Comment by spikels | Hacker News Reader

Comment by spikels | Hacker News Reader

There are two distinct types of data visualizations: ones that seek to understand data and ones that use data to convince.

Tufte believes strongly in the first: we should strive for understanding the data and avoid distorting what the data has to say. He calls this "graphical integrity". This is how scientists use data visualization.

However most data visualization is of the second type: decide what you want to say first then use data visualization to support your point. After all we have products/services/policies to sell. This type of data visualization is so prevalent that if a visualization doesn't make an obvious point we are confused. PowerPoint is simply a product designed for this greater demand.

We should not be surprised that the author of this piece, former Booz Allen consultant to the government now MailChimp "data scientist", seems to be in the more common camp. Perhaps his visualizations are mostly used to sell clients on MailChimp.

It's not just about selling. I was a journalist for many years, and the worst kinds of visualizations and charts were just data dumps. The best kind had a point of view, or tried to make sense of the data and tell a story with it.

Now, if you distorted or hid conflicting data in the process of doing this, that's obviously bad; but data presented with a point of view can be good, because humans innately respond to stories. It's up to the reader/viewer/listener to think critically about the story.

As best I can tell Tufte thinks "data dumps" are the best kinds of visualizations: "Above all else show the data". Unlike salesman, campaigners or reporters Tufte doesn't think good data visualization is about getting a response from the viewer. He thinks we should show the data as best we can so that the viewer can learn from it. I like to think of this as the viewpoint of a scientist or researcher.

This is a rather extreme perspective and probably not effective in most situations where data visualization is used. The techniques used in the famous book "How to Lie with Statistics" are actually mostly about data visualization and are much more practical. This explains their popularity and that of tools that emphasize them like PowerPoint.

From Tufte's point of view we have a lemons market in data visualization: "bad" visualization drives out "good" visualization because it is more effective[1]. Editors, businessmen and politicians will always choose the visualization that empasises their point (i.e. distorts the underlying data). While they are happy with the results Tufte is dismayed.

[1] http://en.wikipedia.org/wiki/The_Market_for_Lemons

Reply on news.ycombinator.com