Saturday, July 07, 2012

Plot the data!

I've heard the injunction "Plot the data" for decades.  Plotting the data is important throughout statistical analyses, from initial looks to surmise what model might fit the data to looking at residual plots to check the model or see what might be a good next step. Plotting the reference mode or reference behavior pattern of a problem is a key beginning step in the application of system dynamics.

At first, graphs were done manually with colored pencils and graph paper, then there were HP's great pen plotters, and then we got today's almost universal system capability of displaying and printing high-resolution graphics.  We're indebted to those such as William Cleveland and Edward Tufte for their research-based insights on ways to generate helpful, honest graphics.

I also understand the utility of tables of data, at least in the right circumstances. In that regard, I like Andrew Ehrenberg's advice in his The Problem of Numeracy, and his advice on reading tables is worth reading, too.

Now there's a bit of research that may help in the decision whether to use a graph or a table for displaying data.  Economists Are Overconfident. So Are You from the HBR Blog Network points to research that indicates that economists do better at understanding regression analyses when only shown graphs than when shown graphs and numbers and much better than when only shown numbers.

So plot the data already!

Labels: , , , , , ,


Blogger Bill Harris said...

After posting this, I found Felix Salmon's How economists get tripped up by statistics at nice explanation of the message!

10 July, 2012 21:28  

Post a Comment

<< Home