Topic 2.6 Notes – Linear Regression Models

1. What a Linear Regression Model Is

When two quantitative variables have a roughly linear relationship, we model that pattern with a simple linear regression model.

The model is the least squares regression line (LSRL):

$$\hat{y} = a + bx$$

Here’s what each piece means:

• $x$ = explanatory variable (the input you choose or observe)
• $\hat{y}$ = predicted value of the response variable
• $a$ = y-intercept (predicted value of $y$ when $x = 0$)
• $b$ = slope (predicted change in $y$ for a 1-unit increase in $x$)

This line is called least squares because it minimizes the sum of the squared residuals, where a residual is:

$$\text{residual} = y - \hat{y}$$

Quick reminder from earlier:

• Correlation (r) tells you strength and direction.
• Regression gives you an actual prediction equation.

Correlation describes. Regression predicts.

2. Interpreting the Slope and Intercept

Before calculating anything, you need to understand what the numbers mean in context. That’s where points are earned or lost.

Slope $b$

Interpret it like this:

For each 1-unit increase in $x$, the predicted value of $y$ changes by $b$ units.

Be specific:

• Include units of x
• Include units of y
• Say whether it increases or decreases

If $b = -2.3$, that means predicted $y$ decreases by 2.3 units for each 1-unit increase in $x$.

On free response, a complete answer sounds like:

For each additional hour of exercise per week, the predicted resting heart rate decreases by 2.3 beats per minute.

No units = lost points.

Y-Intercept $a$

Interpretation:

When $x = 0$, the predicted value of $y$ is $a$.

But here’s the catch: this only matters if $x = 0$ makes sense.

If your data is about house sizes from 1200 to 3000 square feet, then interpreting price at 0 square feet is meaningless. You’re allowed to say:

Since 0 square feet is outside the scope of realistic house sizes, the intercept has no practical meaning.

That earns credit.

3. How to Calculate a Predicted Value

This is the core skill of this topic.

Suppose the regression equation is:

$$\hat{y} = 5.2 + 1.8x$$

You’re asked to predict when $x = 7$.

Step-by-step

1. Substitute 7 for $x$:

   $$\hat{y} = 5.2 + 1.8(7)$$

2. Multiply:

   $$\hat{y} = 5.2 + 12.6$$

3. Add:

   $$\hat{y} = 17.8$$

Now state it properly:

The predicted value of the response variable for $x = 7$ is 17.8 units.

Notice:

• This is $\hat{y}$, not an actual observed value.
• You must include context and units on the AP exam.
• Keep full precision until the final answer.

Students often forget to label it as predicted. The model gives an estimate, not certainty.

4. Interpolation vs Extrapolation

Regression works best within the range of observed x-values.

Interpolation

Predicting for an $x$-value inside the data range.

Example: If your data covers study times from 2 to 15 hours, predicting at 10 hours is interpolation. This is generally reasonable.

Extrapolation

Predicting for an $x$-value outside the data range.

If that same dataset is used to predict at 30 hours, that’s extrapolation.

Why it’s risky:

• The linear pattern may not continue.
• Predictions can become unrealistic.
• The further outside the range, the less reliable it becomes.

On AP free response, you must explicitly say:

This prediction is an extrapolation because the x-value is outside the range of observed data, so it may not be reliable.

If you don’t say it, you won’t get full credit.

5. Common Mistakes to Avoid

• Confusing $y$ and $\hat{y}$. Regression predicts $\hat{y}$.
• Forgetting units in slope interpretation.
• Interpreting the intercept when $x = 0$ isn’t realistic.
• Extrapolating without explaining why it’s unreliable.
• Assuming regression implies causation. Unless it’s a randomized experiment, it only shows association.