EFFECTIVE INFORMATION VISUALIZATION
Guidelines and Metrics for
3D Interactive Representations of Business Data
3 OBSERVATIONS FROM USE OF CONSTRUCTED VISUALIZATIONS*
3.1 Inventory Viewer (Fixed Income Risk Management) (1993+)*
3.2 TSE 300, SP500, and other bid-ask-trade stock visualizations (1993+)*
3.2.2 Wall Variations*
3.2.3 LME and SP timeseries variations*
3.3 SP500 Minimalist Representation (1995)*
3.4 Head Trader (1994+)*
3.5 Sales (1994+)*
3.6 Trade Routing (1994)*
3.7 Annual Report Visualization (1995)*
3.8 Stock Risk Factor Visualizations (1995+)*
3.9 Multiple Timeseries Visualization (1995+)*
3.10 Risk Movies (1995)*
3.11 Management Decision Support*
3.14 Other Constructed Visualizations*
3.15 Summary from Constructed Visualizations*
Visible Decisions, a company founded in 1992, has focused on the creation of information visualizations for corporations and government organizations around the world. As a pioneer in the commercialization of information visualization, this group has had to confront many real-world issues in creating fully functional and usable visualizations for users ranging from computer novices to computer experts. A team at Visible Decisions Inc., led by William Wright and Richard Brath has created over two hundred information visualizations (1998). Each visualization was designed with user input, developed through a number of iterations and built upon past experience. Many different visual and interactive techniques were explored over this time on a case-by-case basis. An overview from some of these visualizations and observations on some of the techniques are given below. Each section represents a different type of visualization.
Figure 3-1 illustrates the first commercialized visualization built by Visible Decisions Inc. Previous VDI visualizations were limited prototypes with very limited interaction. Inventory Viewer was built with requirements collected from business users. It uses real data and it brings together a wide range of interactive techniques. Interactive echniques such as filtering and brushing proved to be very important to the success of the visualization and have been incorporated into almost all visualizations created at Visible Decisions.
Figure 3-1: Inventory Viewer overview
Images from Visible Decisions. <img src="inventoryViewerWhite.gif"> <img src="inventoryViewer95a.gif">
Inventory viewer visualization overview. Initial version shown to left (1993),
revised visualization shown to the right (1995). These images white backgrounds for improved printing. Under normal use they have black backgrounds.
This visualization is used to assess risk in an interactive manner across an entire bond portfolio. A bond portfolio consists of thousands of financial bonds, each with various attributes. The basic layout of the visualization is a grid with date (maturity date of the bond) along one axis, and bond categories (trading desk) along the other axis. The bars at each intersection represent a measure (such as face value, sensitivity or profit) as an aggregate for all bonds in that particular bucket (e.g. all the Canada bonds in the year 2002). Different measures were shown with a different color pair (green and red for positive or negative sum of bond face values; blue and yellow for profit vs. losses, etc.). Only one measure can be shown at a time. A total line along the front aggregates across the different bond categories. Along the back, a yield curve shows interest rate at different dates in the future. The yield curve’s date axis is aligned with the date axis of the grid on the floor. By default, the yield curve shows current interest rates.
Figure 3-2: Inventory Viewer with brush and filter.
Images from Visible Decisions. <img src="inventoryViewer95b.gif"> <img src="inventoryViewer95c.gif">
Left image shows the view zoomed in to a particular area. By pointing the cursor at a bar at a grid intersection, a textual brush appears, showing the values for bonds represented by that bar. Right image shows view rotated to the side with an active filter. The filter interface is shown at lower left with slider bars being used to select a subset of bonds based on value. Only the filtered subset of bonds are visible in red on the grid.
Some novel interactions included:
This was the first commercial visualization produced by Visible Decisions. VDI developers and the users found numerous shortcomings in the visual presentation and interaction:
However, this visualization also revealed numerous successful visualization techniques:
This visualization was revised in 1995. Besides numerous enhancements to data source handling, the new version addressed some of the previous shortcomings:
A series of real-time stock market visualization explored variations of "small multiples" within a 3D information visualization. Occlusion and reference contexts became a concern.
Figure 3-3: Stock visualizations.
Images from Visible Decisions. <img src="tse300.gif"> <img src = "dow30.gif">
This visualization evolved as a general, real-time, stock-market visualization. All variations of this visualization showed real-time stock attributes (e.g. bid price, ask price, last trade price, number of shares for sale at the ask price, etc) for all the stocks that underlie a stock index, such as the Toronto Stock Exchange 300 or the Dow Jones Industrials. The original visualization (TSE 300) shares features common to all:
Figure 3-4: Stock visualizations - bid/ask cubes.
Image by author. <img src="sp500good2.gif">
This visualization evolved into a series of visualizations. Many later variants included walls. Typically walls were located either at the end, or along the sides. Walls were used to locate related market information such as the performance of key market indicators (such as sector indexes) or news headlines.
The original and initial variants only represented the current state of the stock market – they did not represent a period of time. Some variations attempted to represent time using different techniques:
Figure 3-5: Time series within the bid/ask visualizations.
Images by author. <img src="timeseriesGrounded.gif"> <img src="timeseriesUngrounded.gif">
In both images the user has turned off the bid/ask cubes. Left image shows timeseries oriented horizontally and constrained to fit within a roughly cubic volume. Right image shows the time series oriented vertically.
This visualization introduced one remarkable innovation: the connotative small multiple. The use of stacked cubes is sufficiently abstract so that it is both visually simple and easy to comprehend. The TSE and SP500 variants of this visualization were shown to numerous individuals at tradeshows, in presentations and universities. Given a verbal description of the bid-ask cubes, viewers could often identify different conditions within the stock market through viewing the visualization, even though they did not necessarily have any background in the stock market. The SP500 variant visualization resulted in a number of successful commercial sales after presentations to senior managers at banks in Canada and the United States as well as retailers. The concept was sufficiently simple and concise enough for non-financial managers to make an association to their own area of expertise and see the application.
Other innovations with this series of visualizations included:
These visualizations enhanced our understanding of occlusion and reference contexts: Objects must have a clear relationship with a reference context and should not have too much occlusion.
The SP500 visualization was rebuilt in 1995 with entirely new small multiples. The intent was to reduce the number of shapes required to represent the information, thus simplifying the scene and permitting interesting conditions to be found more quickly and/or more data to be shown. This approach was chosen based on a very literal interpretation of Tufte’s maxim to "reduce the non-data ink" [Tuf83].
Figure 3-6: Minimalist S+P 500 stock visualization.
Image from Visible Decisions. <img src="sp500.gif">
This visualization presents all the 500 component stocks that make up the Standard and Poors 500 stock index. A single cube represents each stock. The cube shows 5 different data attributes using 5 different visual attributes (depth, width, height, hue and brightness). Data attributes are arbitrarily mapped to different attributes of the cube. It was assumed that a user would be able to combine all the attribute mappings together into simple visual patterns – such as "bright, big, red cubes indicate a buying opportunity" or, "flat, dark purple squares indicate a lot of activity going nowhere".
Figure 3-7: Stock visualization - minimalist representation.
Image by author. <img src="sp500bad2.gif">
This visualization was shown to numerous individuals at tradeshows, in presentations and at universities. The result did not provide the novice user with any easy method to remember the attribute mappings. The user would typically forget one or two of the mappings while viewing the cubes. This required the user to frequently refer back to the legend thus losing focus on the cubes containing the actual content. In comparison, the previous TSE visualization, section 3.2, used a small multiple composed of many parts (multiple color coded cubes and a line joining them together) with a somewhat connotative mapping; while this visualization used a single cube with arbitrary mappings.
This reduction to a single cube also removed redundant encoding. In previous versions, hue and location relative to each grid cell distinguished between bid, ask and last trade. Also in the previous versions, both width and depth of each box encoded size.
Five different data attributes were always visualized simultaneously on the same cube, whereas, the earlier visualizations grouped only two or three similar attributes together into a single visual object with a specific hue (such as the bid cube representing bid price and bid size).
We suspect that users were able to automatically mask out objects based on component parts for visual search tasks in the earlier (TSE) visualizations in section 3.2; while the minimalist representation presented no component parts forcing the user to work with the cube and all 5 attributes simultaneously. Although such a mapping may result in improved performance for an expert user, the visualization is not easy for the novice user to learn. We are uncertain as to which visual criteria were used for masking in the TSE visualizations (Is it hue? Is it the relative location of the component parts? Is it some combination of shape and hue?).
Since users frequently forgot attribute mappings, the inclusion of a legend was absolutely imperative to making this visualization usable.
The head trader visualization re-uses an organizational device that a user already knows. Head trader is a real-time monitoring visualization that utilizes the floor plan of the brokerage’s trading floor as its primary organizational device.
Figure 3-8: Head Trader.
Image from Visible Decisions. <img src="headTrader.gif">
The head trader visualization evolved from the bid-ask series visualizations. The visualization combined real-time data from markets with a financial firm’s actual holdings. Driven by user requirements, the primary goal was focused on the trader. The users wanted to rapidly assess the current status for the firm (first); for each individual trader (second); and the state of the underlying stocks which the trader traded (third).
The overall organizing device used was the layout of the actual trading floor: the manager knew where each individual trader’s desk was located and the location of each trader on the trading floor was carefully chosen to place each trader in close relationship to other traders. Thus re-using layout of the trading floor was driven by the user and proved to be more effective than reading through an alphabetically sorted list. The user could see patterns between groups of traders emerge much more clearly.
The trading floor metaphor also provided walls that were utilized to show various summary attributes for the overall firm, such as profit/loss and exposure for the entire firm. The use of walls to display summary data closely matched the user’s reality. Many trading floors include pixel boards with various market and firm summary statistics. The users were able to first scan the summary walls, then scan the trader summaries on the floor, and then zoom into any particular trader to see details for individual stocks. Any large exposure or profit/loss, would immediately be seen as a spike whether at the summary, trader or individual stock level.
The decision requirements were clear: the user needed to know status very quickly for a small subset of variables (profit/loss, exposure) at various levels of aggregation (stock, trader, firm). The use of the trading floor layout permitted the manager to very quickly assimilate all the information in a single glance. The particular innovation was the re-use of the floor plan. By reusing an organizational device that the user was already familiar with, the learning effort was reduced, and search tasks were simplified. For example, if the user was interested in the status of a particular trader or type of instrument traded, the user could instantly focus on the portion of the trading floor of interest as opposed to reading through a list of traders’ names or instruments.
A number of trading floor visualizations for different clients were built. User comments before the visualization were: "We’re drowning in data and we need some way to make sense from it". After the visualization the user comments were "We now have a single way to make sense of this information from a single interface".
Stuart Card [Car97], an authority on visualization, presented a taxonomy of visualizations at IEEE Visualization ’97, primarily organized on utilization of representational dimensions (e.g. the spatial dimensions, X, Y ,Z, color dimensions, time dimension, etc). He presented an image of a VDI trading floor visualization as an example of the very innovative multiple use of the spatial dimensions within a single visualization (i.e. the small multiple principle of Tufte[Tuf83]). At a large scale, the spatial dimensions are used to locate traders, at a finer scale they are used to locate stocks, and at a finer scale they are used to represent properties such as profit and loss.
Following a VDI visualization, the NYSE has created a trading floor visualization representing various abstract data including market data and system conditions. All the stock and trading post information is placed directly in the context of an abstraction of the trading floor. In addition, the walls display other contextual information such as indexes and stock tickers. More information about the NYSE’s visualization can be found athttp://www.nyse.com/floor/ramp.html.
A sales analysis visualization evolved popular interaction techniques including animation over a time variable and a water level (i.e. clipping plane). It also explored issues of use of geographic maps within information visualizations.
Figure 3-9: Sales - as it first appeared in 1994.
Images from Visible Decisions. <img src="salesDiscovery.gif">
Figure 3-10: Sales as it appeared in 1996.
Images from Visible Decisions. <img src="salesDiscovery96a.gif">
Figure 3-11: Sales – 1998 version.
Images from Visible Decisions. <img src="salesJava.gif">.
After successfully selling visualization to retailers based on financial visualization demonstrations, a more general sales based visualization was created. The visualization focused on sales and margin values for a particular product for each store in each region. The visualization featured a number of innovations over previous visualizations:
The use of the map and simple bars located on the map made the visualization very easy to understand. As a result, it has become one of the most widely used and most widely referenced VDI visualizations. People from any background understand the visualization. The visualization is regularly shown in most VDI presentations.
The map itself, although only used as a reference to help identify locations, has drawn complaints for being too abstracted: "My state doesn’t look like that"; and "Is that really Minnesota?" Although no one familiar with U.S. geography has had difficulty identifying states presented on the map, offense has been taken. Newer versions have successively refined the map. Why map accuracy is of concern to users, VDI is uncertain.
The use of animation as an analysis device generates enthusiasm. Users have said that the animation presents them with the possibility of being able to comprehend trend on a finer scale than ever before. Many analysts, particularly in the retail sector, do not view trends within their data, other than doing a single point of comparison to the same time period last year. In addition, animation maps intuitively to the measure of time – thus the use of animation for a time variable is very easy for most people to comprehend.
The enthusiasm of users to the water-level filter surprised VDI. Previously, a separate filter that turned on/off bars based on user selected values performed the same function and had been in each visualization since Inventory Viewer. However, the value of the water-level may be its explicit representation: it is obvious that moving the level up/down and obscures objects below it and therefore it is very easy to learn. Or, the value may lie in the location of its operation: as it moves up/down in the context of the visual objects, it facilitates continued focus on the data thereby eliminating the need to look back and forth between the 3D scene and a GUI slider bar. Finally, since the water-level is explicit within the scene, its meaning is easily communicated through static scenes such as used in printouts and presentations.
Each of the images above is presented with the original background colors and have not been modified for print. Notice how the original visualization has a black background that creates high contrast with the foreground elements while later variations have evolved with successively lighter backgrounds.
An experimental trade routing visualization done in 1994 presented 25 different trading organizations and drew lines between different pairs to represent trades from one organization to another. The intent was to create a simple visualization showing trading relationships. The result was a "spaghetti" of lines. To help clarify this, unique colors were chosen for the 25 different organizations. The result was multi-colored spaghetti; that is, it was difficult to see and understand trading relationships – there were too many overlapping lines and too many different colors.
Figure 3-12: Simplification of trade routing visualization.
Images from Visible Decisions. <img src="tradeRouter.gif">
Experiments with this visualization and others show that we have difficulty perceptually separating many different colors – although the threshold for "too many colors" is not known.
In this particular visualization, visual patterns became apparent when the visualization was scaled back to six different organizations with only six different colors.
At the request of a customer, a visualization of their annual report was made. This was made against our recommendation because there was no particular goal or user problem identified. In addition, there was very little data: each measure was not directly comparable to the next as a ratio. No trend or historical values were available either: historical values would be directly comparable within a given measure.
The resulting visualization was of little use. Visual comparison across singular different measures does not yield any value.
This visualization introduced interactive drill-down within the 3-dimensional scene. It evolved, in part, to overcome problems with moiré and scale.
Figure 3-13: Risk factors.
Image from Visible Decisions. <img src="barraGlobalRisk.gif">
A client wished to interactively view and explore risk factors on a stock portfolio. This was conceptually similar to the earlier Inventory Viewer application. However, there were many more risk factors, and many more groupings of stocks. The client also wanted to be able to see individual stocks within the visualization, as opposed to bucketed grouping as was used in Inventory Viewer. A very early version of the application resulted in a 58 x 500 grid. This 58 x 500 grid was too large:
The fixed grid was redesigned. Different variations were experimented with. An expand/contract tree control was used in one variation; drill-down buttons (represented as country flags) for one-level of drill-down were provided in a second variation (the second variation was used for management reporting, the first variation was used by analysts).
The interactive 3D drill-down technique of expanding the visual representation within the scene became an important interactive technique used in many VDI visualizations after this visualization. Graphical drill-down techniques are necessary for viewing visually a level of detail somewhere between the high-level visual summary and low-level details presented in the brush. All VDI visualizations before this time presented only high-level visualization overviews with low-level details delivered in a brush and interactions such as filtering were applied globally across the entire visualization. This was the first VDI visualization to use an interactive drill-down technique that provided the user with some control over the scope of drill-down. Approximately 1 out of 3 VDI visualizations use some form of focused drill-down technique. A variant of this drill-down, as a pop-up graph, can be seen in the sales visualization dating from 1996 (section 3.5).
Timeseries data typically has hundreds to thousands of observations for each object. Comparison across a number of timeseries requires normalization across the timeseries data. This normalization may require user interaction.
Figure 3-14: Timeseries - initial visualization.
Image from Visible Decisions. <img src="timeseries.gif">
Presentations of the TSE 300 series of visualizations revealed to VDI groups of analysts that were primarily interested in the exploration of multiple timeseries simultaneously. The initial attempt (shown above) presented each timeseries side by side. Interactions included brushing (to see the values for any individual observation), highlighting (any particular timeseries could be selected and highlighted), and filtering (filter out all the bars in a timeseries over a particular value). This initial timeseries visualization had numerous problems:
In effect, the interactions available in the visualization were not the interactions required for the analysis of multiple timeseries data.
Later timeseries visualizations (1996 to 1998) have been very successful, because the representations are based on normalized data and appropriate interactions are available:
Figure 3-15: Timeseries - later improvements.
Images by author. <img src="SeeITfxDailyWaterlevel.gif"> <img src="timeseriesStudio.gif">
In the first image (representing daily foreign exchange rates during 1997), each timeseries has been normalized to base "0" at the starting time. The height of the subsequent series is based on the change from this initial starting point. The color is set to the percent change from the previous observation. Even though the nominal exchange rates for the DEM and FRF are significantly different, it can be seen that they track almost identically. The vertical stripes of color and underlying reference grid aid in the visual alignment of different time series; for example, it can be seen that CHF peaks near the same time that DEM and FRF. Also, the waterlevel permits all timeseries to be quickly compared at any particular threshold.
In the second image (representing changes in daily stock prices over 3 years), the user controls normalization. The user can click and drag any point in time to be the "100" base point (shown in the cyan line cutting across all the timeseries). A second point of comparison can be chosen, with a corresponding red/green marker showing the net percent change at that point in time from the initial chosen base point in time (shown at the orange line cutting across all the timeseries). These interactions permit the user to choose and accurately compare between any two arbitrary points in time. The back wall permits selected timeseries to be displayed on the same visual plane. Using this wall, regions of selected timeseries can be accurately compared on the same plane.
For a visualization to be successful, the visualization and interactive techniques must be appropriate to the task. The techniques evolved above are specific to making multiple timeseries comparison usable. These are:
These techniques are used in very detailed VDI timeseries analysis visualizations at banks, stock exchanges and government organizations. After working with a group of senior statisticians at a government organization with significant staffing resources dedicated to timeseries analysis; a senior statistician summarized these techniques as "a very powerful innovation to timeseries analysis". The statisticians were particularly interested with the possibilities of the interactive comparison at arbitrary points in time.
Risk movies focused on ease of use as the visualization was targeted at senior executives with little computer background and limited time available for analysis. To achieve ease of use goals, many techniques were used including:
Figure 3-16: Risk Movies
Image from Visible Decisions. <img src= "riskMovies.gif">
This visualization was the first visualization designed for use by senior management. Many subsequent management report visualizations have been created. The requirements included that the interface be very simple to use and the visual presentation be very simple to comprehend. The visualization presents a bank’s inventory cross-tabbed as a bar chart organized by different types of financial instruments (bonds, t-bills, etc) and by country. The green and red bars show profits and losses respectively. The left wall shows various properties for each country, such as interest rates, stock indexes and exchange rates. The interest rates are represented by a curved line, with which the users were already familiar with and refer to as a yield curve. The right wall shows summaries for various different scenarios: each vertical bar represents one scenario and the entire group gives a summary overview of all the scenarios. A number of the design ideas originated with the users at the firm of Algorithmics.
Figure 3-17: Risk Movies preset viewpoints.
Images from Visible Decisions. <img src="riskMoviesGraphView.gif"> <img src="riskMoviesScenarioView.gif">
Buttons in the GUI interface easily permit the user to move to preset viewpoints, such as the graphs on the left wall and the scenario chart on the right wall.
The user steps through different scenarios by clicking on VCR-like buttons in the toolbar (play, stop, back, forward, etc). The viewpoint is easily adjusted using only 3 buttons in the toolbar. Additional turning of the scene, although not required, can be accomplished by clicking and dragging with the right mouse button. Turning the scene with the mouse provides a consistent rotation about the vertical axis with a left-right mouse movement, and tilt of the scene up-down with up-down mouse movement (zooming and panning are also available through use of the mouse). Any object in the scene can be brushed by pointing the mouse at the object. No further interaction is required to invoke the brush. Brushing typically shows the category of the item being brushed and the value. No further detail is presented.
This visualization replaced a batch processed report of 120 pages with a single screen. The visual report was reviewed more quickly than the paper report.
This visualization is used as a demonstration with to many people both within VDI and at other companies. As of this writing, it is available on the VDI website for general download (www.vdi.com).
Navigation: This visualization was among the first visualizations at VDI to be usable with only GUI-button based navigation. Apprehension to mouse-based 3D-scene navigation had been noticed in many novice users with previous visualizations (in spite of the fast feedback with mouse-based navigation):
Complete navigation via the GUI interface alleviated this fear: all 3D navigation was available from a visible interface.
Intuitive Mappings and Layout: The green and red bars map intuitively to good and bad respectively, and thus map well to profit and loss. People understand the central concept of an inventory with green profits and red losses very quickly. In addition, the green profits project up from the ground plane while the red negative bars protrude down from the ground plane thus using direction connotatively and reinforcing the color mapping. The alignment of each country on the ground and the left wall reinforces the relationship between each country’s macro-economic condition with its impact on the inventory. The yield curve and the nomenclature for the indexes are industry specific and are understood by a financial audience but are not well comprehended by a general audience. Finally, the right wall showing the summary is not intuitively related to the other data by itself. Through animation it becomes more apparent that there is a correlation between the size of vertical slice and the height of the bars on the grid. This right wall diagram is an artifact of the reports that the organization had been using prior to the visualization and although the immediate user community may have understood it, it requires significant explanation/documentation for the new user.
Brushing and Learning: The contents of the brush are simple in this visualization: The first two lines identify the item being brushed (e.g. Canada – Treasury Bill) and the value of the item (Profit: $12,000,000). Observations of new users revealed interesting behavior with the brush:
VDI became aware that the brush was used being as a learning device. By repeated brushing different bars, the users were making assumptions and testing those assumptions against the representations in the scene. Rather than being used as a tool to drill-down to the underlying data, the brush was a critical tool used to refine and validate an evolving conceptual model.
Conclusion: With a clean GUI interface, clear brushing, conceptually simple interactions and intuitive central mapping, this visualization was often the first visualization a new user would learn.
This visualization tackled a complex "what-if" analysis across many measures by making reuse of many organizational devices and linking their interactions together.
Figure 3-18: Decision Support.
Image from Visible Decisions. <img src="oilCo.gif">
This visualization presented complex project management information in a single view, using many organizing devices already familiar to a company.
The organization chart on the back right wall shows different business units. It is used as a selection interface. Any subset of projects can be selected by choosing appropriate sub-groups of the organizational chart.
The charts on the left wall show a summary of the consumption of 10 different resources, such as capital expenditures, over time. The summary is based on the currently selected projects as chosen from the organizational chart. Any individual resource can be selected for further analysis.
The map on the floor shows the amount of the selected resource being consumed at each location at a particular point in time. A VCR style control at the top permits different time periods to be selected. The bars can also be selected to choose subsets of projects at particular locations.
The timeseries along the left front edge show the consumption of the selected resource for each major project. The timeseries are sorted on the resource consumption for the selected time period. Any combination of major projects can be selected and the timeseries for the component projects can be seen along the right front edge. Any component project can be selected and its rate of development can be changed. All summaries are immediately impacted.
The visualization is used by senior executives to:
Prior to use of the visualization, the management team depended upon pre-computed spreadsheets of different scenarios predetermined before the meeting. The review of the data was slow and cumbersome and new scenario suggestions could not be evaluated.
The visualization was useful because it presented the business in the form with which the users were familiar. It used their org charts, their map and their measures and their methodologies from scenario decomposition to working with timeseries data. The different representations were linked through interactions thus reinforcing what was viewed in one representation with the corresponding representation in another representation.
SeeIT was first created in early 1996 as an example visualization which web users could download and explore small datasets through an interactive scatterplot visualization. It include a broad palette of interactions for quick ad hoc analysis including brushing, filtering, tightly coupled displays, user defined mappings for visual attributes and axes and a water-level.
Figure 3-19: SeeIT overview.
Image from Visible Decisions. <img src= "seeIt2.gif">
This visualization is used for ad hoc exploration of arbitrary record based data sets. Record data appears within the data grid shown in the left panel. The visualization is based on a simple mapping of record fields (as identified by the column names in the data grid) to the X,Y,Z axes and shape properties, such as height, width, and color – resulting in a grid with assorted colored shapes placed on it.
Back walls act as shadow walls, similar to the use of shadow walls in the CAD domain. They show the same information in the grid projected onto the wall. Thus the back walls effectively show the distribution of the data in one of the two dimensions used on the floor.
The current mapping of data to the axes is explicitly shown along the edges of the grid in the 3D scene as well as in a legend in the upper right corner of the scene. The mapping of data to the shapes is shown in a legend in the upper left corner of the scene.
Direct scene interactions include:
The visualization was successfully used on the website. Over 2,000 people have downloaded the visualization. SeeIT has been used to present findings at four conferences (that Visible Decisions is aware of), including a presentation that stated that the combination of SeeIT and data mining techniques permitted an automotive manufacturer to save more than $5 million. SeeIT has since been commercialized.
SeeIT presented some enhancements over previous visualizations, including:
Conclusion: SeeIT introduced general user-defined mappings for data to visual attribute connections. This innovation is particularly useful for ad hoc analysis by a sophisticated users, but confusing for a novice user (there are many possible settings, plus offsets and scale factors). The placement of legends in the overlay plane is very useful and has been used in many VDI visualizations since. The removal of the Z axis connection and the default settings helped initial users avoid the danger of creating a visualization with no reference context.
This visualization was created as a concept visualization in late 1996. The visualization represented graphs; where a graph is a large collection of objects and relationships. In this case, the graphs were websites, with individual webpages as objects and hyperlinks as the relationships between objects.
Figure 3-20: Webviz overview.
Image from Visible Decisions. <img src= "webViz1.gif" >
Webviz displays all the connections in a web site in a 3D volume of connections. The graph is displayed as a tree from the root of the web site (shown as a cyan square towards the lower left in the image) with green lines indicating connections to the next level of the tree, with each node shown in cyan. Links that go back to a higher level are shown in blue beneath the tree as a shadow. The user can toggle various node properties such as adjusting the color or the form or the addition of annotation to the nodes to indicate various attributes about the web pages. For example, the brightness of the node indicates the age of the content; cyan squares indicate web pages on the site, while purple triangles indicate links to other web sites; and yellow text indicates the title of web pages within the first or second level of the hierarchy. Zooming in close to a node displays a thumbnail image of the web page on that node:
Figure 3-21: Webviz closeup.
Image from Visible Decisions. <img src = "webViz2.gif">
Initial versions did not include brushing. The visualization was adapted and drill-down windows added to display detailed contents about the page and links to launch web browsers to view the full page. The layout of the hierarchy was challenging. By default, the layout algorithm could result in a volume that might be excessively long or wide, making it difficult to get an overview. Dialog boxes were created which permitted the layout to be customized. Large web sites resulted in trees which had regions of high density. This was addressed through the creation of a drill-down mechanism that permitted portions of the tree to be isolated. Navigation within the web visualization was challenging. Zooming into particular pages and then navigating up/down hierarchy required navigation dependent on the current object of interest. The addition of the thumbnail image aided the user’s ability to identify individual web pages. In addition, thumbnail images help users locate themselves within the web site and help them determine where to navigate to next.
Conclusions: The visualization of large graphs is challenging. Navigation, layout and subset selection of the graph are instrumental interactions to making the graph usable.
Many of the visualizations constructed by Visible Decisions are proprietary. Other experimental visualizations have not had a broad exposure outside of Visible Decisions and thus are inconclusive. Samples of other visualizations with unqualified comments are shown below.
Figure 3-22: Exploded Time Image from Visible Decisions.
Different layers provide different data contexts for the timeseries. Interfaces to turn on/off layers aid comparison of data subsets.
Figure 3-23: Black Scholes
Image from Visible Decisions.
Figure 3-23: Black Scholes
Image from Visible Decisions.The height of a surface can be difficult to judge perceptually. Interactive cross-section lines provide a reference.
Figure 3-24: Risk Page>
Image from Visible Decisions.
Figure 3-24: Risk Page>
Image from Visible Decisions.Significant amounts of text are used for titles, legends, quantitative details and annotations.
Figure 3-25: F500 Stack
Image from Visible Decisions.
Figure 3-25: F500 Stack
Image from Visible Decisions.Many different attributes per item are shown as "wafers" on a stack. Interactions such as highlighting wafers and sorting stacks enable comparison across diverse attributes.
Figure 3-26: Worldwide Performance
Image from Visible Decisions.
Figure 3-26: Worldwide Performance
Image from Visible Decisions.Maps require strategies for areas of congestion, such as the leaders shown above, or selective magnification, or zooming.
Figure 3-27: Radial Hierarchy
Image from Visible Decisions.
Figure 3-27: Radial Hierarchy
Image from Visible Decisions.There are many possible representations for hierarchies. In the above image, the root of the hierarchy is at the center, with subsequent levels further out.
Figure 3-28: Plastic Mold
Image from Visible Decisions.
Figure 3-28: Plastic Mold
Image from Visible Decisions.Five timeseries with 3 dependent variables. The timeseries are plotted based on the dependent variables. It is difficult to precisely locate the time series.
Figure 3-29: Directed Graph
Image from Visible Decisions.
Figure 3-29: Directed Graph
Image from Visible Decisions.A network with many links, each with different properties and direction is challenging to represent. Vertical separation and color coding differentiate links.
Through the construction, use and analysis of numerous information visualizations, many different representational and interaction techniques have been experimented with and considered. The accumulated knowledge has led to a slow evolution of successful techniques, such as brushing, animation, simple navigation, organized visual layouts and connotative representations. This knowledge is summarized as guidelines in the next section.
Next Section 4. Guidelines and Metrics
© Copyright by Richard Karl Brath 1999
© Copyright by Richard Karl Brath 1999