Successfully reported this slideshow.
We use your LinkedIn profile and activity data to personalize ads and to show you more relevant ads. You can change your ad preferences anytime.
Sources of bias when working with visualisations 
David Peebles 
University of Hudders
eld 
November 8, 2014 
David Peebles (University of Hudders
eld) Biases and visualisations November 8, 2014 1 / 16
Where's the `cognitive'? 
Using visualisations prime example of 
`embodied' cognition (Wilson, 2002). 
David Peebles (Univ...
eld) Biases and visualisations November 8, 2014 2 / 16
Where's the `cognitive'? 
Using visualisations prime example of 
`embodied' cognition (Wilson, 2002). 
Situated in real-wo...
eld) Biases and visualisations November 8, 2014 2 / 16
Where's the `cognitive'? 
Using visualisations prime example of 
`embodied' cognition (Wilson, 2002). 
Situated in real-wo...
eld) Biases and visualisations November 8, 2014 2 / 16
Where's the `cognitive'? 
Using visualisations prime example of 
`embodied' cognition (Wilson, 2002). 
Situated in real-wo...
eld) Biases and visualisations November 8, 2014 2 / 16
The embodied cognition-task-artefact triad 
Embodied 
Cognition 
Task 
Artefact 
Interactive 
Behaviour 
Interactive behav...
eld) Biases and visualisations November 8, 2014 3 / 16
Sources of bias when working with visualisations 
Bias. A systematic preference for a 
particular pattern of behaviour in ...
eld) Biases and visualisations November 8, 2014 4 / 16
Sources of bias when working with visualisations 
Bias. A systematic preference for a 
particular pattern of behaviour in ...
eld) Biases and visualisations November 8, 2014 4 / 16
Sources of bias when working with visualisations 
Bias. A systematic preference for a 
particular pattern of behaviour in ...
eld) Biases and visualisations November 8, 2014 4 / 16
Computational aordances (Peebles  Cheng, 2003) 
10 
9 
8 
7 
6 
5 
4 
3 
2 
1 
0 
Amount of Silver and Gold Produced (in t...
eld) Biases and visualisations November 8, 2014 5 / 16
Emergent/salient features 
David Peebles (University of Hudders
eld) Biases and visualisations November 8, 2014 6 / 16
Emergent features (Peebles, 2008) 
Tasks: local (1 feature) and global (all features) comparison. 
Accuracy and latency of...
eld) Biases and visualisations November 8, 2014 7 / 16
Upcoming SlideShare
Loading in …5
×

DECISIVe workshop introduction

326 views

Published on

Title: "Sources of bias when working with visualisations". Introduction to the "Dealing with Cognitive Biases in Visualisations (DECISIVe 2014) workshop at IEEE VIS, Paris, November 2014.

Published in: Data & Analytics
  • Be the first to comment

  • Be the first to like this

DECISIVe workshop introduction

  1. 1. Sources of bias when working with visualisations David Peebles University of Hudders
  2. 2. eld November 8, 2014 David Peebles (University of Hudders
  3. 3. eld) Biases and visualisations November 8, 2014 1 / 16
  4. 4. Where's the `cognitive'? Using visualisations prime example of `embodied' cognition (Wilson, 2002). David Peebles (University of Hudders
  5. 5. eld) Biases and visualisations November 8, 2014 2 / 16
  6. 6. Where's the `cognitive'? Using visualisations prime example of `embodied' cognition (Wilson, 2002). Situated in real-world environment. Inherently involves perception/action. David Peebles (University of Hudders
  7. 7. eld) Biases and visualisations November 8, 2014 2 / 16
  8. 8. Where's the `cognitive'? Using visualisations prime example of `embodied' cognition (Wilson, 2002). Situated in real-world environment. Inherently involves perception/action. Time pressured. Under pressure of real-time interaction with environment. David Peebles (University of Hudders
  9. 9. eld) Biases and visualisations November 8, 2014 2 / 16
  10. 10. Where's the `cognitive'? Using visualisations prime example of `embodied' cognition (Wilson, 2002). Situated in real-world environment. Inherently involves perception/action. Time pressured. Under pressure of real-time interaction with environment. Exploits task environment to reduce cognitive workload by holding, representing, and manipulating information. Knowledge in world combines with knowledge in head. David Peebles (University of Hudders
  11. 11. eld) Biases and visualisations November 8, 2014 2 / 16
  12. 12. The embodied cognition-task-artefact triad Embodied Cognition Task Artefact Interactive Behaviour Interactive behaviour: Emerges from dynamic interaction of goal-directed, task-driven cognition/perception/action with designed task environment. A complex combination of bottom-up stimulus driven and top-down goal and knowledge driven processes. Makes little sense to consider and investigate cognition in isolation. David Peebles (University of Hudders
  13. 13. eld) Biases and visualisations November 8, 2014 3 / 16
  14. 14. Sources of bias when working with visualisations Bias. A systematic preference for a particular pattern of behaviour in a task environment. David Peebles (University of Hudders
  15. 15. eld) Biases and visualisations November 8, 2014 4 / 16
  16. 16. Sources of bias when working with visualisations Bias. A systematic preference for a particular pattern of behaviour in a task environment. Question: How are preferences (and interpretations) created/shaped by the embodied cognition-task-artefact triad? David Peebles (University of Hudders
  17. 17. eld) Biases and visualisations November 8, 2014 4 / 16
  18. 18. Sources of bias when working with visualisations Bias. A systematic preference for a particular pattern of behaviour in a task environment. Question: How are preferences (and interpretations) created/shaped by the embodied cognition-task-artefact triad? Four sources of bias: Computational aordances Emergent/salient features Gestalt principles of perceptual organisation Adaptive behaviour David Peebles (University of Hudders
  19. 19. eld) Biases and visualisations November 8, 2014 4 / 16
  20. 20. Computational aordances (Peebles Cheng, 2003) 10 9 8 7 6 5 4 3 2 1 0 Amount of Silver and Gold Produced (in tonnes) by ‘Precious Metals Plc.’ in 1996 Gold Jan FebMar AprMayJun Jul AugSepOctNovDec Amount Produced Month Silver Informationally equivalent Computationally inequivalent. Require dierent procedures. 10 9 8 7 6 5 4 3 2 1 0 Amount of Silver and Gold Produced (in tonnes) by ‘Precious Metals Plc.’ in 1996 0 1 2 3 4 5 6 7 8 9 10 Gold Silver Jan Feb Mar Apr May Jun Jul Aug Sep Oct Nov Dec When gold is 4, what is the value of silver? Which two months have the same values of silver and gold? David Peebles (University of Hudders
  21. 21. eld) Biases and visualisations November 8, 2014 5 / 16
  22. 22. Emergent/salient features David Peebles (University of Hudders
  23. 23. eld) Biases and visualisations November 8, 2014 6 / 16
  24. 24. Emergent features (Peebles, 2008) Tasks: local (1 feature) and global (all features) comparison. Accuracy and latency of comparison judgements aected by: Representation used. Value being compared. Emergent features created by arrangements of values. David Peebles (University of Hudders
  25. 25. eld) Biases and visualisations November 8, 2014 7 / 16
  26. 26. Emergent features aect distance perception (a) Transport = 1:91 (b) Transport = 2:14 David Peebles (University of Hudders
  27. 27. eld) Biases and visualisations November 8, 2014 8 / 16
  28. 28. Gestalt principles (Ali Peebles, 2013) Laws of perceptual organisation (e.g., proximity, similarity, continuity, connectedness, common fate) aect grouping of graphical elements. Percent Error as a function of Experience and Time of Day Low High 100 90 80 70 60 50 40 30 20 10 ● ● Percent Error Experience Time of Day Day Night Percent Error as a function of Experience and Time of Day Percent Error Low High 100 90 80 70 60 50 40 30 20 10 Experience Time of Day Day Night Line: Novices focus primarily on legend variable (connectedness and similarity). Bar: Novices' attention balanced between legend and x-axis variables. David Peebles (University of Hudders
  29. 29. eld) Biases and visualisations November 8, 2014 9 / 16
  30. 30. Using emergent features and Gestalt principles AA AB 500 450 400 350 300 250 200 150 100 50 Response Time as a function of Task and Stimulus Type Response Time Task Stimulus Type Words Pictures (a) Crossover interaction 100 90 80 70 60 50 40 30 20 10 Percent Error as a function of Experience and Time of Day Percent Error Low High Experience Time of Day Day Night (b) Colour-match graph (a) Experts learn emergent features for rapid pattern recognition. (b) Knowledge of Gestalt principles can be used to design more eective representations (Ali Peebles, 2013). David Peebles (University of Hudders
  31. 31. eld) Biases and visualisations November 8, 2014 10 / 16
  32. 32. Basic elements of visualisation behaviour Cognitive processes: Goal setting and monitoring, initiate visual search, retrieval from LTM, generating inferences, etc. David Peebles (University of Hudders
  33. 33. eld) Biases and visualisations November 8, 2014 11 / 16
  34. 34. Basic elements of visualisation behaviour Cognitive processes: Goal setting and monitoring, initiate visual search, retrieval from LTM, generating inferences, etc. Perceptual processes. Judgement of length, direction, area, position on common scale (Cleveland McGill, 1984). Visual comparisons of length, colour, shape, quantity. David Peebles (University of Hudders
  35. 35. eld) Biases and visualisations November 8, 2014 11 / 16
  36. 36. Basic elements of visualisation behaviour Cognitive processes: Goal setting and monitoring, initiate visual search, retrieval from LTM, generating inferences, etc. Perceptual processes. Judgement of length, direction, area, position on common scale (Cleveland McGill, 1984). Visual comparisons of length, colour, shape, quantity. Physical actions: Eye movement saccades and
  37. 37. xations, mouse clicks and cursor movements,
  38. 38. nger taps and pinches etc. Interface manipulations: Selection/highlighting with mouse; dragging, realigning, rotating, deleting; zooming in and out David Peebles (University of Hudders
  39. 39. eld) Biases and visualisations November 8, 2014 11 / 16
  40. 40. Interactive routines Basic cognitive, perceptual and motor operators combined into interactive routines that take between 0.3 to 1 second to execute: Direct attention to object and encode features/location. Move mouse cursor to graphical object and click on it (see CPM-GOMS model below; Gray Boehm-Davis, 2000). David Peebles (University of Hudders
  41. 41. eld) Biases and visualisations November 8, 2014 12 / 16
  42. 42. Unit tasks and strategy selection Unit tasks: Combinations of interactive routines that perform subtasks. Typical execution time: between 3s and 30s. Select and visually mark subset of data. Locate variable value(s) according to some criterion (e.g., max, min, median etc.). Attend click object Look for 'Done' button Click on 'Done' button Encode object features seek next object Yes No Attend click object Encode object features request retrieval Blended location retrieved, attend to location Relocate object seek next object Relocation Presentation Current phase? Unattended object? Yes No Unattended object? David Peebles (University of Hudders
  43. 43. eld) Biases and visualisations November 8, 2014 13 / 16
  44. 44. Unit tasks and strategy selection Unit tasks: Combinations of interactive routines that perform subtasks. Typical execution time: between 3s and 30s. Select and visually mark subset of data. Locate variable value(s) according to some criterion (e.g., max, min, median etc.). Strategies: Sequences of unit tasks. Attend click object Look for 'Done' button Click on 'Done' button Encode object features seek next object Yes No Attend click object Encode object features request retrieval Blended location retrieved, attend to location Relocate object seek next object Relocation Presentation Current phase? Unattended object? Yes No Unattended object? David Peebles (University of Hudders
  45. 45. eld) Biases and visualisations November 8, 2014 13 / 16
  46. 46. Adaptive behaviour Interactive routine changes reduce interaction cost optimise control system resource allocation. Milliseconds matter (Gray Boehm-Davis, 2000) Attend click object Look for 'Done' button Click on 'Done' button Encode object features seek next object Yes No Attend click object Encode object features request retrieval Blended location retrieved, attend to location Relocate object seek next object Relocation Presentation Current phase? Unattended object? Yes No Unattended object? David Peebles (University of Hudders
  47. 47. eld) Biases and visualisations November 8, 2014 14 / 16
  48. 48. Adaptive behaviour Interactive routine changes reduce interaction cost optimise control system resource allocation. Milliseconds matter (Gray Boehm-Davis, 2000) Ballard et al. (1995) increased information access cost from eye movement to head movement { users shifted from display based access to memory retrieval. Attend click object Look for 'Done' button Click on 'Done' button Encode object features seek next object Yes No Attend click object Encode object features request retrieval Blended location retrieved, attend to location Relocate object seek next object Relocation Presentation Current phase? Unattended object? Yes No Unattended object? David Peebles (University of Hudders
  49. 49. eld) Biases and visualisations November 8, 2014 14 / 16
  50. 50. Adaptation as a source of bias Adoption and adaptation determined by: Cost structure of the task environment (i.e., how quick/easy is it to execute the task). The representational eciency of the visualisation. History of success with the strategy. Adaptation typically beneath conscious awareness and deliberate control. Do users always adapt to an optimal interaction? Not necessarily { only when local (interactive routine level) optimum coincides with global task level optimum (Fu Gray, 2004). Suboptimal interaction and interpretation may result from strategy selection pressures resulting from unconscious choices made at the embodiment level. David Peebles (University of Hudders
  51. 51. eld) Biases and visualisations November 8, 2014 15 / 16
  52. 52. Summary Interpretation of data graphics can be shaped/biased by number of factors: Visual and computational properties of the representation. Adaptive behaviour of the user seeking to minimise eort. Research on simpler data graphics must be extended to: Larger and more complex data sets. Broader class of visualisations. Cover the increasing variety of interactions and manipulations that are being developed. Vital for design of most appropriate representations/task environments for dierent users and tasks and to reduce error. Important role for cognitive science theories and methods in this research (e.g., computational cognitive modelling) David Peebles (University of Hudders
  53. 53. eld) Biases and visualisations November 8, 2014 16 / 16

×