The Correlation project studies the visual perception of **correlation** in data visualizations.** **A data visualization is a graphical representation of a data set. For instance, scatter plots are a common choice of visualization for data with two variables.

In a scatter plot, data is translated into a graphic form by placing points on a cartesian (x-y) coordinate plane according to their values on each variable. Correlation in a scatter-plot corresponds to the degree to which the points form a straight line. Scatter plots represent the variability in a data set with a single **visual variable**, position, but there are others (ie: **size, color, texture, and brightness**) which could be used instead. For example, consider the two **ring strip-plots** below. They represent the same data-set as the scatter plots above, but they use ring-size rather than y-position to represent variability in one of the variables.

Although scatter plots are far more common, there’s no evidence of them being more effective than ring strip-plots, or any other possible alternatives. To rigorously compare visualizations we need measures for how well they enable a viewer to understand the structure of the underlying data – which is why we measure the accuracy and perception with which viewers perceive correlation.

We use two classic methods from psychophysics to derive our measures – *discrimination tasks* using the *staircase method* to measure precision, and a *magnitude estimation task* to measure accuracy. Performance in both respects is regular and well described by Weber and Fechner laws – a linear relationship for discrimination and a logarithmic curve for estimation – regardless of which visual variables are chosen to represent the data.

Our working theory for these results is based on participants using the information entropy of the visualization to make their judgements. Currently we’re studying how different gamma levels impact the perception of correlation in black and white luminance strip plots, and evaluating the effects of mixed populations in scatter plots.

## You Might Like This Project If…

- You are interested in applied research
- You want to learn about computational modelling
- You are interested in data visualization
- You are interested in the limits of visual attention
- You like to program in JavaScript (or would like to learn more about it)

#### Interested in being part of this project? We are currently not looking for new co-pilots to join the Correlation team. Please check back in the next term. Please refer to the section “Application steps” on how to apply.

## Relevant Articles:

- Rensink RA (2012). Invariance of Correlation Perception.
*Journal of Vision*, 2012;12(9):433. doi: 10.1167/12.9.433. - Rensink RA, and Baldridge G (2010). The perception of correlation in scatterplots.
*Computer Graphics Forum*,**29**: 1203-1210. - Rensink RA (2014a). On the Prospects for a Science of Visualization. In W. Huang (Ed.)
*Handbook of Human Centric Visualization: Theories, Methodologies, and Case Studies*. New York: Springer. pp. 147-175.