Predicted value in AP Statistics

In AP Statistics, a predicted value (ŷ, read "y-hat") is the value of the response variable that a regression model estimates for a given value of the explanatory variable. You find it by plugging x into the equation of the line, such as ŷ = a + bx.

What is the Predicted value?

A predicted value is the output of a regression model. You take an x-value, plug it into the equation of the line, and the number you get back is ŷ ("y-hat"). The hat is the whole point. It tells you this is the model's estimate of the response variable, not a value someone actually measured.

Think of the regression line as a machine that turns x-values into best guesses. If a model says ŷ = 8.2 + 1.7x and x = 5, the predicted value is 8.2 + 1.7(5) = 16.7. Every point on the regression line is a predicted value. The real data points scatter around it, and the gap between a real point and the line is the residual. A predicted value is only as trustworthy as the model behind it. If the relationship isn't linear, or if you plug in an x far outside the data you collected (extrapolation), ŷ can be way off.

Why the Predicted value matters in AP® Statistics

Predicted values sit at the center of the regression topics in AP Statistics: 5.3 Linear Regression Models, 5.4 Residuals, and 5.5 Least-Squares Regression. In 5.3 you calculate ŷ from an equation and interpret it in context. In 5.4 you compare ŷ to the actual y to get a residual. In 5.5 you learn that the least-squares line is the one whose predicted values make the sum of squared residuals as small as possible. If you can't confidently produce and interpret ŷ, none of the later regression skills hold up. That's why the idea appears in nearly every regression question on the exam, from quick MCQ calculations to multi-part FRQs.

How the Predicted value connects across the course

Residuals (Topic 5.4)

A residual is just actual minus predicted, y − ŷ. The predicted value is the model's guess, and the residual measures how wrong that guess was for one point. A positive residual means the model underestimated, and a negative one means it overestimated.

Least-Squares Regression Line (Topic 5.5)

The least-squares line is the line whose predicted values miss the data by the smallest total squared amount. So "least squares" is really a rule for choosing the best set of ŷ values. One handy result is that plugging in x̄ always gives a predicted value of ȳ, because the line passes through (x̄, ȳ).

Slope of a Regression Model (Topic 5.3)

The slope describes how the predicted value changes, not how the actual y changes. For ŷ = 42 + 3.5x, each one-unit increase in x raises the predicted y by 3.5. Leaving out the word "predicted" is one of the most common ways to lose credit on a slope interpretation.

Extrapolation (Topic 5.3)

You can compute ŷ for any x, but that doesn't mean you should trust it. Predicted values are reliable only inside the range of x-values used to build the model. Outside that range, you're assuming the pattern keeps going, and the data never showed you that.

Is the Predicted value on the AP® Statistics exam?

Expect predicted values in three main ways.

1. Compute it. MCQs hand you an equation like ŷ = 15 − 2.4x and ask for the prediction at a specific x. Show the substitution and keep units in context.

2. Use it to find a residual. A classic stem gives a plant growth model, ŷ = 8.2 + 1.7x, and an actual growth of 18.3 cm at 5 grams of fertilizer. You find ŷ = 16.7 first, then the residual is 18.3 − 16.7 = 1.6 cm. The model underpredicted that plant.

3. Interpret in context. Questions about models like ŷ = 12 − 0.5x or ŷ = 42 + 3.5x test whether you describe the slope and intercept in terms of the predicted response. Graders look for that word.

Released FRQs build whole questions on regression contexts where this matters. 2017 FRQ Q1 models wolf weight from body length. 2022 FRQ Q1 models bullfrog mass from length. In settings like these, being able to produce a predicted value, compare it to an observed one, and phrase everything in context is what earns full credit.

The Predicted value vs Actual (observed) value

The actual value, y, is what was really measured in the data. The predicted value, ŷ, is what the regression line says y should be for that x. They're usually different, and the difference (y − ŷ) is the residual. A quick check is to look at the symbol. If it has a hat, it came from the model. If it doesn't, it came from the data.

Key things to remember about the Predicted value

• A predicted value, written ŷ, is the response the regression line estimates for a given x, and you get it by plugging x into the equation.

• The hat on ŷ signals that the value comes from the model, while y without a hat is an actual observed value from the data.

• A residual equals the actual value minus the predicted value, so a positive residual means the model underpredicted.

• When you interpret slope, say the predicted response changes by the slope amount for each one-unit increase in x.

• The least-squares regression line always predicts ȳ when you plug in x̄.

• Predicted values for x-values outside the range of the data are extrapolations and should not be trusted.

Frequently asked questions about the Predicted value

What is a predicted value in AP Stats?

A predicted value (ŷ) is the response a regression model estimates for a specific x. For ŷ = 8.2 + 1.7x, the predicted value at x = 5 is 16.7.

Is the predicted value the same as the actual value?

No. The actual value (y) was measured, and the predicted value (ŷ) comes from the line. They usually differ, and that difference, y − ŷ, is the residual.

How do you find a residual using a predicted value?

First compute ŷ by plugging x into the equation, then subtract that from the actual y. For a plant with 5 g of fertilizer and actual growth of 18.3 cm under ŷ = 8.2 + 1.7x, the residual is 18.3 − 16.7 = 1.6 cm.

Do I have to say "predicted" when interpreting slope on the AP exam?

Yes. The slope describes the change in the predicted response, not a guaranteed change in every actual value. For ŷ = 42 + 3.5x, say "the predicted y increases by 3.5 for each 1-unit increase in x," and put it in context.

Can you trust a predicted value for any x?

No. Predictions are reliable only within the range of x-values used to build the model. Plugging in an x far outside that range is extrapolation, and the linear pattern may not hold there.