Jason McCarley - Principles of good graph design Flashcards Preview

Psychology - Honours year > Jason McCarley - Principles of good graph design > Flashcards

Flashcards in Jason McCarley - Principles of good graph design Deck (26):

Outline the differences between and Analog and a digital display.

In a digital display (e.g., table) - differences in physical magnitude are not correlated with differences in values. (differences in number size is not reflected in the visual display.

In an analog display (e.g., graph) - the values are depicted through the physical magnitude of display properties. (differences in number values is reflected in the visual display)


What is an analog display?

In an analog display (e.g., graph) - the values are depicted through the physical magnitude of display properties. (differences in number values is reflected in the visual display)


What is a digital display

In a digital display (e.g., table) - differences in physical magnitude are not correlated with differences in values. (differences in number size is not reflected in the visual display.


When would a table be more appropriate than a graph?

A table is useful if the reader needs to specific values (e.g., stock market - do you want to see if it has spiked or do you want to know exactly how much money you have lost?)


Can format of a data display affect behaviour? (give example.

YES risk-taking behaviour is influences more by a graph than by an alphanumeric representations. A stick-figure representation of your risk makes the information more salient.


List the 5 principles of good graph design

1 - Use appropriate visual dimensions to convey information (position, line length are less susceptible to magnitude judgements distortions/bias)

2 - Make important information easily dicriminable (minimise clutter and adornments)

3 - Choose the graph format that fits you message

4 - Reduce working memory demand

5 - Don't surprise the reader

(also make graphs and text consistent - lead the viewer)


What are the appropriate visual dimensions to convey information in graphs?

The best dimensions are those less susceptible to magnitude judgement distortion (when people perceive magnitude changes incorrectly as being either larger [expansion] or smaller [compression] than they actually are), these include Position and Line length (but not circles,area, brightness, loudness, heaviness).


Should colour be used in graphs?

It can be but it should be used redundantly (lots of people are colour blind). It is most useful for distinguishing objects from the background or other objects (e.g., as in a brain scan to seperate brain regions) - when finite detail/specific values are not needed.

Hue is NOT good for metric data becaue hues do not have a natural small - > large. However, Saturation can be acceptable because purity/intensity of a hue can be used to express metric data - BUT IS SUBJECT TO expansion compression)

Color can be used to represent metric information (e.g., the closer to red the greater the debt, or the deeper the color saturation the higher the
population) or to mark variables (e.g., the red bars represent girls and the blue bars represent boys; Brockmann, 1991; Cleveland and McGill, 1985; Hoffman et al., 1993). However, color does not accurately represent precise quantitative information (Cleveland and McGill, 1985) and may even be misleading when used for continuous data. For example, certain colors are more likely to be interpreted as “higher” or “lower” in displays such as contour plots, influencing how well viewers are able to imagine these plots three-dimensionally. Furthermore, colors may be interpreted as representing
categorical data even when they are intended to convey continuous information (Phillips, 1982).

color has certain benefits. One major use of color can be to group elements in a display. For example, color can help viewers group data in scatterplots (Lewandowsky and Spence, 1989); for example, different colors
can be used for “boys” and “girls” in a scatterplot of height versus age. Similarly, if two pieces of data in separate graphs (such as two pie charts) are
compared, they are more likely grouped if they are the same color (Kosslyn,
1994). Another potential benefit of color is to reduce the difficulty viewers face in keeping track of graphic referents because of the demands imposed
on working memory (e.g., remembering that the line with open circles represents
a temperature of 70± in Figs. 6(a) and (b); Fisher, 1982; Kosslyn, 1994; Schmid, 1983). Indeed, in a recent study examining viewers’ eye fixations as they interpreted graphs, we showed that viewers must continuously
reexamine the labels to refresh their memory (Carpenter and Shah, 1998). If those dimensions are represented with a meaningful color choice, such as red for warmer temperatures and blue for cooler temperatures, it might
help viewers keep track of variable names (Brockman, 1991).

the use of semantically related features is highly dependent on assumptions shared by the graphic designer and graph reader. For example, green
means profitable for financial managers but infected for health care workers (Brockmann, 1991). Nonetheless, the use of semantically related features
may be especially beneficial when presenting graphs to children, who, in general, have fewer working memory resources than adults (Halford et al., 1998). In conclusion, color can provide helpful cues especially with respect to helping viewers keep track of quantitative referents, but is inadequate as the only source of precise quantitative information.


What should not be included on graphs (i.e., to reduce clutter)

Anything that is not necessary, including: shadows, 3D (can occlude/degrade information and be difficult to intepret) , thick lines, heavy grid lines, filled backgrounds.

These make it difficult to read.


What are the 4-types of 'message' people are usually trying to get across with graphs?

* Point reading - single values/data points "What was factory A's production in 1958?"
* Local comparison/1st order comparison - "Was factor B's production greater in 1996 or 1997?"
* Global comparison/2nd order comparison - "Was factory Bs total production in 1996 and 1997 less than factory As production in these years?"
* Synthesis - What will factory As production be in 1999, is factory As production increasing/decreasing? (trends)


What is the proximity-compatability principle?

graph format is necessarily better overall than any other format. Instead, there is an interaction of task and graphic format, called the proximity compatibility principle byCarswell andWickens (1987; see also Simkin and Hastie, 1986). Integrated, object-like displays (e.g., a line graph) are better for integrative tasks, whereas more separable formats (e.g., bar graphs) are better for less integrative or synthetic tasks such as point reading (Carswell and Wickens, 1987).


If you want readers to mentally isolate
data points, present them in a visually isolated form. If you want readers to mentally integrate data
points, present them in a visually integrated form.


(Divided attention between two information sources may be necessary for the completion of one task. These sources must be mentally integrated and are defined to have close mental proximity. Information access costs should be low, which can be achieved in many ways (e.g. proximity, linkage by common colors, patterns, shapes, etc.). However, close display proximity can be harmful by causing too much clutter.)

The PCP depends critically on two dimensions of proximity or similarity: perceptual proximity and processing proximity.

Perceptual proximity (display proximity) defines how close together two display channels conveying task-related information lie in the user's multidimensional perceptual space (i.e., how similar they are).

Mental or processing proximity defines the extent to which the two or more sources are used as part of the same task. If these sources must be integrated, they have close processing proximity. If they should be processed independently, their processing proximity is low. The principle proposes a compatibility between these two dimensions. If there is close processing proximity, then close perceptual proximity is advised; conversely, if independent processing is required, distant perceptual proximity is prescribed.

Displays relevant to a common task or mental operation (close task or mental proximity) should be rendered close together in perceptual space (close display proximity


What is a "point-reading" 'message'?

single values/data points "What was factory A's production in 1958?"


What is a "local comparison/1st order comparison" 'message'?

"Was factor B's production greater in 1996 or 1997?"


What is a "Global comparison/2nd order comparison" 'message'?

"Was factory Bs total production in 1996 and 1997 less than factory As production in these years?"


What is the "synthesis" 'message'?

TRENDS - What will factory As production be in 1999, is factory As production increasing/decreasing?


When would you use a bar graph?

When ytou want to encourage attention to isolated points. (absolute values)


bar graphs emphasize discrete comparisons (Carswell andWickens, 1987; Shah et al., 1999; Zacks and Tversky, 1999). Furthermore, bar graphs of multivariate data appear to be less biasing than line graphs.

Viewers are much less biased in their descriptions of the relationships in this bar graph and describe the effects of room temperature and noise level on achievement
test scores equally often (Shah and Shellhammer, 1999).


When would you use a Line graph?

Emphasise trends, make interactions easier to spot. Encourages the perception of continuous variables on the X-axis, even if variables are actually discrete so be careful. Can make us attend to trends but ignore other comparisons or effects. (absolute values)


Viewers are more likely to describe X-Y trends when viewing line graphs over bar graphs (Carswell
et al., 1993; Shah et al., 1999; Zacks and Tversky, 1999)

Viewers are alsomore accurate in retrieving x–y trend information from line graphs than from bar graphs (Carswell and Wickens, 1987).

Even when two discrete data points are plotted in a line graph, viewers (college students at Stanford) sometimes describe the data as continuous.

This emphasis on the x–y trends can lead
to incomplete interpretations of data when the data are complex (for example, multiple lines on a display representing a third variable).


When would you use a pie chart? (and divided bar chart)

Emphasize proportions, encourage part-whole comparisons.

Used for presentation of proportion data. Pie charts best for proportions because divided bar graphs require adding up information for different parts of the bar.

Divided bar charts best when proportions AND absolute values are needed.


Should graphs have a legend?

Yes, but you can just label the lines if that is feasible - reduces working memory by reducing load and need to scan.

Another choice a graph designer makes is whether to use a legend (or key) or to directly label graph features (such as lines and bars) according to their referents. Because legends or keys require that graph readers keep referents in memory, legends pose special demands on working memory; see, for example Fig. 6(b). Thus, the conventional wisdom is that graph designers
should avoid using legends (except when labels would lead to too much visual clutter) and instead label graph features directly with their referents (Kosslyn, 1994) as in Fig. 6(a). Again, this advice may be particularly important when presenting graphs to children.


What things on a graph might surprise the reader?

When the Y-axis does not cover a plausible range of values. when spacing on X-axis is not even (e.g., 1 year - 1 week).

Choose your axes and other graph properties in a way that conforms to the readers’ expectations.


What is a magnitude estimation task?

subjects assign numbers to describe physical stimulus properties
such as size, brightness, or loudness.
• If perceptions are unbiased, ratings are proportional to the physical magnitude.
• Response compression exists when differences in magnitude are underestimated.
• Response expansion exists when differences in magnitude are overestimated.


What is Steven's power law?

describes the relationship between physical magnitude and perceived

P = kSn,

where P = perceived magnitude, k = an arbitrary constant, S = true magnitude, and n = Stevens’

n < 1: response compression
n = 1: unbiasedness
n > 1: response expansion


What are the implications of Steven's power law for graph design?

we should try to represent data with
properties that don’t cause response compression or expansion.


What are the three major components processes for graph comprehension?

what do they imply?

1. Encoding the visual array and identifying the important visual features (such as a curved line). - accuracy and grouping of the information is influenced by the inherent biases and limitations of our perceptual apparatus.

2. Relating the visual features to the conceptual relations that are represented by those features - this component is influence by the information encoded in step 1 and the individuals experience/way of interpreting it. Some mappings between visual features and referents are more intuitive. (e.g., horizontal lines for distance travel, but vertical lines for amount [as in taller piles])

3.The third component process of graph comprehension is that viewers must determine the referent of the concepts being quantified (e.g., Population,
Rural South, etc.; in Fig. 1) and associate those referents to the encoded functions (Bertin, 1983). As discussed in more detail later, students’ and/or novices’ interpretations are often colored by their expectations of the content (Shah, 1995; Shah and Shellhammer, 1999).

These three processes imply that three factors play an important role in determining a viewer’s interpretation of data: the characteristics of the visual display (bar or line graph, color or black and white, etc.), knowledge about graphs (graph schemas), and content (e.g., age vs. height, time vs. distance).


What does it mean to say there is a trad-off between specific and intergratic data in visual displays?

there is a trade-off between the ability to accurately perceive specific quantitative facts and the ability to get a more qualitative gist of relationships depicted in the data. A table, for example, allows people to get single point values most accurately but provides the least integrative information (Guthrie et al., 1993).


How might a viewers knowledge about graphs affect how they encode and remember pictures and diagrams?