Abstract
Graphical representations of political issues and economic trends are an increasingly popular means of conveying information to the public. However, graphs have the potential to shape public opinion by visually emphasizing or downplaying the information they convey. I randomly assign subjects to view graphs that represent the same underlying information but that differ in relative emphasis: one is consistent with a textual account of rising inequality, while the other de-emphasizes the same information by increasing the scale of the Y-axis. My results indicate that graphical frames provide powerful contextual cues: for Republicans and conservatives, exposure to the de-emphasizing graph results in a 40% decrease in expressed support for intervention against inequality relative to Republicans and conservatives in the control condition, despite the fact that both groups read the same textual information. My findings reveal how an increasingly important and unexamined form of political communication affects public opinion, also suggesting promising avenues for future research.
Politicians, campaigns, advocates, and the news media use graphical representations of complex policies to convey information and persuade the public. Barack Obama’s 2013, 2014, and 2015 State of the Union addresses, for instance, were presented online via “enhanced broadcasts” that each contained a slideshow featuring graphs that depicted economic time series and policy consequences. But while political graphs are popular forms of political communication (Grabe and Bucy, 2009), academic research on graphical representation has focused almost entirely on scientific graphs (Cleveland and McGill, 1985; Schwabish, 2014; Tufte, 2001), ignoring graphical communications from campaigns and news media. Data visualization experts often recommend complexity and precision, but political campaigns and advocates often strive for simplicity and persuasion. 1
While developed lines of research examine the effects of visual elements in surveys (Couper et al., 2004; Fuchs, 2009; Tourangeau et al., 2013) and political imagery more generally (Gadarian, 2014; Petersen, 2011; Prior, 2014; Sullivan and Masters, 1988; Swigger, 2012), existing scholarship on political graphs does not identify the specific visual attributes that lead to opinion change. Graphs that compare income share for the top and bottom 20% of earners lead conservatives to be more likely to identify inequality as “bad,” though no more likely to support government intervention against it (Piston, 2014). Providing graphical representations of climate change, job growth, and insurgent attacks in Iraq lowers reported misperceptions about each issue (Nyhan and Reifler, 2013). Graphs also shape whether individuals accept information. Tal and Wansink (2014) show that including scientific graphs alongside textual information makes that information more credible and persuasive, especially for individuals who express support for science in general. However, these studies do not compare multiple graphs that all represent the same information; as a result, it is unclear what specific features or characteristics of graphs affect opinion formation.
In this study, I provide textual information about increasing income inequality to a group of experimental subjects and randomly assign them to view different graphical representations of the same information. I then assess redistributive preferences, identifying the effect of graphical emphasis on redistributive opinion formation. My results indicate that graphical frames have large effects: for Republicans and conservatives, the inclusion of a graph that de-emphasizes the information presented in text results in a 40% decrease in expressed support for intervention against inequality.
Graphical framing and opinion formation
Political stimuli shape the attitudes expressed by individuals (Chong and Druckman, 2010; Druckman, 2014). By combining both images and words, graphs engage both verbal and visual processing systems (Paivio, 1986) and increase individuals’ abilities to organize relevant information into coherent mental representations (Mayer, 2005; Prior, 2014). As a result, individuals retain and recognize visual information more efficiently and for longer periods of time than if they were exposed to verbal information (Grabe and Bucy, 2009: 17). However, graphs also necessarily frame issues. The graphs I examine create “framing effects” in the most precise sense of the term; since they are logically equivalent ways of representing the same information, they differ only in relative emphasis (Druckman, 2004). I expect that these effects are likely to persuade individuals to express opinions consistent with relative graphical emphasis.
All graphs lead the viewer to perceive a spatial relationship from a particular point of view; there are no universal criteria for determining whether a graph is leading, misleading, unfair, or biased. The very act of illustrating a complex policy issue with a particular line or shape circumscribes alternative interpretations and excludes competing representations. By emphasizing or deemphasizing the contrast between quantities, graphs provide visual emphasis frames that guide opinion formation. Such graphs should also encourage particular kinds of issue evaluations. When a graph deemphasizes information conveyed in text, I expect that individuals will be less likely to think that the information is politically important. Thus, their evaluations of the issue should change.
At the same time, individuals engage in motivated reasoning; they tend to evaluate new sources of information in a way that is consistent with preexisting beliefs (Taber and Lodge, 2006), especially when they perceive information to align with partisan and ideological commitments (Slothuus and de Vreese, 2010). If graphical emphasis is more consistent with partisan or ideological priors than textual information, graphs should have a larger influence on reported opinions. I expect that when graphical emphasis coincides with an individual’s preexisting partisan and ideological considerations about an issue, the individual will be more likely to engage in motivated reasoning, discounting textual information in favor of the graphical representation.
In order to test the effects of graphical emphasis, I provide the same textual information about rising inequality to a sample of experimental participants, and then randomly assign subjects to view graphs that depict the textual information in different ways.
The first graph, appearing in the top panel of Figure 1, shows the pre-tax income share for the top 1% of earners from 1913 to 2012 with a truncated Y-axis of [5, 25]. The graph is based on Piketty and Saez’s (2014) research, and is available to the public on Saez’s website. The graph in the lower panel of Figure 1 depicts identical data, but with a Y-axis ranging from [0, 100]. As a result, the large contrast between the postwar years and the present is deemphasized; the overall pattern appears more like a straight line. The top graph emphasizes a change over time by restricting the Y-axis to [5, 25], while the bottom graph de-emphasizes that change by including the complete Y-axis. Of course, much of the space represented in the lower graph is empty – the top graph conveys subtle changes over time much more precisely.
Graphs of income inequality of time.
Data and methods
From April 28–May 6 2014, I contracted with Survey Sampling International (SSI) to draw a sample of 540 individuals, which were meant to be representative of the US as a whole by gender, ethnicity, age, and education. 2 This sample is more ideologically representative of the US public than convenience samples of students or online labor market samples. However, it is slightly less conservative and Republican than the US public: 30% of the sample identified as conservative, comparable to the 40% of conservative identifiers in the 2012 American National Election Study (ANES) (weighted). Further, 27% of the SSI sample identified as Republican, compared with 39% of 2012 ANES respondents. Subjects were randomly assigned to three groups: a control group that saw no graphs and two treatment groups that each saw one of the graphs depicted in Figure 1.
At the beginning of the survey, all three groups read the following text: Some say that income inequality has been increasing in America. In the 1920s, the top 1% of the income distribution earned about 20% of all income. From the 1940s through the early 1980s that number decreased to about 10%, but now the top 1% earns a little more than 20% of all income.
Group 2 viewed the text-consistent graph (top panel, Figure 1), which emphasizes the change over time described in the text by focusing on the area of the graph between 5% and 25%. Group 3 viewed the de-emphasis graph (bottom panel, Figure 1). Group 1 received only the textual information. After exposure to the stimuli, all respondents were asked: “Do you favor, oppose, or neither favor nor oppose the government trying to decrease income inequality?” Respondents indicated favor or opposition on a seven-item Likert scale which I recode 0–1, and 54% of respondents expressed support for decreasing income inequality. 3
I expect that exposure to the graph that de-emphasizes the information conveyed in the text will increase the number of respondents who oppose intervention. Table 1 shows the results for individuals in each of the three groups.
Average inequality intervention opinion by treatment group.
The table shows that exposure to the de-emphasis graph had a statistically significant negative effect on support for government intervention against inequality, relative to the text only group (p < .01, two-sided test) and the unchanged graph group (p < .05 two-sided test). The effect of the unchanged graph relative to the text only group is not statistically significant (p = .56).
On average, individuals who were exposed to the de-emphasis graph were about 10 percentage points less supportive of intervention against income inequality, plus or minus 3.4 points, confirming Hypothesis 1. Support for intervention among those who saw the text-consistent graph is not statistically distinguishable from support among those that read the text alone. By contrast, the de-emphasis graph minimized the change in income distribution depicted in the information-consistent graph (and conveyed in the text) and, in so doing, decreased support for intervention against income inequality.
As Hypothesis 2 suggests, I expect that this effect will be more pronounced for subsets of the sample. Specifically, I expect that the de-emphasis graph will have larger effects on intervention support among Republicans and conservatives than among Democrats and liberals. In order to evaluate this claim, I examined average responses across subsamples of the data.
As Figure 2 shows, all respondents became less supportive of intervention after seeing the graph with the full Y-axis – the de-emphasis graph. But seeing the de-emphasis graph has a significant effect only for Republicans and conservatives; among those subjects, support drops by 20.6 and 19.9 percentage points, respectively. This large and statistically significant effect dwarfs the modest shift observed for Democrats and liberals; it represents a decrease in overall support of 40% in each case.
Average inequality intervention support by de-emphasis graph exposure.
To understand how the strength of Republican and conservative identity affects these results, I estimate Ordinary Least Squares (OLS) regression equations, which appear in Table 2. As the Table shows, exposure to the de-emphasis graph had a modest but significant effect on all respondents’ opinions, as discussed above. But for strong Republicans and strong conservatives, exposure to the de-emphasis graph has much more dramatic effects: the average treatment effect for strong Republicans is a 0.26 decrease in support for intervention, while for Republican leaners, the effect is a more modest decrease of 0.13. For extreme and slight conservatives, the comparable effects are −0.29 and −0.14, respectively. As Republican and conservative identification increases, subjects appear increasingly likely to engage in motivated reasoning and accept the de-emphasized graphical representation of inequality information rather than the textual account.
Inequality intervention opinion among those predisposed against information.
p<.05; **p<.01.
Discussion and conclusion
In this research note, I provide the first systematic evidence of how graphical style affects opinion formation. All subjects in the experiment read the same textual information, but the emphasis that particular graphs conveyed profoundly shaped reported opinions. I found a modest, but significant effect for all subjects exposed to the de-emphasis graph, confirming Hypothesis 1. But the effect was especially pronounced for individuals who were likely to be predisposed to the position advanced by the de-emphasis graph, confirming Hypothesis 2. This result strongly suggests that graphs provide individuals with powerful interpretive frames; those who are predisposed to accept the information contained in a graph do so when asked to evaluate an issue.
Further research on graphical style (following Cleveland and McGill, 1985) will refine investigation into the aspects of a particular graph that frame issues and shape evaluations most effectively, even for respondents with less motivation to look carefully. An important limitation of the studies reported here concerns temporal scope; I am unable to test how long effects persist and if those effects shape future political behavior. If simple graphs create durable, image-based frames for imagining political issues, it is likely that individuals will rely on those images long after exposure. Indeed, these graphs could provide the primary cognitive representations for otherwise abstract and difficult-to-imagine political and economic issues, especially because individuals process visual information more efficiently than textual information.
Walter Lippmann first described public opinion using visual terms: “The pictures inside the heads of these human beings, the pictures of themselves, of others, of their needs, purposes, and relationship, are their public opinions” (1922: 18). Yet research on public opinion and political communication almost uniformly focuses on verbal representations of political issues. Graphs provide a framed picture of a political issue; a fixed point of view that, in turn, shapes the pictures in people’s heads. As the proliferation of political images continues via new media and electronic political communications, scholars must take a closer look at graphs and how they affect public opinion.
Footnotes
Acknowledgements
I thank Lynn Sanders, Nick Winter, Eric Patashnik, Spencer Piston, Jon Kropko, Craig Volden, Paul Freedman, Heidi Schramm, Akitaka Matsuo, Victoria Shineman, Nicholas Nicoletti, and participants in the University of Virginia American Politics Working Group for their comments and suggestions.
Declaration of Conflicting Interests
The author(s) declared no potential conflicts of interest with respect to the research, authorship, and/or publication of this article.
Funding
The author(s) received no financial support for the research, authorship, and/or publication of this article.
