Designing with the User in mind

1. Designing with the User in mind Jamie Starke

2. Sizing the Horizon: The Effects of Chart Size and Layering on the Graphical Perception of Time Series Visualizations • J. Heer, N. Kong, M. Agrawala (2009). CI 2009 • Rethinking Visualization: A High-Level Taxonomy • Melanie Tory and TorstenMoller. InfoVis 2004

3. Sizing the Horizon: The Effects of Chart Size and Layering on the Graphical Perception of Time Series Visualizations • J. Heer, N. Kong, M. Agrawala (2009). CI 2009 • Rethinking Visualization: A High-Level Taxonomy • Melanie Tory and TorstenMoller. InfoVis 2004

5. Why? • Analysts often need to compare a large number of time series • Finance • Stocks, Exchange rates • Science • Temperatures, Polution levels • Public Policy • Crime Rates

6. Goal • Effective Presentation of multiple time series • Increase the amount of data with which human analysts can effectively work • Maximize data density (Tufte)

7. Goal • Effective Presentation of multiple time series • Increase the amount of data with which human analysts can effectively work • Maximize data density (Tufte) Increased Data Density DOES NOT IMPLY Increased Perception

11. Graphical Perception • Color hue ranks highly for nominal (category) data but poorly for quantitative data • Bertin

12. Line Charts http://coralreefwatch.noaa.gov

13. Line Charts Overlap reduces legibility of individual time series http://coralreefwatch.noaa.gov

14. Line Charts Overlap reduces legibility of individual time series Small Multiples? http://coralreefwatch.noaa.gov

15. Stacked Time Series http://www.babynamewizard.com

16. Stacked Time Series Not informative aggregation for many data types or negative values http://www.babynamewizard.com

17. Stacked Time Series Not informative aggregation for many data types or negative values Comparisons involve length rather than more accurate position judgements http://www.babynamewizard.com

18. Animation http://graphs.gapminder.org

19. Animation Animation results in significantly lower accuracy in analytic tasks compared to small multiples of static charts http://graphs.gapminder.org

20. Horizon Graphs

21. Horizon Graphs

22. Horizon Graphs

23. Horizon Graphs

24. Horizon Graphs Both use Layered Position encoding of values

25. Horizon Graphs Comparison across Band requires mental unstacking Both use Layered Position encoding of values

26. Horizon Graphs Comparison across Band requires mental unstacking Both mirror and offset show promise for increasing data density Both use Layered Position encoding of values

27. Evaluation • How much does chart sizing and layering have on speed and accuracy of graphical perception • 2 experiments • Tasks: Discrimination and estimation tasks for points on time series graphs • Determine the impact of band number and horizon graph variant (mirrored or offset) on value comparisons between horizon graphs • Compare line charts to horizon graphs and investigate the effect of chart height on both • Used 80% trimmed means to analyze estimation time and accuracy

28. Discrimination and Estimation tasks

29. Discrimination and Estimation tasks Which is bigger?

30. Discrimination and Estimation tasks Which is bigger? What is the Absolute Difference?

31. Experiment 1: Questions • How does the choice of mirrored or offset horizon graph affect estimation time or accuracy? • How does the number of bands in a horizon chart affect estimation time or accuracy?

32. Experiment 1: Hypotheses • Offset graphs would result in faster, more accurate comparisons than mirror graphs, as offset graphs do not require mentally flipping negative values • Increasing the number of bands would increase estimation time and decrease accuracy across graph variants

33. Experiment 1: Bands

34. Experiment 1: Estimation Error

35. Experiment 1: Estimation Error No significant difference between 2 and 3 bands

36. Experiment 1: Estimation Error So Significant difference between Offset and Mirror charts No significant difference between 2 and 3 bands

37. Experiment 1: Estimation Time

38. Experiment 1: Estimation Time Estimation time increases as the bands increase

39. Experiment 1: Observations • As band count rose, participants experienced difficulty identifying and remembering which band contained a value and that performing mental math became fatiguing • Working with ranges of 33 values in the 3-band condition was more difficult than working with the ranges in the 2 and 4 band that were multiples of 5

40. Experiment 2: Questions • How do mirroring and layering affect estimation time and accuracy compared to line charts? • How does chart size affect estimation time and accuracy?

41. Experiment 2: Hypotheses • At larger chart heights line charts would be faster and more accurate than mirror charts both with and without banding, and mirror charts without banding would be faster and more accurate than those with banding • As chart heights decreased, error would increase monotonically, but would do so unevenly across chart types due to their differing data densities.

42. Experiment 2: Chart Type

43. Experiment 2: Estimation error

44. Experiment 2: Estimation error Disadvantage of line chart compared to both mirrored charts

45. Experiment 2: Estimation error Accuracy decreased at smaller chart heights Disadvantage of line chart compared to both mirrored charts

46. Experiment 2: Estimation error Accuracy decreased at smaller chart heights 2 band remained stable at lower heights Disadvantage of line chart compared to both mirrored charts

47. Experiment 2: Estimation Error

48. Experiment 2: Estimation Error 2-Band has lower baseline error rate, but higher virtual resolution at a the same resolution

49. Experiment 2: Estimation Error 2-Band has lower baseline error rate, but higher virtual resolution at a the same resolution Banded mirrored charts had nearly identical error levels at matching virtual resolution

50. Experiment 2: Estimation Time