SAT Line of Best Fit: Predictions, Interpolation, Extrapolation, and Residuals

A line of best fit summarizes the overall linear trend in a scatterplot.

It does not need to pass through every data point.

Its purpose is to model the relationship well enough to:

• describe direction;
• estimate a rate of change;
• make predictions;
• compare actual values with predictions.

If the model is:

$$\hat y=mx+b,$$

then:

• $m$ describes the predicted change in $y$ for a 1-unit increase in $x$;
• $b$ is the predicted y-value when $x=0$, if that interpretation is meaningful.

The SAT often tests whether you can use the line without overclaiming what it means.

Why line-of-best-fit questions matter

College Board's Problem-Solving and Data Analysis domain includes two-variable data and models, scatterplots, linear models, and interpretation of slope, intercept, and predictions.

Questions may ask you to:

• estimate an equation;
• use a given regression equation;
• predict $y$ from $x$;
• interpret slope/intercept;
• compare prediction with actual data;
• distinguish interpolation from extrapolation;
• reason about residuals;
• assess whether a linear model is appropriate.

For the broader topic, see SAT Scatterplots and Models.

A fitted line is not required to pass through data points

Suppose the plotted points are scattered around a line.

A best-fit model tries to balance those deviations.

If you force the line through two arbitrary points from the scatterplot, you may get a line that does not represent the overall trend well.

On SAT graph-estimation questions, use points on the drawn best-fit line, not random data points near it.

Worked example 1: use the given equation

A line of best fit is:

$$\hat y=4.2x+18.$$

Predict $y$ when:

$$x=10.$$ $$\hat y=4.2(10)+18.$$ $$=42+18=60.$$ $$\boxed{60}$$

The hat over $y$ reminds you this is a predicted value.

Interpret the slope with units

Suppose:

• $x$ = hours of training;
• $y$ = score points.

Model:

$$\hat y=3.5x+52.$$

Slope:

$$3.5.$$

Interpretation:

The model predicts score increases by about 3.5 points for each additional hour of training.

Do not say:

Score increases by exactly 3.5 for every student.

A line of best fit summarizes a trend, not a deterministic rule.

Worked example 2: intercept interpretation

Model:

$$\hat y=2.4x+75$$

where $x$ is age in years from 10 to 16.

The mathematical intercept is:

$$75.$$

But $x=0$ lies outside the observed age range.

So saying:

The model predicts a value of 75 at age 0

is mathematically what the intercept encodes, but it may not be a meaningful real-world interpretation because the model was built for ages 10–16.

The SAT may test whether an intercept is meaningful in context.

Interpolation vs extrapolation

Interpolation

Prediction within the range of observed x-values.

Usually safer.

Extrapolation

Prediction outside the observed x-range.

Usually less reliable because the observed relationship may not continue.

If data were collected for:

$$5\le x\le20,$$

predicting at:

$$x=12$$

is interpolation.

Predicting at:

$$x=50$$

is extrapolation.

Worked example 3: identify extrapolation

A model is based on temperatures from:

$$10^\circ C\text{ to }35^\circ C.$$

Using it to predict at:

$$25^\circ C$$

is:

interpolation

Using it at:

$$60^\circ C$$

is:

extrapolation

The equation still produces a number at 60, but reliability is more questionable.

Estimating slope from a drawn best-fit line

Suppose a drawn line passes approximately through:

$$(2,10)$$

and:

$$(8,34).$$

Slope:

$$m=\frac{34-10}{8-2} = \frac{24}{6} = 4.$$

Estimated model:

$$y=4x+b.$$

Use:

$$(2,10):$$ $$10=8+b.$$ $$b=2.$$

So:

$$\boxed{\hat y\approx4x+2}$$

Choose clear points on the line itself, ideally grid intersections far apart to reduce reading error.

Why far-apart points are better for estimating slope

If you estimate slope using two very close graph points, a small reading error can strongly distort:

$$\Delta y/\Delta x.$$

Using points far apart makes the same visual uncertainty a smaller proportion of the total change.

That is a practical graph-reading habit.

Worked example 4: choose the more plausible line

Suppose a scatterplot shows:

• clear positive trend;
• points roughly centered around a moderate upward line.

Candidate equations:

A. $y=-3x+10$
B. $y=0.1x+1000$
C. $y=3x+10$
D. $y=20x-500$

A best-fit line should match:

• direction;
• rough steepness;
• location through the center of the data.

If the visual trend rises about 3 y-units per x-unit near intercept 10, choice C is plausible.

This is estimation, not exact point-fitting.

Actual vs predicted value

Suppose at:

$$x=6,$$

the model predicts:

$$\hat y=42,$$

but actual data point is:

$$y=47.$$

Residual:

$$47-42=5.$$

For the full sign logic, see SAT Residuals.

Worked example 5: underprediction

Model:

$$\hat y=5x+12.$$

At:

$$x=7,$$

prediction:

$$47.$$

Actual:

$$53.$$

The model underpredicts by:

$$6.$$

Residual:

$$\boxed{+6}$$

Line of best fit and correlation direction

Positive slope usually corresponds to positive linear association.

Negative slope corresponds to negative linear association.

But slope magnitude is not the same thing as correlation strength.

A steep line does not necessarily mean a stronger relationship than a shallow line because slope depends on the measurement units.

Correlation strength concerns how tightly points cluster around the trend, not simply how steep the trend is.

Worked example 6: units change slope

Suppose height is measured in centimeters instead of meters.

The numerical slope changes by a factor of 100.

The strength of the relationship does not suddenly become 100 times stronger.

This is why slope and correlation strength are different concepts.

Line of best fit does not prove causation

Suppose ad spending and sales have a strong fitted linear relationship.

The equation may be useful for prediction.

It does not by itself prove:

Increasing ad spending causes exactly the predicted increase in sales.

Other variables, selection effects, or study design can matter.

See Correlation vs Causation.

When a linear model is inappropriate

A line may be inappropriate if:

• the scatterplot is clearly curved;
• residuals show a systematic pattern;
• the relationship has a threshold or saturation;
• the x-range includes very different regimes.

Do not force linearity because the SAT gives you coordinate axes.

Worked example 7: curved data

Suppose points trace a clear U-shape.

A horizontal line may have small average error in one sense, but it misses the structure.

Better conclusion:

A nonlinear model is more appropriate.

A residual plot would likely show a systematic curved pattern.

Outliers and the best-fit line

An outlier can pull a fitted line.

If one point lies far from the rest, ask whether:

• it is a vertical outlier;
• it is far out in x;
• it changes the apparent trend.

The SAT may ask which point least follows the model or has the largest residual.

Predictions are estimates, not certainties

If the model predicts:

$$72,$$

the actual observation can be:

$$69,\ 75,\ 80,$$

or another value.

A line of best fit describes expected trend, not guaranteed outcomes.

Words such as:

• approximately;
• predicts;
• estimates;
• expected;

are appropriate.

Words such as:

• must;
• always;
• exactly;

are often too strong.

Regression equation vs hand-drawn line

A regression equation is produced by a statistical fitting method.

A hand-drawn line of best fit is an approximation.

SAT questions may provide either.

If an equation is given, use it.

If only a graph is shown, estimate carefully.

Do not replace a supplied model with your own arbitrary two-point line.

Desmos strategy

If a question provides tabular x-y data and asks for a regression model, Desmos can be useful for exploratory or practice work.

On the actual digital SAT, use the built-in graphing calculator when appropriate, but be sure you understand:

• what the slope means;
• what the intercept means;
• whether the model is linear;
• whether the requested prediction is interpolation or extrapolation.

For calculator strategy, see Desmos for the Digital SAT.

Common traps

The most common errors are:

• using two random data points rather than the fitted line;
• assuming every point lies on the model;
• extrapolating far beyond the observed data without caution;
• interpreting slope without units;
• treating the intercept as meaningful when $x=0$ is outside context;
• assuming a steep slope means a strong correlation;
• treating a best-fit equation as proof of causation;
• using exact language for an approximate model.

A reliable SAT best-fit routine

1. Identify x and y variables and units.
2. Determine trend direction.
3. If equation is given, use it directly.
4. If estimating from a graph, select clear points on the fitted line.
5. Interpret slope as predicted change in y per unit x.
6. Check whether the prediction is interpolation or extrapolation.
7. Compare actual vs predicted using residuals if asked.
8. Avoid causal conclusions unless study design supports them.

SAT-style practice

Question 1

$$\hat y=2x+15.$$

Predict at $x=12$:

$$24+15=39.$$ $$\boxed{39}$$

Question 2

Data x-range is 10 to 40.

Prediction at $x=30$:

interpolation

Prediction at $x=75$:

extrapolation

Question 3

Model predicts 90, actual is 84.

Residual:

$$84-90=-6.$$ $$\boxed{-6}$$

Question 4

Scatterplot is strongly curved.

Should a linear best-fit model automatically be used?

No

Question 5

Model slope is 4.2 dollars per item.

Interpretation:

Predicted cost rises by about $4.20 per additional item.

FAQ: SAT line of best fit

What is a line of best fit?

A line that summarizes the overall linear trend in two-variable data.

Does it pass through every point?

No.

What is interpolation?

Prediction within the observed x-range.

What is extrapolation?

Prediction outside the observed x-range.

Why is extrapolation risky?

The observed relationship may not continue beyond the data range.

How do residuals relate to the best-fit line?

Residual equals actual y minus predicted y from the line.

Does a line of best fit prove causation?

No. It models association/prediction.

How do I interpret the slope?

As predicted change in y for each 1-unit increase in x, with units.

How to practice this on MastaPrep

The quickest way to lock in a skill like this is to practice it on its own, then again a few days later, mixed in with other questions. That is how MastaPrep is built. A placement check finds your weakest SAT Math areas, the study plan targets them, every question has a worked explanation, and missed questions come back until you get them right without help.

See how MastaPrep works or try free SAT practice questions. The free plan needs no credit card.

Related MastaPrep guides

• SAT Scatterplots and Models
• SAT Residuals
• SAT Regression: Slope and Intercept
• Correlation vs Causation
• Desmos for the Digital SAT

Final takeaway

A line of best fit is a model, not a rule every data point obeys.

Use it to estimate, interpret trend, predict within context, and compare actual with predicted values. Then keep the boundaries clear:

prediction ≠ certainty

and:

association ≠ causation

---

Official SAT references:
https://satsuite.collegeboard.org/sat/whats-on-the-test/math/types/problem-solving
https://satsuite.collegeboard.org/practice/student-question-bank/math

Trademark note: SAT® is a registered trademark of College Board. Desmos® is a trademark of Desmos Studio PBC. MastaPrep is independent and is not affiliated with or endorsed by College Board or Desmos.