Notes on Mike Dewar’s talk — Open Value
I watched a talk by Mike Dewar on understanding NYTimes readers behavior.
There are a few points that struck me.
- He argues that we should not only open source or open algorithm. We should be upfront about value — why we do the specific visualizations or models.
- One of the values is not to aggregate or sample the data. Rather it tries to visualize data to generate questions. This is quite different from the perspective of survey samples.
- Question-generating visualizations can be different from question-answering visualizations! This echoes with Hadley’s point about visualizations are to surprise analysts.
- He treats session as unit of analysis.
- He also suggested to segment users not by demographics or taste but by reading behavior. He hopes to make these segmentation algorithm more mainstream.
- His idea of open value seems like developing a theory behind what our data analysis is doing.
Here’s the link to the talk