A scatterplot displays paired measurements. Each point represents one individual or situation with an x-value and a y-value. Looking at the point cloud helps you describe a relationship and decide whether a line gives a useful model.
Describe direction, form, and strength
A positive association means larger x-values tend to go with larger y-values. A negative association means larger x-values tend to go with smaller y-values. No clear association means the plot shows no consistent tendency.
Form describes whether the pattern is approximately straight or curved. Strength describes how closely the points follow that pattern. A strong relationship can be curved, so “strong” does not automatically mean a straight-line model is appropriate.
For study hours and test scores, each point must pair one student's hours with that same student's score. Sorting the hours and scores separately would destroy the pairings and invent a different relationship.
More study hours are associated with higher scores in these observations, but the points do not lie exactly on one line.
Describe this figure
| Hours | Score |
|---|---|
| 1 | 52 |
| 2 | 60 |
| 3 | 61 |
| 4 | 75 |
| 5 | 82 |
Example: interpret a model's slope
A fitted line for a plant's height is where is time in weeks and is height in centimeters. The slope predicts an average increase of 2.5 cm for each additional week. The intercept predicts height 8 cm at week zero.
At week 6, the predicted height is cm. This is a model prediction, not a guarantee that every plant will have exactly that height.
The units of slope are vertical units divided by horizontal units: centimeters per week. A slope without its context tells only part of the story.
Draw a line through the overall pattern
An informal line of best fit should pass through the middle of the point cloud, with points reasonably balanced above and below. It need not pass through every point or through the origin.
To estimate its equation, select two convenient points on the fitted line, not necessarily two original observations. Compute slope as change in divided by change in , then use one point to find the intercept.
For a line through and slope is Substituting gives Choosing two unusual data points far from the main pattern can produce a poor model.
Example: distinguish observation from prediction
The line predicts 23 cm at week 6, but a measured plant is 25 cm tall. Its observed value is 2 cm above the prediction. That difference, observed minus predicted, is called a residual.
A positive residual places the point above the line; a negative residual places it below. Small residuals throughout the observed range suggest that the line describes the data reasonably well, though they do not prove a cause-and-effect relationship.
Predict within the range cautiously
Estimating between observed x-values is interpolation. Predicting far beyond them is extrapolation and can be unreliable. A plant's early growth rate may not continue indefinitely, so a line fitted over six weeks may be poor at predicting height after five years.
Association alone does not prove causation. Students who study more may differ in other ways, and a third variable can affect both measurements. A scatterplot describes a pattern in the data, not the mechanism that produced it.
Try it yourself
A line predicts At the observed y-value is 29. Find the prediction and residual.
Answer: The prediction is 32. The residual is so the observed point lies 3 units below the line.
What to remember
Keep paired observations together, describe the overall pattern, interpret slope with units, and distinguish a model's prediction from an actual observation.