Regression failure modes: exact fit or best fit?

Your feedback matters:

Lesson progressPractice problems 0/5

Difficulty: Advanced
Estimated time: 55 minutes
Domains: Problem Solving and Data Analysis Advanced Math
Techniques: RegressionVerificationResidualsExact FitBest FitModel Choice

What you’ll learn

Distinguish exact function conditions from measured or rounded data and report the result appropriately.
Define a residual and interpret its sign.
Verify an exact fit against every original condition.
Use the model family named or justified by the prompt.
Recognize when regression hides solutions or lacks enough information and switch methods.

Prerequisites

Why this matters on the SAT

Know what the fitted model promises

Regression can produce an equation for almost any table. The important question is what that equation means.

For exact function values, the model must satisfy every stated condition.
For measured data, a line or curve of best fit is an estimate and usually misses some points.
For some questions, regression hides information the SAT asks you to report.

SAT example

A researcher records the following measurements. The data are described as having a roughly linear trend.

$x$	$1$	$2$	$3$	$4$
$y$	$10$	$15$	$19$	$26$

Which equation is the most appropriate model for the data?

A
$y=5.2x+4.5$
B
$y=\frac{2}{3}x^3-\frac{9}{2}x^2+\frac{83}{6}x$
C
$y=4.5x+5.2$
D
$y=5.2x-4.5$

Fast Desmos solution

Enter the table and compare y_1~mx_1+b with y_1~px_1^3+qx_1^2+rx_1+s. The linear regression gives

m=5.2\qquad\text{and}\qquad b=4.5.

The cubic passes through all four observations, but that does not make it the most appropriate model. The prompt identifies a roughly linear trend, so the answer is A.

Words such as measured and recorded are supporting clues. Here, roughly linear trend and most appropriate model are decisive. Read the complete claim instead of classifying from one word.

Calculator loads as you approach

The cubic interpolates every observation, but the linear model matches the trend named in the prompt.

Name the job before fitting

Classify what the prompt expects before you read a regression result. The same calculator command can represent two very different claims.

Treat the fit as exact when…

the table gives values of a function;
a graph or model passes through stated points;
several equations or function values must all be true; or
the relationship has a stated exact form with enough independent conditions.

Treat the fit as an estimate when…

the values were measured, observed, recorded, or rounded;
the scatterplot has a roughly linear or curved trend;
a line or curve of best fit is used; or
the question asks for a prediction, approximation, or best estimate.

Regression is the wrong tool when the question asks for all solutions, the greatest or least solution, or the number of solutions; when a known function rule only needs one evaluation; when the conditions do not determine every fitted parameter; or when the prompt names a different model or method that regression would replace.

For exact conditions, every original condition must hold. For best fit, the model should follow the overall pattern and does not need to pass through every observation.

Check your understanding:

Classify each case as exact fit, best fit, or wrong tool: (1) three stated values of a quadratic function; (2) measured temperature and battery data with a roughly linear trend; (3) an equation asking for the sum of all real solutions.

1. Exact fit means every condition survives

A residual is the observed output minus the model’s predicted output:

\text{residual}=y_{\text{observed}}-y_{\text{predicted}}.

For exact conditions, every residual should be $0$ , apart from tiny numerical display noise.

Worked example

A quadratic function $f$ has the exact values shown.

$x$	$-2$	$0$	$2$
$f(x)$	$3$	$5$	$15$

Determine the equation that defines $f$ .

Fit y_1~ax_1^2+bx_1+c. Desmos reports

a=1,\qquad b=3,\qquad c=5.

Define the fitted function and residual list with f(x)=ax^2+bx+c and d_1=y_1-f(x_1). The residual list is

d_1=[0,0,0].

The zero residuals show that the candidate fits these rows. Complete the verification by substituting into the original model:

\begin{aligned} f(-2)&=(-2)^2+3(-2)+5=3,\\ f(0)&=5,\\ f(2)&=2^2+3(2)+5=15. \end{aligned}

Regression output is numerical. The substitutions establish that the exact function is

f(x)=x^2+3x+5.

Check your understanding:

A stated function table has residuals $[0,0,0.002]$ after regression. Can you accept the displayed model as an exact fit without another check?

Common mistake:

Calling a close fit exact. A small residual can be acceptable for measured data, but not for an exact function condition. If the prompt says the points lie on the model, every original condition must be verified.

Calculator loads as you approach

The fitted quadratic passes through every exact function value, and every residual is zero.

2. Best fit allows nonzero residuals

Worked example

A researcher records measured practice-time data and fits the line

\widehat y=3.9x+48.3,

where $\widehat y$ means the predicted value of $y$ .

$x$	Observed $y$	Predicted $\widehat y$	Residual $y-\widehat y$
$1$	$52$	$52.2$	$-0.2$
$2$	$57$	$56.1$	$0.9$
$3$	$59$	$60.0$	$-1.0$
$4$	$64$	$63.9$	$0.1$
$5$	$68$	$67.8$	$0.2$

Interpret what the nonzero residuals say about this best-fit model.

The residuals are not all $0$ , and that is not an error. For this linear model, Desmos chooses the slope and intercept that make the total squared vertical error as small as possible. You do not need to calculate that total by hand on the SAT.

Desmos can also plot residuals from the regression controls. A residual:

above $0$ means the observed point is above the model;
below $0$ means the observed point is below the model; and
equal to $0$ means the model passes through that point.

Try it yourself:

Change the score at $x=4$ from $64$ to $62$ . Predict whether the fitted line will move up or down near that point, then inspect the updated residuals and reset the example.

Check your understanding:

After changing the value at $x=4$ from $64$ to $62$ , does the fitted line move up or down near that point, and what sign does the new residual have?

Common mistake:

Forcing every observation onto the model. A line of best fit summarizes the trend. Changing or discarding measured points merely because their residuals are nonzero would confuse a model prediction with an exact function rule.

Calculator loads as you approach

This residual plot uses a tight viewport to show positive and negative residuals around y = 0.

3. Fit the model family the prompt supports

Desmos may display $r$ , $R^2$ , or RMSE. These statistics summarize a selected model, sometimes on a transformed scale in Log Mode; they cannot override the prompt, choose the model family, or turn rounded observations into exact values.

Refer back to the opening calculator. The cubic has enough free parameters to pass through all four measurements, while the linear model follows the roughly linear trend named by the prompt. Do not choose a more complicated model only because it bends closer to a small sample.

Check your understanding:

A prompt describes measured data as roughly linear. A linear regression has small nonzero residuals, while a cubic passes through all four observations. Which model should you use, and why?

Common mistake:

Choosing the most flexible model. A higher-degree curve can imitate a simpler pattern over a small table. Match the stated or justified model family first, then inspect the fit.

4. Know when regression answers the wrong question

Worked example

Consider

p^2=9.

What are all real values of $p$ that satisfy the equation?

Entering 9~p^2 asks regression to assign one fitted value to the parameter $p$ . It is not a request to list every solution.

Graphing x^2=9 shows both solutions:

x=-3\qquad\text{and}\qquad x=3.

Use graphing and solution counting when the prompt asks for all solutions, their sum, the greatest or least solution, or how many solutions exist.

Regression cannot invent a missing condition

One point does not determine a line with two unknown parameters. For $y=mx+b$ , the point $(2,7)$ creates one equation,

7=2m+b,

but infinitely many pairs $(m,b)$ satisfy it. A regression result cannot create the missing independent condition.

Check your understanding:

The equation $(x-2)^2=25$ has two real solutions. The question asks for the greater solution. Should you use regression or graphing, and what is the answer?

Common mistake:

Trusting one regression candidate as the complete solution set. Regression estimates one value for each fitted parameter. It does not promise to enumerate every solution of an equation.

Calculator loads as you approach

Regression assigns one candidate to p, while graphing exposes both real solutions.

5. Verify before you answer

Use this checklist after every regression:

Name the job. Is the prompt exact, best fit, or not a regression task?
Match the model. Did the function family come from the prompt or a justified pattern?
Check the information. Are there enough independent conditions for the fitted parameters?
Inspect the fit. For exact conditions, verify every row. For measured data, inspect the trend and residuals.
Answer the prompt. Report an exact value only when exact conditions support it. Use estimate language for a best-fit prediction.

Fit statistics are supporting evidence. The prompt and original conditions remain the acceptance test.

Example: Choose a rounded exponential model

Worked example

The table shows a set of measurements.

$x$	$0$	$2$	$4$	$6$
$y$	$400$	$256$	$164$	$105$

The $y$ -values are rounded to the nearest whole number.

An exponential model has the form

y=ab^x,

where $a>0$ and $0<b<1$ .

Which model is most consistent with the data?

A
$y=400(0.64)^x$
B
$y=400(0.8)^x$
C
$y=256(0.8)^x$
D
$y=400(0.8)^{x/2}$

Step 1

Classify the job

The outputs are measured and rounded, and the prompt asks for the model most consistent with them. This is a best-fit problem.

Do not require the model to reproduce every rounded value exactly.

Step 2

Fit the stated model

Enter the table, then fit y_1~a*b^(x_1).

Automatic Log Mode reports values close to

a\approx399.93\qquad\text{and}\qquad b\approx0.8002.

These estimates point to $a=400$ and $b=0.8$ in the choices. A fit statistic cannot override the rounded original values or the model family named in the prompt.

Calculator loads as you approach

The exponential fit is close to a = 400 and b = 0.8 because the measurements were rounded.

Step 3

Check the candidate in context

For choice B,

y=400(0.8)^x.

At the supplied inputs, this model predicts

400,\quad256,\quad163.84,\quad104.8576.

Rounding gives

400,\quad256,\quad164,\quad105,

which matches the measured table. The answer is B.

The unrounded model values do not equal the rounded observations in every row, and they are not expected to.

Common mistake:

Demanding exact equality from rounded data. Rounded measurements represent intervals of possible original values. Choose the model that is consistent with the rounded observations instead of forcing exact equality to the displayed integers.

Finish the solution

The exact conditions and regression are ready. Verify the fitted parameters and evaluate the requested expression.

Verify the exact exponential model

Finish the solution

The function $f$ is defined by

f(x)=a\cdot3^x+b,

where $a$ and $b$ are constants. If

f(1)=11\qquad\text{and}\qquad f(2)=29,

what is the value of $a+b$ ?

First steps

Treat the two function values as exact conditions.
Enter the points $(1,11)$ and $(2,29)$ in a table.
Fit the stated model with y_1~a*3^(x_1)+b.
Desmos reports the candidate values $a=3$ and $b=2$ .

Finish it

Your answer

Calculator loads as you approach

Verify both exact conditions, then evaluate a + b.

Practice problems

SAT practice problems

For each problem, identify the regression job before typing. Start from a clean calculator and preserve the wording about exact values, predictions, and estimates.

Interpret a residual in context

Practice problem

A line of best fit predicts a response $y$ from an input $x$ :

\widehat y=2.5x+6.

When $x=4$ , the observed response is $14$ . Which statement correctly interprets this observation?

Calculator loads as you approach

Evaluate the prediction, then interpret observed minus predicted in context.

Verify an exact quadratic model

Practice problem

The number of items in a simulation is modeled by a quadratic function $P(t)$ . The table gives three exact function values.

$t$	$0$	$2$	$4$
$P(t)$	$50$	$66$	$50$

Which equation defines $P(t)$ ?

Calculator loads as you approach

Fit or test the stated quadratic values, then verify every row exactly.

Diagnose an underdetermined model

Practice problem

A table records the exact value $f(2)=7$ for a linear function $f(x)=mx+b$ . A student enters that one row and fits y_1~mx_1+b. Which conclusion is justified?

Calculator loads as you approach

One exact point supplies only one equation for the two unknown coefficients.

Find every solution before answering

Practice problem

What is the sum of all real solutions to

x^2-4x-21=0?

Your answer

Calculator loads as you approach

Graph the equation to expose both solutions before adding them.

Finish the lesson

5 practice examples left

Finish the remaining questions correctly to complete this lesson.

Quick recap

Read the prompt before reading the regression output.
Exact function values and stated conditions require every residual to be $0$ and every condition to verify.
Measured, observed, or rounded data can have nonzero residuals because a best-fit model summarizes a trend.
A residual is observed output minus predicted output.
Use the model family named or justified by the prompt. A strong fit statistic does not choose the model for you.
Regression cannot invent missing conditions or list every solution of an equation.
Report exact-fit conclusions as exact and best-fit predictions as estimates.

Next lesson

Meet every condition

Turn every stated equation, restriction, and relationship into one verified model.

Start next lesson

Free Practice

What you’ll learn

Prerequisites

Know what the fitted model promises

Fast Desmos solution

Name the job before fitting

Treat the fit as exact when…

Treat the fit as an estimate when…

1. Exact fit means every condition survives

2. Best fit allows nonzero residuals

3. Fit the model family the prompt supports

4. Know when regression answers the wrong question

Regression cannot invent a missing condition

5. Verify before you answer

Example: Choose a rounded exponential model

Classify the job

Fit the stated model

Check the candidate in context

Finish the solution

Verify the exact exponential model

First steps

Finish it

Practice problems

SAT practice problems

Interpret a residual in context

Verify an exact quadratic model

Diagnose an underdetermined model

Find every solution before answering

5 practice examples left

Quick recap

Meet every condition

Free question bank practice problems