Visualizing Relationships - Data Points: Visualization That Means Something - page 181

Now here's the difference between correlation and causation. If you
increase the price of gas, will the price of milk go up by default? More
important, if the price of milk did go up, was it because of the increase in
the gas price or was it an outside factor, such as a dairy strike?
It's difficult to account for every outside, or confounding factor, which
makes it difficult to prove causation. Researchers spend years figuring
stuff like that out. You can, however, easily find and see correlation, which
can still be useful, as you see in the following sections.
Correlation can help you predict one metric by knowing another. To see
this relationship, return to scatterplot and multiple scatterplots.
More with Points
In Chapter 4, “Visualizing Patterns over Time,” you used a scatterplot to
graph measurements over time, where time was on the horizontal axis
and a metric of interest was on the vertical axis. This helped spot temporal
changes (or nonchanges). The relationship was between time and another
factor, or a variable. As shown in Figure 6-1, however, you can use the
scatterplot for variables other than time; you can use a scatterplot to look
for relationships between two variables.
If two metrics are positively correlated (Figure 6-2, left), dots move higher
up as you read the graph from left to right. Conversely, if a negative cor-
relation exists, the dots appear lower, moving from left to right, as shown
in the middle of Figure 6-2.
Sometimes the relationship is straightforward, such as the correlation
between peoples' height and weight. Usually, as height increases, weight
increases. Other times the correlation is not as obvious, such as that
between health and body mass index (BMI). A high BMI typically indicates
that someone is overweight; however, muscular people for example, who
can be athletically fit, could have a high BMI. What if the sample population
were body builders or football players? What would relationships between
health and BMI look like?