Why do we use linear models if real-world relationships are often curved?

Linear models are simple, interpretable, and work well over the limited range of data you collect. A relationship may be slightly curved overall, but linear within the range you measure—much like Earth appears flat on a local scale even though it is round. If a linear model fits your data well, it is a useful tool for prediction and interpretation. If the relationship is strongly curved, you would see that in the scatter plot (points would form a curve, not cluster around a line), and you would choose a different model.

What is the difference between the line of best fit and the linear model?

They are the same thing. The line of best fit is fitted to your data using a method like least squares, and once it is fitted, that line becomes your linear model. You use it to make predictions and to understand the relationship between the variables.

If my prediction is way off when I check it against real data, what went wrong?

Several things could have happened. First, the relationship in new data might differ from the original data (the real world changes). Second, you might have extrapolated beyond your data range, where the model is less reliable. Third, the original data might have had a weak linear relationship to begin with—if the scatter plot was very scattered, the line would not predict individual cases accurately, even if it captures the overall trend. Always check the strength of the original relationship and the range of your data.

Yes. A zero slope means the line is horizontal, and the output variable does not change as the input variable increases. This indicates no linear relationship. For example, if customer age has no relationship with purchase amount, the fitted line would have a slope near zero.

M8MATH-10.3

Using a Linear Model with Bivariate Data

Learn to interpret slope and intercept in linear models, use them to make predictions, and recognize when predictions extrapolate beyond collected data.

What you'll do in this lesson

A voice-first session with the Crimsora tutor on Using a Linear Model with Bivariate Data, then targeted practice and FRQs — with the tutor adapting to where you get stuck.

Start Using a Linear Model with Bivariate Data free

What this lesson covers

When biologists want to understand how sunlight affects plant growth, they collect data from many plants, fit a line through the scatter plot, and use that line to predict what will happen. That fitted line is called a linear model. In this lesson, you'll learn how to read the slope and intercept from a linear model, understand what they tell you in the real world, use the model to predict new values, and—just as importantly—when NOT to trust a prediction because it goes beyond the data you actually collected.

Understanding Slope in a Linear Model

The slope of a fitted line tells you the rate of change: how much the output variable changes for each unit increase in the input variable. In the plant example, a slope of 1.5 means that for every additional hour of daily sunlight, the model predicts a plant will be about 1.5 cm taller. Slope always has units: in this case, centimeters per hour. When you read a fitted line from a graph or equation, the slope describes a real-world relationship. A steep slope means small changes in the input produce big changes in the output. A gentle slope means the output changes slowly. A negative slope means the variables move in opposite directions—as one goes up, the other goes down. The key is that slope represents per-unit change, which is why it appears in predictions: to estimate an unknown output, you multiply the slope by the change in input from a reference point.

Understanding the Intercept in Context

The intercept (specifically, the y-intercept when the line is written as

y = mx + b

) is the output value when the input is zero. In the plant model, the y-intercept might represent the height a plant would have with zero hours of daily sunlight. However, the intercept does not always make practical sense. If the input variable cannot be zero—for example, if you are modeling a plant's height based on the number of weeks it has been growing, you cannot have zero weeks and measure a real plant—then the intercept is just a mathematical feature of the line, not a quantity you would observe. When interpreting a linear model, always ask whether the intercept's input value (

x = 0

) actually exists in the real situation. If it does, the intercept tells you a meaningful starting value. If it does not, note it but do not interpret it as a realistic prediction.

Making Predictions with a Linear Model

Once you understand the slope and intercept, you can use the linear model to predict an output value for a new input value. If the model is

y = 1.5x + 2

(where

x

is hours of sunlight and

y

is plant height in centimeters), you can predict the height of a plant with 8 hours of sunlight:

y = 1.5(8) + 2 = 12 + 2 = 14

cm. Substitute the new input value into the equation, compute, and you have your prediction. Predictions are most reliable near the center of the data—in the range of input values you actually measured. The linear model is fitted to describe the relationship you observed; it stands on that foundation of data.

Extrapolation and the Limits of a Model

Extrapolation means using a model to predict beyond the range of data you actually collected. If your data on plant sunlight ranged from 4 to 12 hours per day, and you use the model to predict height at 20 hours of sunlight, you are extrapolating. Extrapolation is risky because you do not know whether the linear relationship continues outside the data range. Perhaps plants stop growing taller at extreme sunlight levels, or the relationship becomes curved. Your linear model cannot tell you this—it only fits the data you have. When extrapolating, always flag it explicitly: say "This prediction assumes the relationship continues beyond the data range" or "This is an extrapolation and may not be reliable." Predictions within the range of collected data are called interpolations and are generally more trustworthy. A good practice: always identify the minimum and maximum input values in your data, and note when a prediction falls outside that range.

Checking Your Interpretation

To interpret a linear model accurately, follow these steps. First, identify the variables—what does

x

represent, and what does

y

represent, including units. Second, read the slope and write it with units, translating it into "per-unit change" language. Third, consider whether the intercept makes real-world sense. Fourth, when making a prediction, substitute the input value into the model equation and compute carefully. Fifth, check whether the input value is within the range of your data. If it is, you have an interpolation. If it is beyond the range, label it as an extrapolation and note the limitation. This systematic approach prevents misinterpretation and catches unreliable predictions before you make claims based on them.

Key terms

Linear model.: A line fitted to a scatter plot that describes the relationship between two variables; often written in the form $y = mx + b$ .
Slope.: The rate of change represented by $m$ in $y = mx + b$ ; tells you how much the output variable changes for each unit increase in the input variable, always with units.
Intercept (y-intercept).: The value of $y$ when $x = 0$ , represented by $b$ in $y = mx + b$ ; may or may not have a meaningful interpretation depending on whether $x = 0$ makes sense in the real context.
Interpolation.: A prediction made for an input value that lies within the range of data actually collected; generally more reliable than extrapolation.
Extrapolation.: A prediction made for an input value outside the range of collected data; risky because the linear relationship may not continue beyond the data.
Per-unit change.: The amount the output variable changes for a single unit change in the input variable; this is what slope expresses.
Fitted line.: A line drawn through or near points on a scatter plot to model the relationship between two variables.
Bivariate data.: Data involving two variables, often displayed in a scatter plot to show the relationship between them.

Worked example

A marine biologist collects data on the length (in meters) of sea turtles and their age (in years). She fits a line to the scatter plot and finds the model

\text{length} = 0.4 \cdot \text{age} + 0.2

. Her data came from turtles aged 2 to 20 years. (a) Interpret the slope. (b) Does the y-intercept make sense in this context? (c) Predict the length of a 15-year-old turtle. (d) Is that prediction reliable?

(a) The slope is 0.4, with units meters per year. This means that for each additional year of age, a sea turtle is expected to be approximately 0.4 meters (or 40 centimeters) longer.

(b) The y-intercept is 0.2, which would represent the length of a turtle with age 0 years. A newborn or unborn turtle cannot be measured in the same way as the turtles in the study, so this intercept does not have a meaningful real-world interpretation. It is a mathematical feature of the line, not a prediction you would trust.

(c) Substitute age = 15 into the model:

\text{length} = 0.4(15) + 0.2 = 6 + 0.2 = 6.2

meters. The predicted length of a 15-year-old sea turtle is 6.2 meters.

(d) Yes, this prediction is reliable because age 15 falls within the range of data collected (2 to 20 years). This is an interpolation, not an extrapolation, so we can trust the model's prediction. If the biologist had asked about a 50-year-old turtle, that would be an extrapolation and much less trustworthy.

Practice questions

A scientist measures the temperature (in degrees Celsius) of water as it sits in the sun, and plots temperature against time (in minutes). The fitted line is

T = 0.8t + 15

, where

t

is time and

T

is temperature. The data came from observations over 45 minutes. What does the slope 0.8 tell you?

The water was 0.8 degrees Celsius when the timer started.
The water temperature increases by 0.8 degrees Celsius every minute.
The water has been in the sun for 0.8 minutes.
It takes 0.8 minutes for the water to warm up by 1 degree Celsius.

Answer: The water temperature increases by 0.8 degrees Celsius every minute.

Slope is the rate of change: it tells how much the output (temperature) changes for each unit increase in the input (time). A slope of 0.8 means 0.8 degrees Celsius per 1 minute. The other choices misinterpret slope as an intercept (choice A), a time value (choice C), or an inverse rate (choice D).

You have a linear model that predicts a student's test score based on the number of hours studied, fitted from data where students studied between 2 and 8 hours. Would you be more confident predicting a test score for a student who studied 5 hours or one who studied 12 hours? Explain.

Answer: The prediction for 5 hours would be more confident. 5 hours is interpolation because it falls within the range of data collected (2 to 8 hours), so the model has been fitted on similar data. 12 hours is extrapolation, going beyond the range, so there is no guarantee the linear relationship continues at that level of study time.

Interpolation—predicting within the data range—is more trustworthy than extrapolation because your model was fitted on that type of data. Beyond the range, the relationship may change shape, level off, or behave differently in ways the linear model cannot capture. Always check whether a prediction is inside or outside your data range.

A clothing store tracks daily sales revenue (in dollars) against the number of customers who enter the store. The linear model is

R = 45n + 120

, where

R

is revenue and

n

is the number of customers. (a) What does the y-intercept (120) represent, and does it make sense? (b) What does the slope tell you?

Answer: (a) The y-intercept (120) would represent revenue when 0 customers enter. This does not make practical sense because if nobody enters the store, there would be no sales revenue. It is a mathematical artifact of the fitted line, not a real quantity. (b) The slope (45) tells you that for each additional customer, revenue is expected to increase by 45 dollars.

The intercept only makes sense if the input value of zero is realistic in the context. Here, zero customers would mean zero transactions, so a positive revenue intercept is illogical. However, slopes always have an interpretation: they describe per-unit change in the output. The slope 45 dollars per customer is meaningful and usable for prediction.

FAQ

Why do we use linear models if real-world relationships are often curved?: Linear models are simple, interpretable, and work well over the limited range of data you collect. A relationship may be slightly curved overall, but linear within the range you measure—much like Earth appears flat on a local scale even though it is round. If a linear model fits your data well, it is a useful tool for prediction and interpretation. If the relationship is strongly curved, you would see that in the scatter plot (points would form a curve, not cluster around a line), and you would choose a different model.
What is the difference between the line of best fit and the linear model?: They are the same thing. The line of best fit is fitted to your data using a method like least squares, and once it is fitted, that line becomes your linear model. You use it to make predictions and to understand the relationship between the variables.
If my prediction is way off when I check it against real data, what went wrong?: Several things could have happened. First, the relationship in new data might differ from the original data (the real world changes). Second, you might have extrapolated beyond your data range, where the model is less reliable. Third, the original data might have had a weak linear relationship to begin with—if the scatter plot was very scattered, the line would not predict individual cases accurately, even if it captures the overall trend. Always check the strength of the original relationship and the range of your data.
Can a slope be zero?: Yes. A zero slope means the line is horizontal, and the output variable does not change as the input variable increases. This indicates no linear relationship. For example, if customer age has no relationship with purchase amount, the fitted line would have a slope near zero.

Learn this with a teacher, not a page

The Crimsora tutor teaches Using a Linear Model with Bivariate Data live — explaining on a whiteboard, asking you questions, and adapting to where you get stuck.

Try a live session free