This page will contain information associated with our publication for InfoVis 2014:
Error Bars Considered Harmful: Exploring Alternate Encodings for Mean and Error.
Questions requests not covered by this page can be directed to the paper authors.
Make Your Own
There is also Processing code for generating png figures, available here (zip file, 863 kB). You will need the Apache Commons Math Java library for these exmaples. A version of the library that can be dropped into the "libraries" folder of your Processing sketch folder is available here.
We conducted a series of experiments for this paper. This website provides details about these experiments not included in the paper for reasons of space or relevancy, as well as information for reproducibility.
A sample version of the experiment run in this paper (with slightly fewer stimuli for the purpose of convenience) is kept running here.
All stimuli used in our experiment are available here.
Simplified versions of the experiments (without consent forms, training, and with participant feedback) will be available here, as well as the associated stimuli for the experiment sets.
Data tables are available here. The collected columns are listed below. Other columns (such as the cdf,pdf, and p-value) are derived from these values.
|id||A pseudo-random anonymous participant idenitifier.|
|index||The question number of this particular trial for the participant.|
|risk||The participant's self-reported risk tolerance (higher=more tolerance).|
|dotDelta or delta||The difference between means, or the difference between the mean and the potential outcome|
|encoding||The type of encoding seen (0=bar,1=gradient,2=violin,5=modified box plot).|
|wording||The problem framing seen (0=election, 1=weather, 2=finance).|
|margin||The full length of the 95% confidence interval, *10|
|prediction||What the participants predicted. For one sample experiments, -1=the outcome will be lower than the potential, 1=the outcome will be higher. For two sample, -1=A will be higher, 1=B will be higher|
|confidence||How confident the participants were in their prediction, from 1=least, 7=most|
|outcome||How likely or severe an outcome the participants predicted. For the one sample test (1= the potential outcome is very unlikely, 7=the potential outcome is very likely). For two sample, -3=A will be overwhelmingly higher than B, 3=B will be overwhelmingly higher than A.|