For each student in a music class, record the hours spent practicing in a week and the score at the recital, then plot one dot per student. If the dots drift up from left to right, more practice tends to go with higher scores: a positive correlation. If they drift down, as with outdoor temperature and a home's heating bill, the correlation is negative. If they form a shapeless cloud, as with shoe size and spelling score, there is little or no correlation.
Direction is only half the story; strength is how tightly the dots follow a line. Picture two classes whose dots both rise. In one, every dot sits close to a straight line, so a student's practice time predicts their score well, and r might be 0.9. In the other the dots rise but scatter widely, and r might be 0.4. A value of exactly 1 or −1 means every dot lies on a line; 0 means no straight-line pattern at all.
Correlation is not causation. Fires where more fire trucks show up tend to do more damage, but the trucks do not cause the damage: bigger fires bring more trucks and also do more damage. A third factor can make two things rise together. Correlation also measures only straight-line patterns: a thrown ball's height plotted against time makes a perfect arch, yet its r can be close to 0.
Think of it like this. Two correlated variables are like two dancers. With positive correlation, when one steps forward the other usually steps forward too; with negative, one steps back as the other steps forward. The tighter they stay in step, the closer r is to 1 or −1. Dancing in step does not tell you who is leading, or whether both are just following the music.
Core idea. Correlation describes how two variables move together; it does not by itself prove cause.