Interesting paper entitled "Error Bars Considered Harmful: Exploring Alternate Encodings for Mean and Error" by Michael Correll and Michael Gleicher, both from the Department of Computer Sciences, University of Wisconsin-Madison, USA.
Essentially, their work tackles the public perception of error bars, noting that bar charts using conventional error bars suffer from two main problems :
Their proposed solution is that a more nuanced way of representing error probability is used, summarised in this image from the paper:
It's all interesting stuff, and it is great to read that the public comes as being able to robustly interpret data - so long as the data meets them half way!
Essentially, their work tackles the public perception of error bars, noting that bar charts using conventional error bars suffer from two main problems :
Within-the-bar bias: the glyph of a bar provides a false metaphor of containment, where values within the bar are seen as likelier than values outside the bar.
Binary interpretation: values are within the margins of error, or they are not. This makes it difficult for viewers to confidently make detailed inferences about outcomes, and also makes viewers overestimate effect sizes in comparisons.
Their proposed solution is that a more nuanced way of representing error probability is used, summarised in this image from the paper:
 |
| Ways of representing probability of error |
It's all interesting stuff, and it is great to read that the public comes as being able to robustly interpret data - so long as the data meets them half way!
No comments:
Post a Comment