NOS-a-Palooza: Correlations and Causality

The notion of statistics is enough to send people into a cold sweat. Stats was that math class that we all had to take but that few of us enjoyed. Many are cynical when it comes to statistics. "Oh, they can proove anything with statistics." Like 'em or not stats are a necessary evil in science where we deal with averages and try to demonstrate differences. Stats help us to look at two or more average measurements and determine if in fact they are different.

One statistical element that is useful is called the corelation co-efficient (r). The correlation Coeffient is used to tell how solid a relationship is between two variables. The maximum value is 1.0 and indicates an absolute relationship.

The graph to the left has a correlation coefficient of 0.9 (nearly perfect). The danger is to attach significance to this perceived relationship without using common sense. Its one thing if this was a graph of the distance travelled in a given amount of time. Clearly the farther you go, the longer it takes. This makes sense.

In reality, this graph is a plot of yearly steel production since 1900 in tons (horizontal axis) against the length of dresses (vertical axis). Clearly then the height of hemlines is directly related to the amount of steel produced worldwide. If we are to believe this, then the best way to increase steel production is to wear shorter dresses. Foolishness, they arent related but this demonstrated the danger of data and statistics. Both can be used for evil.

Note tha tthe sata points are not in a nice straight line. There is some degree of scatter. This indicates that the relationship is not perfect but that a trend clearly exisits.