Using Visualizations to Better Communicate Scientific Findings


Examples of non-standard charts designed to convey both proportions and uncertainty.


Summary of the Internship Topic

Part of the scientific method is based on i) running experiments (e.g., have human subjects carry out tasks and measure their performance), ii) doing statistical analyses of the data collected and then iii) communicating the results of these analyses in a research article. Although lots of work has been done by statisticians in developing reliable and powerful statistical tools, often the statistics are miscommunicated and the findings misinterpreted.

This project will investigate how information visualization and human-computer interaction can help communicate the data behind scientific findings effectively. At least two types of approaches can be considered:

1) Using new interactive visualizations as a pedagogical tool, to help scientists build an accurate intuitive understanding of difficult statistical concepts (e.g., statistical significance, correlations, or confidence intervals) and of their conventional visual representations. One example would be to generate a fictitious scientific article where the data, analysis methods and presentation techniques can be interactively manipulated.

2) Improving existing visual representation techniques, e.g., by exploring and studying hybrid chart types that can faithfully and simultaneously convey several types of statistical information such as proportions (ratios between means), variances (how much the data is spread around means), and uncertainty (probability distribution of the means). No such unified and truthful statistical representation exists yet.

Part of this work will consist in designing interactive or static visualizations, testing them experimentally (either with scientists or through crowdsourcing), and disseminating them to scientists. Although the population of users targeted are initially scientists, the student can also consider studying techniques that are targeted to the general public (e.g., how to faithfully communicate scientific findings in a newspaper article, especially regarding effect sizes).

Requirements

The student should have a genuine interest in -- and if possible some basic knowledge of:

  • Information visualization
  • Human-Computer Interaction
  • The scientific method and statistical data analysis
  • Experimental psychology

A decent level of oral and written English and some programming skills are required. Speaking French is not required.

Advising

The student will be advised by Pierre Dragicevic. Possible informal collaboration with Geoff Cumming, emeritus professor at La Trobe University.

Links

More advanced readings