Lying with statistics

Last week I told you about an article that brought back some old memories on how to lie with statistics.

One of the most common tricks used is to show small changes as huge by not using zero as the base in a diagram. This trick is used all the time in newspapers.

A more advanced trick is to use index years and then pick a time period that shows either an increase or a decrease depending on what you want to show. In the following example different years were chosen as the index = 100 year. And since there was a recent decline choosing a shorter period to show a huge decline is very effective. Especially combined with the trick above.

If you are a hard core statistician you also know that in a bar graph it is typically not only the height of the bar that counts but really the area of the bar that defines its value. In the following example the area of the green (right) bar is almost double that of the red (left) bar. Yet it is easy to interpret them as only a small increase.

And these are only a few tricks possible that works well on graphs. You also have the option of excluding important data. For example you can say that "the risk of being infected by [insert your choice of horrible disease here] is 50% higher if [insert something you like to do here]". But what if the initial risk was only 0.01%. That means the increase risk is 0.015%. That does not sound so horrible does it? But it is 50% higher risk...