Here are my notes on and from CSI 703: Statistical Scientific Visualization / Human centered Data Science. The course was really important in training me in data literacy and cognitive visual perception, that I apply to everything – from my note-taking, Tableau dashboards, to Instagram stories (Although sure, those are still very heady with information and my annotated thoughts)
Presentation and Exploration of the visuals is for the viewer’s perception and cognition. It is the “Aha! I see!” moment.
Terrence W. Deacon said the transformation of the senses into the inner symbolic language that Steven Pinker, Harvard psychologist calls “mentalese”.
In the Isaac Asimov short story, Sally, the cars seem to have a language of their own which makes the protagonist question the existence of a mentalese in the car’s “cararity”. This is the mentalese of the cars. The hypothetical mental system, resembling language, in which concepts can be pictured and combined without the use of words.
In Eternal Sunshine of the Spotless mind, the brain is reworked much like in Cognitive Behavioral Therapy methodology into forgetting certain triggers; in the movies’ case, the loved ones’. This is all to say, the brain processes the mental images according to what is communicated.
I instantly feel attracted to the interdisciplinary nature of a visual journalist’s book which also includes their life experiences. Curiosity in the world around me, I possess, infinitely. I heard of Cairo, and read about him for my previous blog post. The senses and thinking mechanisms are similar.
I like the idea of systematic and exciting intellectual chaos. Cairo knows how to lasso me into this field straight off the bat. Rudolf Arheim Visual Thinking – Western Philosophy Parmenides and Plato distrusted the senses – Is that used in Marketing I wonder.
Telea, Munzner, and Cairo all agree that the purpose of visualization is to gain insight into a problem. While Cairo balances a practical application in visual journalism with philosophical themes of communication in visualization, Telea and Munzner have a more pedagogical take on the same subject. They all concur that infinite curiosity is a requirement in this field and Cairo even goes on to demonstrate that with his UN data on population statistics. He calls for rational optimism for information architects, as the archaic term for data visualization engineers would be, to close the gap between comprehending the data available without the human brain trying to process the observations on its own. Telea has a noticeably mathematical approach to the field, given that the fields encompassing Visualization include but are not restricted to – Computer Science, Cognitive Science, Perception Science, Engineering in the form of Signal theory, Imaging, Computer Graphics, and Statistics. Such a multi-disciplinary subject seems especially fascinating but that is not to say computers can do it all on their own. The goal of visualization being insight is an entirely human endeavour which can only be aided by computational tools. As Munzner elaborates, human judgement and decision-making plays one of the most important roles in the process of visualization and even afterwards. For enhanced engagement in this, interactivity can manipulate data for effective storytelling even as validation of such manipulation techniques can be difficult.
I find Cairo’s persuasions fascinating due to my interest in Data Journalism and Digital Humanities, but Telea’s Visualization pipeline and 3D modeling or data mapping was creatively stimulating too. I had not previously considered the range of possibilities in data visualization. I learned the structure of visualization applications with OpenGL graphics library in C++ to have a sequential data flow design. I had also not encountered the term Vis Idiom Design and Grouse vis tool before and found the terminology in each context interesting.
There are many data visualization tutorials out there. For each programming language tool and style. What does not exist however, is the natural tendency to visualize ethically, morally, and aesthetically.
“Aesthetics do matter, but aesthetics without a solid backbone made of good content is just artifice.”Alberto Cairo
From The Representation of Numbers, I find the relation between numerical displays being visual in human cognition fascinating for its interdisciplinary nature. Visualization is as much an art of scientific application as much processing perceived information intuitively in a dynamic display. External stimuli contribute to distributed cognitive tasks. The process of representation of Arabic numerals, commonplace numbers and the multiplication of such numbers versus many other systems of numerical representations was very interesting. There are various sub-processes at play, like internal memorial processes and external perceptual representations and the central control which coordinates between the two. The cognitive study of how a visualization would be helpful in understanding what kind of symbolic and numerical representation would be more comprehensible in systems. Munzner and Telea focused on data, datasets, their types, and their usage for their efficient implementation in visualization. The type of data dictates the way it can be visualized. Munzner offers key versus value semantics as a classification of attributes. Both refer to continuous and discrete data and their various forms including even time-series data. They are an updated version, of sorts, from The Eyes have It paper’s take on data type taxonomy. It focuses mainly on the visual information seeking mantra which includes overviewing the data, focused zooming, filtering, and picking up the details depending on their demand.
While Cairo focused on how the human eye works to perceive the world around us, Munzner concentrated on the visuals in graphs. Cairo’s diagram on how perception really works in figure 5.12 is especially interesting when we learn about how visualization can be manipulated in various ways to better aid the perception process in Munzner’s figure 10.9 and figure 10.10. There were a lot of technical details covered on the attributes of colour, shape, and structure of data here. It must be noted that several of the links in Munzner’s book are now obsolete.
A tour through Visualization Zoo expanded the types and fields within which visualization can be found in. There are many techniques to visualize complex forms of data. Each type is important for specific questions to be answered. It also depends on the economic model of the visualization as is explained in the Value of Visualization. Visualization is a subjective study given all the different factors involved and its paradoxical nature for validation.
As someone who has only a novice understanding of the US elections process, I had to do a lot of extra reading, which is indicative of the intended target audience. I noticed that the visualization is tilted and in 3D, which seemed off-balanced to me. The visualization storytelling is linear. The breakdown of district wise data in each state helped understand on a smaller scale but it is not comparable.
I find the philosophy of the Gestalt principles by Kurt Koffka fascinating because of the interplay it brings to the forefront in between human psychology and physiology. There are many principles which explain how easily we can be tricked or manipulated visually. These are similarity, proximity, connection, and enclosure. The human mind fills in any gaps in knowledge with supposedly rational explanations to help us perceive everything around us to produce meaningful information that we can actually utilize for our own purposes. This, potentially, is very useful in the field of data visualization.
The paper, On the Theory of Scales of Measurement expounds the measurement of human sensory output from an objective standpoint. It seeks to classify them according to the order of magnitude of the output. This reminded me of the table we came across in class where visual processing took the greatest precedence in understanding a visual display of information. Each category of this scale picks up on a utility with which the information to be presented, is best represented. For example, nominal scale is for any amount of numbers, while ordinal scale requires data to be in an order for its linearity to be clear. Interval scale is quantitative and ratio scale is relational or dependent on other factors. The paper then goes on to make an interesting point on scale being relative itself to multiple external and internal factors like bias, or precision levels, which cannot be measured individually in each varying case to help in data visualization research.
The Cognitive Coprocessor Architecture for Interactive User Interfaces by Robertson et al. paper is from 1989 and it shows, given the advances in the fields of NLP and interactive visualization as well as the fundamental cognition differences in human understanding of technology itself. The paper discusses a newly developed architecture encouraging information visualizations to be more animated. This is to solve the multiple agent problem and animation problem, which are best described as the disproportionate “direct conversations” possible among a user, the user interface, and the application in a human-computer interaction in a seamless manner. Animating this might alleviate the tension between these components, however, it requires heavier computational resources and time from both the user and the visualizer. The paper proposes the Information Visualizer prototype based on the Cognitive CoProcessor architecture as the solution which balances these requirements with functionality by including an animation loop within the user discourse machine, or the user interface integrated with 2D and 3D animated visuals and interactive buttons extensions for asynchronous agents.
The Animation: Can It Facilitate? paper by Tversky et al. argues for effective graphics to follow the congruence principle. That is not to say that animated graphics are better at conveying information compared to the static graphics, according to research. This is because they violate the Apprehension principle of good graphics where animations might be too complex to be truly understandable. The paper goes on to argue that the reason animated visualizations were successful compared to static versions in certain cases, was due to the superior visualization quality itself. Congruence principle and Apprehension Principle together form a justification against unnecessary or stylistic animations of graphics in visualization. While interactivity might play a dominant role in perception and comprehension it also allows for a deeper level of engagement with the material for the viewer.
Munzner re-titles basic elements in visualization to make them formal and technical. A mark can be considered the most basic unit of a visualization – a geometric object that occupies spatial dimensions. A channel is the way the mark is perceived regardless of the dimensionality. Points, Lines, and Areas are marks that can be used as items or nodes. Containment space and connections are marks used as links. Channels can be varied in identity and magnitude. The effectiveness of a channel varies by the aligning spatial position, or no aligning the positions at all. Similarly, area, depth, luminance, saturation, region, motion, hue, and shape can also be varied for better perception. Channels can help distinguish and group similar types of elements from dissimilar elements. Other size channels include size, angle, curvature. Other identity channels are shape, and motion. Texture and stippling are a mixture of multiple channels. They can all contribute to better understanding of the type of data presented highlighting their separability.
Munzner further argues against unnecessary usage of 3D visualizations especially when the costs in computational, processing, time, and cognitive effort is not justified. The costs are enhanced by occlusion, perspective distortion, shadows, lighting, familiar size, stereoscopic disparity, etc. which contribute to the 3D-depth cues of the visual. The familiar planar structure of a 2D visual is from the spatial positions of legible coordinates. On the other hand, the depth effects of 3D planes are not immediately intuitive on a 2D viewing surface unlike the way humans perceive the real world. 3D shapes do have the advantage of representing shape objects better than in 2D. Moreover, a visualization in 2D must likewise be justified as opposed to a 1D visual. Lists and tables in 1D can inform the viewer of the important details, unless the relationships and inferences are explicitly observed better via a 2D framework. Munzner also expounds the effects of animation. What is visually apparent in front of a viewer is more likely to be comprehensive compared to relying on their memory given the cognitive load it would add to the process of understanding a visual. Thus, a side-by-side view of a graphic might be more persuasive for a viewer than an animated visual. Techniques employing change blindness can work to the visualizer’s advantage. Finally, 3D visuals, as well as virtual reality visuals face the encumbrance of compromising resolution for the sake of immersion. A visual that works in black and white, might as well stay so, as Maureen Stone promoted: “Get It Right in Black and White” which echoes Edward Tufte’s minimalism plea in visualization.
Munzner discusses visual encoding design choices to arrange tabular data spatially. Arrange design choice spatial channels for visual encoding. Keys are an independent attribute with an unique index to look up items in a table and Values are the dependent variable for a cell in a table. Categorical Regions are the space used for categorical attributes. Design Principles for Visual Communication explains that the best visualizations use the human facility for visual understanding to the maximum possible extent. The streamlining for a good visualization starts with identifying the best suited domain-specific requirements, then encoding those principles into the interface, after which the effectiveness is evaluated with user feedback. The paper applies this method to highly specific domains like map-making and technical illustrations, to great effect. Design of data figures understands that in life sciences communication, authors encode information that readers decode at-a-glance. With respecto the history of visualizations, French cartographer Jacques Bertin provided a theoretical framework on the visual properties of graphical elements. Cleveland and McGill measured people’s abilities to process the perceptual tasks. Low-Level Components of Analytic Activity in Information Visualization expounds the techniques for guiding the viewer to focus on what the visualizer wants to convey.
Getting the viewer to perform the specific, “low-level” analytic activities while actually taking in the visual amplifies the usage of such cognitive processing techniques, while also providing a common context for such information dissemination. Jacques Bertin said the deduction of relationships is a matter of permutations. John Tukey developed techniques like exploratory data analysis to extract hypotheses from data, tackling visuals from a statistics side with box plots, rootograms, and Pareto charts. Wehrend and Lewis methodology created a mapping of techniques to the problems of cognitive tasks. Roth and Mattis methodology of visual presentations has the user seeking out comprehension themselves. Zhou and Feiner methodology involves informing to enable understanding of relationships. Shneiderman supports a taxonomy based on user goals and the tasks they have to perform to achieve them. Card focused on high level tasks breaking down into visual structure for its theory. Chi categorizes visualization techniques by their models. The paper provides ten analysis task types categorized by the data and viewer’s goals. Called affinity diagramming analysis, they are
- Retrieve Value
- Filter Compute Derived Value
- Find Extremum
- Determine Range
- Characterize Distribution
- Find Anomalies
- Some of these are more complicated than others.
- People might not know what they want.
- Usability tests for performance tests.
- Some recording of everyone else’s feedback.
Is it a visualization task or a computation task? Choose the right aggregators to get what you want. Is aggregation visual or computation task? It is part of the viz tool, but? Tableau can drag and drop, but the skin is constant. The customization is computational, this part can be annotated. http://vis.stanford.edu/wrangler/
A Multi-Level Typology of Abstract Visualization Tasks proposes a multi-level typology of all the various tasks involved in visualizing that bridges the gap between low-level and high-level types. It is dependent on the reason for the presentation (why?), its methodology (how?), and input-output parameters (what?) to analyze behavioral tasks. A Design Space of Visualization Tasks suggests interfaces for better usage of design interfaces with respect to the kind of users of different expertise levels. Depending on the goals, target to be achieved, type of data, and the tasks grouped together affecting the cognitive load, the interface usage may range from new users to sophisticated users. It includes consistent design practices, picking the design space that fits the domain, and the right visualization framework.
Discusses a powerful technique for an optimal graph layout called the constrained graph layout. It can be applied to a range of practical examples which have three stages – The initial layout is designed with Pivot Multidimensional Scaling method, run for many iterations without constraints and then with mathematical constraints non-overlapping models until a convergence criteria is reached.
For multiple networks to be visualized at once, ManyNets offers a solution in Social Network Analysis, using Visual Information Seeking Mantra (overview, zoom and filter, after which details can be accessed on demand, as we read from Schneiderman’s The Eyes Have It: A Task by Data Type Taxonomy for Information Visualizations) leading to faster, efficient cognitive tasking for understanding the nodes and edges in a network.
Data can be visualized as a graph if an inherent relationship can be formed between the elements. Graph visualizations have several issues like low comprehensibility for larger graphs, compromising performance and viewability. Several Gestalt principles bring up effects in graph visualizations which are actually unintended, due to the cognitive expectations of a viewer. Although not mentioned in the paper, I believe visualizing time in a graph design is also difficult. The time range of one node, related to another is not obvious with its relationships or even the weight of its edges; however, this will not reflect the link density or the edge weight.
Munzner: Chapter 3, 6.10, 9
Apart from a domain-specific framework, an abstract framework details the reason, actionable items, and targets of a vis tool which is generalised to its keywords. Munzner, evidently focuses on the function of the visualization over its form. There are many variations of the layouts by which network graphs can be represented. Force-directed graphs are non-deterministic, as in, they appear different each time they are rendered. Randomness in such dynamic appearance, brittle settings, and the visual as well as time scalability also does not work in favour of force-directed graphs. Adjacency matrix layout is another representation of network graphs.
The Rise of Interactive Graphics by Cairo: Chapter 9
Multimedia informational graphics enabled by Adobe Director and Macromedia Flash were the vanguard of the graphics movement in the early years. Interactive design is not just the digital but also the physical user research. Just how much of this might be incorporated into sci-fi films, I wonder. Mimicking real-world actions to achieve a similar purpose in the digital-sphere, makes the option to do so visibly. Constraints on such actions eases the user into the required actions. Consistent visualizations reduce cognitive load, and enhance navigation. Interactive viz can be instructional, dialogue-oriented, persuasive, or exploratory. Cairo advocates for storyboarding the interaction aspect of the visualization to fit the viz mantra in.
Animated Transitions in Statistical Data Graphics. Heer & Robertson. IEEE InfoVis 2007 argues that while animations are over-hyped, transitions between two graphs can be animated to the visualizer’s advantage. The types of transitions range from point of view, rescaling, highlighting specific data, reordering, reorienting time, type of charts, or the data style itself.
Effectiveness of Animation in Trend Visualization. Robertson, Fernandez, Fisher, Lee, Stasko. InfoVis 2008 postulates that the results of trends analysis with animation, as in the case of GapMinder World is an intriguing study for deriving analysis itself, which is possible with static visualizations as well. Trends may be upwards, downwards, reversed or cyclic, which can be statistically inquired as a trend line in regression. For analysis of trends for unfamiliar data itself, the design, multiple individual views, aggregating the data points, are important factors to reduce clutter and confusion.
Animated Exploration of Graphs with Radial Layout, Ping Yee, Danyel Fisher, Rachna Dhamija, and Marti Hearst. InfoVis 2001 justifies that a node’s degree of connectivity and network distance, nearness to other nodes, and other such social network theory and communication concepts are better represented in a radial tree layout or target sociograms than in 3D Cone Trees or Hyperbolic Browsers, which are fascinating to view in their own right. I would imagine sci-fi films making ditzy use of such visually extravagant layouts in AR / VR holographic models. For the purposes of viewer understanding in a more complete sense, the point of view and interacting with the angle of view would influence the inclination of the visual. Linear interpolation, or the shortest straight line between two or more constrained nodes. The radial tree nature is represented by one half of a neighbour node being interactively selected to highlight all the new distances from that focus node. The dynamic animation of inner focus to outer focus of a network gives it the radial tree structure with nodes of different sizes via breadth-first search of the parent-child relationship. I found the marital and business ties of the Florentine family tree a fascinating use case, however, I could see how the visualization adds value to a fact of knowledge that could be derived from a text block on the family networks.
According to Munzner, the response time in an interactive visualization is of immense consequence in the era of high speed internet and video media. The feedback in the latency cycle has to be visual so that the viewer is engaged throughout the sequence. Retrieving data from cache memory is preferable compared to waiting a longer amount of time. Using lower resolution of pixels with visualization mantras like Get it right in B/W and Function first, form next enhances the effectiveness of the visualization.
Visual Queries for Finding Patterns in Time Series Data. Hochheiser & Shneiderman introduces Timeboxes in the Time-Searcher application for dynamic, interactive visualization of time-series data specifically in a modern GUI. It shows two dimensions of a query over the multiple timeboxes. The dynamic data selection in
Generalized Selection via Interactive Query Relaxation. Heer, Agrawala, Willett. CHI 2008 discusses the nuances of selecting an item for fulfilling a specific query. I would have considered point and click with cursor to be the most constant feature of such interactions. However, this paper shows me that dynamic data can be better interpreted with detailed, definite, and referenced attributes in the visualization. The location and of such point-and-place attributes are specific to the visualization in a “deictic” way but can be linked to other attributes too. This is potentially very useful in network graphs. I would be very interested to see this paint-brush technique included in my semester end project.
The design space analysis of narrative visualization chart is an interesting exercise in understanding the popular trends in visual elements. In my project, I suppose I will be using an interactive slideshow format for timeline and continuous graphs. Maps and networks each get their due in a drill down approach, as I understand them.
The flow of storytelling with data is important in getting the point across or even for insight to be formed. For my own project, the parallelism of general-to-specific transition would indicate the possible explicit impact of narrowing down focus of each map, possibly.
Narrative storytelling with visuals is something I came across with Cole Nussbaumer Knaflic’s work on storytelling with data. Here the focus is on the inference being available to the user via persuasive interpretative techniques. I found the concept of Jacques Bertin’s semiotics fascinating; which is itself a mesh of literary, political, critical studies used to understand signs.
I had not previously considered how rhetorical discussion for analysis might play into understanding the subject matter, and in turn, the inference as well. The prioritization of such influences is an important “storytelling device” much like plot points in a novel or film I suppose.
References and Further Readings
Data Wrapper Book Clubs:
Alberto Cairo Touriño—Nerd Journalism_ How Data and Digital Technology Transformed News Graphics-Universitat Oberta de Catalunya (2017).pdf. (n.d.).
Alberto Cairo—The Truthful Art_ Data, Charts, and Maps for Communication-New Riders (2016).mobi. (n.d.).
Amar et al. – Low-Level Components of Analytic Activity in Infor.pdf. (n.d.). Retrieved December 24, 2020, from https://www.cc.gatech.edu/~john.stasko/papers/infovis05.pdf
Amar, R., Eagan, J., & Stasko, J. (n.d.). Low-Level Components of Analytic Activity in Information Visualization. 7.
(Best American) Gareth Cook (foreword), Robert Krulwich (intro), Nicholas Felton, Randall Munroe, Nathan Yau, Giorgia Lupi, Stefanie Posavec—The Best American Infographics 2016-Mariner Books (2016).epub. (n.d.).
Brehmer, M., & Munzner, T. (2013). A Multi-Level Typology of Abstract Visualization Tasks. IEEE Transactions on Visualization and Computer Graphics, 19(12), 2376–2385. https://doi.org/10.1109/TVCG.2013.124
Cairo, A. (2013). The functional art: An introduction to information graphics and visualization. New Riders.
Design of data figures | Nature Methods. (n.d.). Retrieved December 24, 2020, from https://www.nature.com/articles/nmeth0910-665
Heer, J., Bostock, M., & Ogievetsky, V. (2010). A tour through the visualization zoo. Communications of the ACM, 53(6), 59. https://doi.org/10.1145/1743546.1743567
Knaflic, C. N. (2015). Storytelling with data: a data visualization guide for business professionals. Hoboken, New Jersey: John Wiley & Sons, Inc.
Lupi, G., & Posavec, S. (2016). Dear data (First edition). Princeton Architectural Press.
Munzner, T. (2014). Visualization Analysis and Design (0 ed.). A K Peters/CRC Press. https://doi.org/10.1201/b17511
Schulz, H.-J., Nocke, T., Heitzler, M., & Schumann, H. (2013). A design space of visualization tasks. IEEE Transactions on Visualization and Computer Graphics, 19(12), 2366–2375. https://doi.org/10.1109/TVCG.2013.120
Shneiderman, B. (1996). The eyes have it: A task by data type taxonomy for information visualizations. Proceedings 1996 IEEE Symposium on Visual Languages, 336–343. https://doi.org/10.1109/VL.1996.545307
Telea, A. (2015). Data visualization: Principles and practice (Second edition). CRC Press, Taylor & Francis Group.
Tufte, E. R. (1999). The visual display of quantitative information (17 print). Graphics Press.
van Wijk, J. J. (2005). The Value of Visualization. 8.
Visualization Lab | Design Principles for Visual Communication. (n.d.). Retrieved December 24, 2020, from http://vis.berkeley.edu/papers/designprinciples/Zhang, J., & Norman, D. A. (n.d.). The Representation of Numbers. 24.