The independent variable (the one you're using to make a prediction from) goes on the horizontal axis; the dependent variable (the one being predicted) goes on the vertical axis. This choice matters: swapping them implies the dependent variable is somehow causing the independent one, which is usually nonsensical in context (e.g. plotting engine size against fuel economy, not the other way round).

A complete answer always needs both direction and strength, stated together in context:

• Positive correlation: both variables increase together (points trend bottom-left to top-right).
• Negative correlation: one increases as the other decreases (points trend top-left to bottom-right).
• No correlation: points are scattered with no visible trend — a line of best fit is meaningless here and should not be drawn.

• It must be a single straight line, ruled — never curved or freehand.
• It must extend across the full range of the plotted data.
• It should be positioned so roughly equal numbers of points lie above and below it along its whole length (balance).
• It does not need to pass through the origin — its position is dictated entirely by the data, not by geometry.
• Ignore any clear outlier when judging the balance — an isolated point that breaks the pattern would otherwise distort the line away from the genuine trend of the rest of the data.

Locate the known value on one axis, trace perpendicular to the line of best fit, then read the corresponding value off the other axis.

Interpolation — predicting a value that falls within the range of the plotted data — is reliable, because it's backed by an observed trend in that exact region. Extrapolation — predicting a value outside that range — is unreliable, because it assumes the same linear trend continues indefinitely, when in reality the relationship could level off, change direction, or be affected by factors the data never captured.

Two variables can move together strongly without one directly causing the other — both might instead be driven by a hidden third variable. A complete answer names a plausible third factor: e.g. ice cream sales and sunburn cases rise together not because one causes the other, but because both are driven by hot, sunny weather.

A scatter graph plots the age of 10 used cars (years) against their resale value (£1000s). The points show a clear downward trend. One car, unusually well-maintained, sits far above the others. Describe the correlation, and explain how the outlier should be treated when drawing a line of best fit.

1. The downward trend means the variables move in opposite directions: this is a negative correlation, and the points cluster fairly closely to a line, so it should be described as strong negative correlation between age and resale value.
2. The unusually high point is an outlier — it should be visually ignored when positioning the line of best fit, since including it would pull the line away from the genuine trend shown by the rest of the data.