Evaluating the Effect of Style in Information Visualization
- 1. Evaluating the Effect of Style in Information Visualization Andrew Vande Moere KU Leuven, Belgium Martin Tomitsch The University of Sydney, Australia Christoph Wimmer T.U.Wien, Austria Christoph Boesch T.U.Wien, Austria Thomas Grechenig T.U.Wien, Austria
- 2. Goal •visualizationimpact of style in information to measure • by comparing 3 different ‘design alternatives’ • in terms of visual and interactive style • style demonstrators based on real-world examples • then contrasted resulting insights against each other
- 3. Dataset •must be ‘agnostic’ to stylistic approach • e.g. dance music vs. cancer statistics •The New York Times news articles • containing terms ‘hope’ or ‘fear’ (4,644) • title, abstract, date, page number, news desk • extra 24 descriptive keywords
- 4. “Reversible” “Factual” Gapminder (2007) Many Eyes (2007) OECD eXplorer (2009)
- 5. “Irreversible” “Meaningful” Bitalizer (2008) Poetry on the Road (2004) Texone (2005)
- 6. Partly “Reversible” Partly “Factual” Digg Swarm (2007) ReMap (2009) We Feel Fine (2006)
- 10. Study •1. style validation study • did our 3 demonstrators correspond to the according style examples? •2. online evaluation study • between-subject design • recruitment through mailing lists on information visualization, HCI, blogs, social media, etc.
- 11. Participation 44 persons 50 persons 44 persons 3.05 level average expertise Analytical Magazine Artistic
- 12. Interaction 18m09s (average) 12m49s (average) 11m55s (average) 181.0 interactions 87.7 interactions 88.9 interactions (average) (average) (average) Analytical Magazine Artistic
- 13. No Reported Insights 1 participant 11 participants 9 participants Analytical Magazine Artistic
- 14. Interface “Insights’ 6% 6 insights 12% 13 insights 29% 27 insights Analytical Magazine Artistic
- 15. Insight Typology •insight classification (Chen et al.*) • 2 independent coders (34% agreement) • revisited classification (89% agreement) • then decided together (100% agreement) •“meaning”: added new insight class • all connotations to ‘content’ (*) Y. Chen, J. Yang and W. Ribarsky, “Toward Effective Insight Management in Visual Analytics Systems,” IEEE Pacific Visualization Symposium (PacificVis'09), IEEE, 2009, pp. 49-56.
- 16. Insight Analysis ANA MAG ART Difference 24% (24) 26% (26) 17% (11) Cluster 22% (22) 15% (15) 9% (6) Distribution 11% (11) 12% (12) 17% (11) Compound 9% (9) 14% (14) 11% (7) Trend 8% (8) 4% (4) 8% (5) Outliers 6% (6) 10% (10) 15% (10) Value 6% (6) 1% (1) 0% (0) Association 5% (5) 3% (3) 6% (4) Meaning (*) 3% (3) 4% (4) 14% (9) Extreme 4% (4) 6% (6) 0% (0) Categories 2% (2) 1% (1) 0% (0) Rank 1% (1) 2% (2) 3% (2)
- 17. Insight Analysis ANA MAG ART Difference 24% (24) 26% (26) 17% (11) Cluster 22% (22) 15% (15) 9% (6) Distribution 11% (11) 12% (12) 17% (11) Compound 9% (9) 14% (14) 11% (7) Trend 8% (8) 4% (4) 8% (5) Outliers 6% (6) 10% (10) 15% (10) Value 6% (6) 1% (1) 0% (0) Association 5% (5) 3% (3) 6% (4) Meaning (*) 3% (3) 4% (4) 14% (9) Extreme 4% (4) 6% (6) 0% (0) Categories 2% (2) 1% (1) 0% (0) Rank 1% (1) 2% (2) 3% (2)
- 18. Insight Analysis Rating (1 - 5) ANA MAG ART uncertain - confident 4.10 (1.11) 4.21 (0.87) 4.17 (0.95)
- 19. Insight Analysis Rating (1 - 5) ANA MAG ART uncertain - confident 4.10 (1.11) 4.21 (0.87) 4.17 (0.95) difficult - easy 3.78 (1.17) 3.63 (1.29) 4.00 (1.24)
- 20. Insight Analysis Rating (1 - 5) ANA MAG ART uncertain - confident 4.10 (1.11) 4.21 (0.87) 4.17 (0.95) difficult - easy 3.78 (1.17) 3.63 (1.29) 4.00 (1.24) shallow - deep 3.18 (1.10) 2.93 (1.08) 2.54 (1.17)
- 21. Insight Analysis Rating (1 - 5) ANA MAG ART uncertain - confident 4.10 (1.11) 4.21 (0.87) 4.17 (0.95) difficult - easy 3.78 (1.17) 3.63 (1.29) 4.00 (1.24) shallow - deep 3.18 (1.10) 2.93 (1.08) 2.54 (1.17) shallow – deep (expert rating) 2.44 (0.78) 2.36 (0.70) 2.28 (0.64)
- 22. Insight Analysis Rating (1 - 5) ANA MAG ART ugly - beautiful 3.48 (0.85) 3.08 (1.03) 3.11 (1.02) obtrusive - fluid 3.27 (0.95) 3.08 (1.01) 2.80 (1.00)
- 23. Insight Analysis Rating (1 - 5) ANA MAG ART ugly - beautiful 3.48 (0.85) 3.08 (1.03) 3.11 (1.02) obtrusive - fluid 3.27 (0.95) 3.08 (1.01) 2.80 (1.00) ambiguous - clear 3.39 (1.17) 1.98 (0.89) 2.00 (0.86) difficult - easy to understand 3.55 (1.04) 2.08 (1.07) 2.14 (1.07) intended inform – express 2.80 (1.15) 3.54 (1.18) 3.66 (1.06) useless - useful 3.61 (0.95) 2.70 (1.09) 2.45 (0.90) frustrating - enjoyable 3.43 (1.00) 2.54 (1.16) 2.34 (1.06) unusable - usable 3.77 (0.91) 2.78 (1.13) 2.64 (1.12) boring - engaging 3.43 (0.93) 3.10 (0.95) 2.80 (1.00) non-functional - functional 3.93 (0.82) 2.80 (1.18) 2.50 (1.13) tool - art 2.30 (1.07) 3.32 (1.19) 3.68 (0.93)
- 24. Discussion •insight classification • based on very short descriptions (M=17.86) • methodology missing to benchmark insights against each other •‘controlling’ style • are the 3 conditions representative? • e.g. similarities MAG / ART
- 25. Conclusions •style impacts perception of usability • in particular for embellished versus non- embellished styles • analytical style was perceived as more understandable, clear, enjoyable, engaging, useful, functional, ...
- 26. Conclusions •style does not impact insight depth • participants were able to overcome huge incomprehensibility issues of ART • and in a minimum amount of time
- 27. Conclusions •style has impact on ‘kind’ of insights • analytical focus of facts versus meaning of content, explanation of reasoning, ... • driven by e.g. graphic incorporation of content, fluidity of interface, ...
- 28. Guidelines •to accurately benchmark insights... • make distinction between analytical characteristics of an insight and its meaning • motivate participants to report insights in a more expansive way • e.g. insight categorization,... • allow participants to report usability issues in parallel with insights • consider alternative ways of insight analysis • e.g. card sorting, affinity diagramming,...
- 29. Thank you! Andrew Vande Moere andrew.vandemoere@asro.kuleuven.be @infosthetics
- 30. Introduction Stage continue >
- 32. Survey Stage