Data Visualization a Gateway to Statistics?

Yes! Personally, I'm a big fan of data visualization as a way of mining complex information. Pictures of data tend to contain more information than text summaries and also make associations much easier to recognize. Humans may be much less unbiased and logical than we like to think, but we're awesome pattern-detecting machines (whether there's something there or not).



Despite its limitations, subjective intuition is an important (though often unacknowledged) part of science. While you'll need watertight statistics to publish that paper, chance observation and hunches often get you to the discovery in the first place. I especially appreciate informatic approaches that fuse the best of objective and subjective approaches. This is where visualization can really contribute to data mining:

1. Project various summaries and visualizations of your data
2. Search for any patterns you can imagine
3. Idiot-check your ideas with objective statistical approaches

I really grew to appreciate this approach as I began wading through piles of non-targeted metabolomic data during my postdoc. It's never practical to annotate your data with every random fact that might be important, but they often jump out at you when you give them a chance.

It's a nice validation of the hours you spent pouring through the literature when you suddenly notice a cluster of features that you faintly remember sharing some known relationship.


The original interview with Wattenberg & Viegas

* MZmine2 is a great example of scientific data visualization done well. It guides you through the tricky process of peak picking and alignment of LC/MS spectra. Its image-heavy interface gives you an intuitive feel for when you've got things lined up right and are maximizing signal to noise. Most other programs I've used are more or less black boxes that leave you wondering what you actually did to your data. 

** Image from Circos