6 Visualizing data
There is an old statistical joke that goes like this. Before you do anything else, perform the IOT test – the Intraocular Trauma Test. To this end, plot the data. If the conclusion hits you between the eyes, the results are significant.
The oldest reference I found for the IOT is from 1963 attributes it to Joseph Berkson (Edwards, Lindman, and Savage 1963), a famous statistician, known among others for describing the Berkson’s paradox – spurious correlations caused by selection bias.
In other words: quite often a good visualization is everything you need to show your point.
Unfortunately, there is also an opposite side to this coin. We humans are incredebly adapt at finding patterns even where there are none. So if we can’t see our point on a diagram (e.g. difference between the groups on a box plot, or correlation on an x-y plot), then most likely there is no meaningful difference.
Statistics is mostly useful for all the cases in between: where we can spot the difference when looking at the plot, but it is not so large as to cause the intraocular trauma.
Since scientific and statistical illustrations play such a powerful role, a substantial part of your communication with your bioinformatitian will refer to the plots and figures they produce for you. As always, communication is the key, and having a common language helps. In this chapter, I will describe a few of the most common problems and misunderstandings.