About Correlations
Let's say you are going to look at the link or "correlation" between "Item A" and "Item B." Correlation statistics use a scale of -1.0 to +1.0. A perfect negative correlation is shown as -1.0. This means that when Item A goes up one step, Item B goes down one step, or vice versa—correlation by itself does not show which movement causes the other. At the other extreme, a perfect positive correlation of +1.0 means when either item goes up one step, so does the other. A 0.5 correlation would mean that when one item moves one full step, the other moves only a half step. A correlation of 0 would mean there is no connection at all between the movements of the two. In short, the distance of the number from zero shows the strength of the link, and the sign (positive or negative) shows whether the items move in the same or opposite directions. When comparing two correlations, both the sign and the size of the number matter.
Be careful when consultants or journalists report on correlations. Often people make the mistake of assuming one correlated item caused the second, but again, a correlation by itself says nothing about which movement caused the other. Other types of research and statistics are needed to establish a cause—even if the timing seems obvious.
Also, not all correlations are considered useful. Scientists use a statistic called "confidence level" to see if the result could have been the result of random chance, based on the size of the correlation and number of data points. Some use a higher confidence level to declare a correlation "statistically significant"—not due to chance—and others a lower one, but they all agree correlations below that lower confidence level are not valid. In one study, a -0.2 correlation would be significant; in another, it would not; and in some cases, it depends on which confidence level the author considered significant. Most nonscientists don't know the difference. To play it safe, TeamTrainers uses the higher confidence level.
E-mail
RSS
Twitter