From an article on gun violence and legislative restrictions on firearms, found via AQF. The article shows how to do correlation right—with graphs, with several dimensions of analysis, and with very little interest for the correlation coefficients or the p-values themselves. A nice example of the intra-ocular trauma test.
Last year, I had the idea of using the Allourideas pairwise comparison software to crowdsource a ranking of countries by their degree of “democracy” in 2010. I asked students in my Dictatorships and Revolutions class to set the ball rolling, and then posted the link to the widget here, allowing anybody to vote. Surprisingly, in just a couple of months of voting the results were interestingly close to standard indexes of freedom or democracy: the crowdsourced ranking had a correlation of 0.84 with Freedom House’s widely-used ratings of political and civil liberties for 2010, and the basic crowdsourced ranking was generally plausible (see this post for further analysis of these results) . In fact, by now the correlation has increased to 0.86 (8556 votes total), which is about as good the correlation between Freedom House and Polity IV (0.87).Abandoned Footnotes: Crowdsourcing a Democracy Index – 2012 edition
“Think about it. Imagine how strong the “true” relationship must be if it shows up even when using only rough proxies for the “true” levels of inequality and immobility. In light of Winship’s criticisms, the high correlation in this chart is all the more remarkable. If his gripes are correct, then graph understates the correlation between inequality and mobility.” (via Freakonomics » Is Higher Income Inequality Associated with Lower Intergenerational Mobility?)
Taken from the NYTWrites text mining application, based on who writes about what for the New York Times. Call it Zeitgeist if you prefer. Irene Ros, a visualization research developer at the Visual Communication Lab, a division of IBM Research, has done many other beautiful things with computer technology.
You might be thinking that the underlying method of thematic proximity must use some form of correlation to sort out the connections and their respective strength. You would be right to think so.