,48)(4, 48) in inches, and 6(4)+24=486(4) + 24 = 48.

Predictions and Residuals

Predict From the Line, Not From a Point

A predicted value comes from the line of best fit. With an equation, plug in the given xx. With a graph, go up from xx to the line, then across to the vertical axis.

A scatterplot with a line of best fit through (0, 5) and (10, 20). A dashed green guide goes up from 6 on the horizontal axis to the line, then across to 14 on the vertical axis. A gray data point sits at (6, 18), above the line. 0 1 2 3 4 5 6 7 8 9 10 11 12 0 4 8 12 16 20 24 28 x y predicted: 14 data point: 18

At x=6x = 6, the line is at 14, so the model predicts 14. The data point at x=6x = 6 sits at 18: that's an actual value, not a prediction. Answer choices love to offer the nearest point's value, so don't take the bait.

With the equation y=1.5x+5y = 1.5x + 5, the same prediction is 1.5(6)+5=141.5(6) + 5 = 14.

A Residual Is Actual Minus Predicted

A residual tells you how far a point sits above or below the line:

A scatterplot with the line y = 1.5x + 3. Each point is joined to the line by a short vertical segment: blue for points above the line, orange for points below it. The point (4, 12) is 3 above the line, and the point (6, 10) is 2 below it. 0 1 2 3 4 5 6 7 8 9 10 0 4 8 12 16 20 x y +3 −2 above: positive below: negative

residual=actual−predicted\text{residual} = \text{actual} - \text{predicted}

  • A point above the line has a positive residual: the model predicted too low. The point (4,12)(4, 12) has 12−9=312 - 9 = 3.
  • A point below the line has a negative residual: the model predicted too high. The point (6,10)(6, 10) has 10−12=−210 - 12 = -2.

Always subtract in that order and keep the sign. Predicted minus actual gives the right size with the wrong sign, and that's a popular wrong answer.

Slope and Intercept in Context

The Slope Is a Predicted Rate

The slope of a line of best fit is the predicted change in yy for each increase of 1 in xx, with both units.

A line of best fit for plant height against weeks since planting, rising 3 centimeters for each week. A small orange step shows 1 more week and 3 more centimeters; a large blue step shows 5 more weeks and 15 more centimeters. 0 1 2 3 4 5 6 7 8 9 10 0 5 10 15 20 25 30 35 40 Weeks since planting Height (centimeters) 5 more weeks +15 cm = 5 × 3 1 week +3 cm slope: 3 cm per week, predicted

Here the slope is 3: each extra week, the predicted height goes up 3 centimeters. A correct answer choice gets three things right:

  1. Predicted. The line describes the trend, not every plant. "Each plant grows exactly 3 cm a week" is wrong.
  2. Units, in order: centimeters of height per week, not weeks per centimeter.
  3. Direction and size. A negative slope is a predicted decrease. Over 5 weeks the predicted change is 5×3=155 \times 3 = 15 cm, not 3.

The Intercept, and When It Means Nothing

The intercept of a line of best fit is the predicted yy-value when x=0x = 0. It can be a real starting amount, such as a plant's predicted height on the day it was planted.

A scatterplot of reaction time against age for adults aged 20 to 60, shaded as the range of the data. The line of best fit is solid over that range and dashed back to age 0, where it meets the vertical axis at 200. 0 10 20 30 40 50 60 70 0 40 80 120 160 200 240 280 320 360 Age (years) Reaction time (milliseconds) ages in the data: 20 to 60 at age 0: 200 no one that age is in the data

It means something only when x=0x = 0 is inside the data's range, or close to it. These data cover adults aged 20 to 60, so the intercept (200 milliseconds at age 0) comes from stretching the line far past the data. Nobody that age was measured, so here the intercept has no real meaning, and that's what the right answer choice will say.

Choosing a Model

Same Difference or Same Ratio

When xx goes up in equal steps, the yy-values tell you which model fits.

Two tables. In the first, x is 0, 1, 2, 3 and y is 5, 8, 11, 14: y goes up by 3 each step, the same difference, so the pattern is linear. In the second, y is 2, 6, 18, 54: y is multiplied by 3 each step, the same ratio, so the pattern is exponential. x 0 1 2 3 y 5 8 11 14 +3 +3 +3 same difference linear x 0 1 2 3 y 2 6 18 54 ×3 ×3 ×3 same ratio exponential
  • The same difference each step (add the same amount): linear.
  • The same ratio each step (multiply by the same factor): exponential.

Check the differences first: 8−5=38 - 5 = 3, 11−8=311 - 8 = 3. Then the ratios: 6÷2=36 \div 2 = 3, 18÷6=318 \div 6 = 3. A table that keeps rising, but by different amounts, is not linear. This test needs equal steps in xx; if the xx-values skip around, compare steps of the same width.

In a word problem, the same amount each year (adds $500) is linear, and the same percent each year (grows 4%) is exponential. Careful: a percent of a quantity, such as 6% of xx, is just 0.06x0.06x, and that's linear.

Reading an Exponential Model

An exponential model is y=a⋅bxy = a \cdot b^x. The number aa is the value at x=0x = 0, and bb is the factor for each step of 1.

The curve y = 20 times 1.5 to the power x, through (0, 20), (1, 30), (2, 45) and (3, 67.5). Each step of 1 across multiplies y by 1.5. 0 1 2 3 4 0 10 20 30 40 50 60 70 80 x y 20 30 45 67.5 ×1.5 ×1.5 ×1.5 starts at 20: a = 20 ×1.5 each step: b = 1.5

Here y=20(1.5)xy = 20(1.5)^x: it starts at 20 and is multiplied by 1.5 each step, giving 30, 45 and 67.5.

  • Growth: b>1b > 1. Growing 50% each step means b=1+0.5=1.5b = 1 + 0.5 = 1.5.
  • Decay: 0<b<10 < b < 1. Losing 20% each step means b=1−0.2=0.8b = 1 - 0.2 = 0.8.

The base is never the percent itself: growing 50% is not b=0.5b = 0.5, and losing 20% is not b=0.2b = 0.2.

A Curve That Turns Is Quadratic

Points that rise and then fall, or fall and then rise, fit neither a line nor an exponential model. They fit a quadratic model, whose graph is a parabola.

A scatterplot of revenue against ticket price that rises, levels off near a price of 10 dollars, and then falls. A blue curve that opens downward follows the points, with its highest point near (10, 100). 0 2 4 6 8 10 12 14 16 18 20 0 20 40 60 80 100 120 Ticket price (dollars) Revenue (hundreds of dollars) highest point near (10, 100) rises, then falls: opens down, so the x² term is negative

In vertex form, y=a(x−h)2+ky = a(x - h)^2 + k:

  • If the curve opens down (rises, then falls), aa is negative; if it opens up, aa is positive.
  • The turning point, the vertex, is (h,k)(h, k). Here it is near (10,100)(10, 100), so the model is y=−(x−10)2+100y = -(x - 10)^2 + 100.

Watch the sign inside: a vertex at x=10x = 10 gives (x−10)2(x - 10)^2, not (x+10)2(x + 10)^2. The full algebra of quadratics is in Skill 09: Nonlinear Functions.

Exponential Growth Eventually Passes Linear Growth

A linear quantity adds the same amount each step. A growing exponential quantity multiplies, so the amount it adds keeps getting bigger.

A line, y = 100 + 30x, and an exponential curve, y = 40 times 1.2 to the power x, from x = 0 to 16. The line starts higher, but the curve rises faster and passes the line between x = 14 and x = 15. 0 2 4 6 8 10 12 14 16 0 100 200 300 400 500 600 700 Years Amount adds 30 each year grows 20% each year the curve passes the line after about 14 years

The line starts at 100 and adds 30 a year. The curve starts at only 40, but it grows 20% a year: it adds 8 in the first year and more every year after. After about 14 years it passes the line, and it stays above it from then on.

This is always true: a growing exponential quantity, with a positive start and any growth rate above 0, eventually passes any linear one, whatever their starting values. To find roughly when, graph both and see where they cross, or compare values in a table.

Graph y=100+30xy = 100 + 30x and y=40(1+r)xy = 40(1 + r)^x, where rr is the yearly growth rate. Start with r=0.2r = 0.2, the 20% curve above: they cross near x=14x = 14. Set r=0.1r = 0.1: the curve grows more slowly, but it still catches the line, now near x=35x = 35. A smaller growth rate only makes you wait longer.

What a Model Cannot Tell You

Past the Data, a Prediction Is a Guess

A model is only trustworthy over the range of data it was built from. Using it far outside that range is called extrapolation: you're assuming the trend keeps going, and nothing in the data shows that it does.

A scatterplot of puppy weight against age for puppies 2 to 10 months old, shaded as the range of the data, with a line of best fit. The line is solid over the data and dashed beyond it, out to 40 months, where it reaches 170 pounds. 0 5 10 15 20 25 30 35 40 0 20 40 60 80 100 120 140 160 180 Age (months) Weight (pounds) data: 2 to 10 months 170 pounds at 40 months? the data stop at 10 months

These puppies were 2 to 10 months old. At 40 months the line predicts 170 pounds, but puppies stop growing, and data that end at 10 months can't show that. A prediction far outside the data's range is less reliable than one inside it. Answer the question that's asked: if it wants the model's value, compute it; if it asks how much to trust that value, say it's unreliable because it's outside the data.

An Association Is Not a Cause

An association, even a strong one, does not show that one quantity causes the other. Something else may drive both, or the cause may run the other way.

A diagram. A box labeled size of the fire sits at the top, joined by lines to two boxes below it: firefighters sent and damage done. A dashed line joins the two lower boxes, labeled associated, but neither causes the other. size of the fire firefighters sent damage done raises both associated, but neither causes the other

Fires that get more firefighters tend to do more damage. That's not because firefighters cause damage: a bigger fire brings more firefighters and does more damage. To show that one variable causes another, you need an experiment that assigns subjects at random (Skill 16: Evaluating Statistical Claims). From a scatterplot alone, the right answer keeps the association and drops the cause.

Keep going

Check Your Skills, the strategy for each kind of question, worked examples and practice from Easy to Hard are in the Quaero SAT Course.

More Problem-Solving and Data Analysis