Catalog & Classify

Who are they?

Created in 1973 by Herman Chernoff, this data visualization system allows multiple variables to be displayed at once. This whimsical system may challenge our understanding and possibly preconceived ideas of what a dataset can look like. Variables are represented by facial features on a caricature of an individual. Perhaps a fun substitute to the classic bar, line, and circle graph, Chernoff faces attempts to incorporate a personal touch to stale geometric numbers and shapes. Created based on the assumption that we, as humans, can easily recognize and read each others’ faces; therefore, we, as data consumers, should be able to recognize the small differences when these features represent diverse variables.

How do they work?

Condensing and organizing the data into categories, each feature on the face will visually represent a category. The range of variables within each category can be differentiated by the size, position, or color of the facial feature. For example, one could indicate pupil size to indicate the range of data for how many times it has snowed in a week. If the pupil is larger, it might indicate that it has snowed more frequently, the precise numeric data can be clarified in a legend, while a smaller pupil might represent less snowfall.

The type of data Chernoff faces can display are quite varied, anything from a range to precise observations. There isn’t too much limitation to the variety of data these faces can visualize, as long as there is a category associated with a facial feature then it can most likely be visually displayed.

There are many different combinations when designing a Chernoff face, which is why organization for this visual representation system is quite imperative. Legends are also essential when defining what each facial feature will represent. Because it is a visual system, it is important to clearly label and categorize to reduce the risk of visual pollution and encourage succinct communication.

Advantages?

Visual comparison. The advantages of Chernoff faces are their quick and accurate depictions of comparative data. Each data or variable does not need to be compared to its counterparts in the same category, as Chernoff faces provide an overall image for quick visual comparisons. Chernoff faces are also easily compared without reading or comprehending numbers, letters, coordinates, etc. They predominantly focus on the distinct visual differences between the illustrated faces rather than numeric differences. These visual representations, varying in shape, position, color, and more, can make spotting differences in data immediate and accessible. In addition, the concept of pareidolia accentuates the favorability that humans have with faces. The concept of pareidolia is the tendency to perceive a stimulus as an object known to the observer, in other words we tend to see faces and patterns in inanimate objects or hear hidden messages in audios. This makes Chernoff faces more exciting and intriguing compared to the longstanding datasets we’ve grown accustomed to in our basic statistics courses.

Disadvantages?

Although Chernoff faces were created based on the assumption that humans can easily detect minute differences in facial expressions, these illustrations for data representation are more challenging to read than a friend’s outward emotions. With seemingly endless variations and representations of facial features, Chernoff faces can become challenging to understand. Elements of these features (position, size, color, shape, etc.) can become muddled and visually overwhelming. There can be various categories to be aware of when reading and comparing the faces. Too many and they can alter the appearance of the faces, sometimes creating unwanted attention towards a particular variable. As social creatures, we also tend to focus on the emotions of inanimate faces, such as reading a Chernoff face as “sad” and negatively connotated because of the downward slope of the “mouth”. Reading faces can also lead to unintentional stereotyping and distortions in a caricature, especially when they are representing groups of people or a specific individual. There is also the controversy that Chernoff faces can misrepresent a “race” by utilizing visual stereotypes and data associated with a certain group of people. It does boil down to an individual’s interpretation of these faces emphasizing the subjectivity connected with human faces/expressions. Chernoff faces also have the disadvantage of personal hierarchy, as individuals tend to have their own hierarchy when identifying a face, some features are more prominent and visible to others than to ourselves. While we can recognize faces quickly, effectively differentiating and comparing features is much more difficult than Herman Chernoff theorized.

Good (or 'decent') Examples

· Fig 1: depicts a "good" example of a Chernoff face because of its clear incorporation of map and faces without visually muddling the data. Overall, it is reasonably understandable due to the simplistic variations in facial features with clarification and context in the legend placed in the bottom left.

· Fig 2: depicts a "good" example of a Chernoff face key or legend, it thoroughly organizes the facial differences into categories and clearly marks the range of data it represents

· Fig 3: depicts a "good" example of a Chernoff face because there is clear separation of legend, variables and facial features, as well as labeling the participants in the study (my only suggestion would be the addition of color to accentuate further clarity of categories)

Bad Examples

· Fig 4: depicts a "bad" example of Chernoff faces because of the lack of clarity. The legend is difficult to understand, the changes in facial expressions are very subtle, the pairing of the map and faces is visually overwhelming, and Trump's tears makes it seem as if it is negatively connotated and he lost all those states...also they are portrayed with similar skin tones, and we all know what color his should be

· Fig 5: depicts a "bad" example of Chernoff faces because it is so convoluted and confusing! There is little to no variation in the facial features and the size of the map and counties makes it seem overcrowded and constricted.

· Fig 6: depicts a "bad" example of Chernoff faces because, well, a graph AND faces? I'm not even sure where to start with this critique.

Resources

https://eagereyes.org/criticism/chernoff-faces

https://flowingdata.com/2010/08/31/how-to-visualize-data-with-cartoonish-faces/

Parallel-Coordinate Plots

What are they?

Parallel coordinate plots visualize multiple factors and, more importantly, show their relationships to one another. Each variable is given its own vertical axis, and the values for each variable is typically represented as an integer that falls along the vertical axis for said variables. Commonly the variables that are most related move from right to left to demonstrate relationships between them more easily given Western cultures read from right to left.

When they work best

Parallel-coordinate plots work best for showing the relationships between multiple variables. It's important to arrange the vertical axis in such a way that they clearly show the relationships between variables and the effects of the variables have on the values assigned to what is being measured.

Downsides

If too many 'items' are being measured by the variables, the visualization can quickly become cluttered and difficult to read and identify the key findings. The best way way to address this downside is using a technique called "brushing" where a key line or series of lines are highlighted and the surrounding lines are faded into the background. This allows the viewer to isolate the sections and plots that matter the most.

Bad Examples

Figure 1: This is parallel-coordinate plot could be better. In this example, the vertical axis are ordered in such a way that it's more difficult to see subsequent relationships since the middle axis convolutes the interpretation of the relationships between one another.

Source: https://www.data-to-viz.com/graph/parallel.html

Figure 2: Given the absence of color or other labels in this parallel coordinate plot, it's impossible identify the different models or separate singular lines from one another. This plot suffers from clutter and cannot be read or establish what 'items' are which as they move along the visualization.

Source: https://eagereyes.org/techniques/parallel-coordinates

Figure 3: This parallel-coordinate plot suffers from trying to measure too many 'items,' in this case, countries. It is impossible to identify which line easily corresponds to which country and the natural disasters that occurred in the given year.

Source: https://www.mathworks.com/help/matlab/ref/parallelplot.html

Good Examples

Figure 4 is the same information as Figure 1, however, the vertical axises have been rearranged to better read from left to right and understand the relationships between the axises being analyzed.

Source: https://www.data-to-viz.com/graph/parallel.html

Figure 5 is relatively clear. Though there is some overlapping points, there are few enough lines to easily identify which color corresponds to which produce and which type and amount of carbohydrate they contain.

Source: https://docs.tibco.com/pub/spotfire/6.5.0/doc/html/para/para_what_is_a_parallel_coordinate_plot.htm

Figure 6 addresses the issue of clutter very well. It pulls out the key line apart from the others using color to help the viewer identify which is the key line being measured. However, without data labels, a viewer must look to the top of the graph in order to understand what each of the data points represents numerically.

Source: https://www.vizwiz.com/2019/07/parallel-coordinates.html

Star Plots

What are they?

Star plots are a type of visualization that, similar to parallel-coordinate plots, measure multiple variables and allow the viewer to see the relationships between the points being measured and represented. They are called star plots specifically because as the points radiate from the center, similar to how a star's points move outwards from a central locus.

When they work best

A star plot works best for identifying outliers in relationships and when the data is more similar/consistent with other relationships.

Downsides

The primary downside of a star plot is the number of variables that can be visualized. Once too many variables are being measured, it can become increasingly challenging to identify the relationships between them.

Bad Examples

Figure 7 is trying to measure too many variables along the outside of the plot. Given the number of variables being measured, it's more challenging to see the minute relationships between the points. However, in this figure, it is easier to see the 'outliers' but almost impossible to read the smaller details in that aren't those specific outliers.

Source: http://rstudio-pubs-static.s3.amazonaws.com/5795_e6e6411731bb4f1b9cc7eb49499c2082.html

Similar to Figure 7, Figure 8 is also measuring too many variables. Though it is only measuring 2 'items' (in this case, Blue Collar Communities vs. Prospering Suburbs) it is very challenging to see what the relationships are in the variables.

Source: http://rstudio-pubs-static.s3.amazonaws.com/5795_e6e6411731bb4f1b9cc7eb49499c2082.html

Good Examples

Interestingly enough with Figure 9, it's focusing on the relationships between the points of the pentagons instead of demonstrating the numbers themselves. In this sense, without statistical data, it is actually easier to 'read' the relationships of the different points on the pentagons

Source: https://fivethirtyeight.com/features/the-5-key-constituencies-of-the-2020-democratic-primary/

Similar to Figure 9, Figure 10 does not focus on numerical data but again focuses on the relationships of the points on the char and how the two 'items' (Carl and Lucy) compare to one another. The two items being measured and limited number of axis make this a good example of a star plot.