Foreground:background salience: Explaining the effects of graphical displays on risk avoidance

https://doi.org/10.1016/S0749-5978(03)00003-7Get rights and content

Abstract

The purpose of this research was to determine the mechanisms underlying the graphical effect identified by Stone, Yates, and Parker (1997), in which graphical formats for conveying risk information are more effective than numerical formats for increasing risk-avoidant behavior. Two experiments tested whether this graphical effect occurred because the graphical formats used by Stone et al. highlighted the number of people harmed by the focal hazard, causing the decisions to be based mainly on the number of people harmed (which we label the “foreground”) at the expense of the total number of people at risk of harm (which we call the “background”). Specifically, two graphical formats were developed that displayed pictorially both the number of people harmed and the total number at risk, and use of these display formats eliminated the graphical effect. We thus propose that the previously discussed graphical effect was in fact a manifestation of a more general foreground:background salience effect, whereby displays that highlight the number of people harmed at the expense of the total number of people at risk of harm lead to greater risk avoidance. Theoretical and practical implications are discussed.

Introduction

The issue of how best to communicate risk information has become an increasingly important part of the risk assessment discipline (see, e.g., Fischhoff, 1995; Fisher, 1991). Although the risk communication field is very broad, most of the work has focused on one of three goals: (a) increasing knowledge about the risks, (b) modifying risk-relevant behavior, and (c) facilitating cooperative conflict resolution (Lipkus & Hollands, 2000; Rohrmann, 1992). The focus of the present work is on the second of these goals, modifying risk-relevant behavior. Consider, for example, communication attempts designed to influence teenagers not to take drugs or to practice safe sex. Gaining an understanding of what techniques are the most effective for these sorts of goals can assist these intervention attempts in a number of situations. More specifically, the present work examined methods for modifying risk-relevant behavior by means of alternative methods of displaying risk magnitude information.

Much of the research on modifying risk-relevant behavior has examined situations where the probability of the undesired event is quite small. Indeed, a large number of the hazards that have led to intervention attempts (e.g., of dying from an overdose of a drug or contracting AIDS from practicing unsafe sex) have low probabilities of occurring per instance. However, when these actions accumulate over time and people, the resulting effects become quite serious. In addition, most people have limited practical experience dealing with low-probability events. A number of researchers have thus suggested that people have difficulty reasoning on the basis of these low-probability risks (see, e.g., Camerer & Kunreuther, 1989; Covello, von Winterfeldt, & Slovic, 1986; Fisher, 1991; Fisher, McClelland, & Schulze, 1989; Halpern, Blackman, & Salzman, 1989; Lipkus & Hollands, 2000; Magat, Viscusi, & Huber, 1987; Roth, Morgan, Fischhoff, Lave, & Bostrom, 1990; Stone et al., 1994, Stone et al., 1997). For example, Fisher et al. (1989) suggested that people either dismiss low-probability events entirely or else focus primarily on the size of the expected loss (similar arguments are given by Halpern et al. (1989), Magat et al. (1987), and Stone et al. (1994)). This type of argument is in keeping with prospect theory (Kahneman & Tversky, 1979), which suggests that small probabilities are either “edited” to zero or else overweighted, as well as with fuzzy trace theory (Reyna & Brainerd, 1991), which suggests that people encode the gist of the available information (e.g., that the probabilities are quite small) and, whenever possible, reason on the basis of this gist rather than on finer distinctions among the probabilities.

These arguments have motivated a variety of different proposed techniques for conveying low-probability risk magnitudes. In one classic study, Slovic, Fischhoff, and Lichtenstein (1978) told participants either that the chance of experiencing at least one disabling injury when driving without a seatbelt is .00001 for each trip or that the probability is .33 over 50 years of driving. Although these statistics are formally equivalent, the latter frame, aggregated over 50 years of driving, produced more risk-avoidant behavior on the part of the participants. More recently, Stone et al. (1994) showed that presenting risk information in relative risk form (e.g., that a safer product reduces the risk to half that of another product) led to more risk-avoidant behavior than simply giving the risk magnitudes for the two products (see also Baron, 1997; Halpern et al., 1989). Siegrist (1997) showed that, under certain conditions, providing risk information via a frequency format (e.g., 600 out of 1,000,000 people will die) leads to more risk-avoidant behavior than simply providing the risk magnitudes in probability (incidence rate) form. Weinstein, Kolb, and Goldstein (1996) used the time intervals between expected events to communicate risk magnitudes, and showed that as long as the time intervals were long, presenting them in addition to the risk magnitudes led to a smaller perceived need for action (less risk avoidance).

A number of authors (e.g., Covello, Sandman, & Slovic, 1988; Keeney & von Winterfeldt, 1986) have proposed that the presentation of risk information in graphical form should be an especially effective means of increasing risk-avoidant behavior. Stone et al. (1997) demonstrated empirically the efficacy of such recommendations, by showing that graphical techniques for displaying risk information can indeed be more effective than simply providing numerical information for highlighting the risk reduction accorded by a safer product. Specifically, they provided information about the risk associated with either tire blowouts or the development of periodontal disease. For example, they told participants that with “Standard Toothpaste,” 30 out of 5000 people would develop periodontal disease. They then informed participants about “Improved Toothpaste,” which is identical to Standard Toothpaste except for the fact that it reduces the risk to 15 out of 5000. In their first experiment, the number of people developing periodontal disease was displayed either numerically by the numbers “30” and “15” or graphically by means of stick figures illustrating the people developing periodontal disease. Participants were willing to pay more for the safer product when the risk was displayed via the stick figures format than by the numbers display. In two subsequent experiments, Stone et al. demonstrated that using stick figures to represent the risk was not essential to the effect, in that similar effects held for other graphical displays, in particular, asterisks and bar graphs. (See Fig. 1 for the numbers and asterisks formats used in the Stone et al. study.)

Although Stone et al. showed that depicting risk information graphically as opposed to numerically is a useful technique for increasing risk-avoidant behavior, they only speculated as to the mechanisms responsible for this “graphical effect.” As Lipkus and Hollands (2000) discuss, this somewhat atheoretical approach to the study of graphical displays of risk is common, despite the fact that theoretical explanations for these sorts of effects are necessary both to integrate the findings on communicating risk information as well as to determine which types of graphical formats would be most effective in specific situations. The work of Stone et al. (1997) showed that for the particular modes of presentation they used, the choice of graphical format did not matter. However, it is important to understand precisely the mechanism underlying the graphical effect, so that any boundary conditions on this effect can be anticipated.

The goal of the present research was to gain a deeper understanding of the reasons behind the graphical effect identified by Stone et al. Consider again the task confronting the participants in Stone et al.’s study. For both the “Standard” and “Improved” products, they were given risk information in the form of a ratio, that x out of y would develop periodontal disease. Using the terminology introduced by Halpern et al. (1989), we refer to the number of people developing periodontal disease (x) as the “foreground,” and the number of people at risk (y) as the “background.” For example, for the toothpaste product used by Stone et al., the background was 5000 for both the Standard and Improved products, and the foreground was 30 for the Standard product and 15 for the Improved product. In this and the other scenarios used by Stone et al., the information in the foreground (30 vs. 15) suggests a fairly strong risk reduction, but the information in the background suggests that the risk reduction is not so strong (since the number of people at risk of harm is quite large, making the resultant incidence rates relatively small). To the extent that attention is drawn to one of these components more than to the other, then, it should be possible to make the risk reduction appear either large or small.

Thus, we suggest that the “graphical effect” discussed by Stone et al. was in fact a manifestation of a more general “foreground:background salience effect,” in that the graphical conditions used by Stone et al. were effective for increasing risk avoidance because they highlighted the number of people harmed at the expense of the number of people at risk of harm. More specifically, we are hypothesizing that in their numbers condition, participants reason based on relatively equal consideration of the foreground and background information. Given that the ratio of the foreground to the background is small, participants see the initial risk as being “low.” They thus perceive the improvement with the safer product as being relatively small, and offer to pay only a small amount more for the safer product. Fig. 2a illustrates the extreme case where the foreground and background receive perfectly equivalent consideration, though it is possible that in actuality the foreground and background do not receive exactly equal weight when participants are presented with a numerical display.

In the graphical conditions used by Stone et al., however, we propose that participants’ attention is drawn immediately to the foreground information. This hypothesis is based on the assumption that the graphical display of the foreground information is more salient than the numerical description of the background information, and that, in accord with the conclusions reached by Sanfrey and Hastie (1998), information that is particularly salient will have the strongest effect on any decision based on that display (see also Jarvenpaa, 1990). Thus, participants’ representations of the decision situation will be dominated by the foreground information, leading them to perceive the original risk as being “moderately large” and the reduction in risk as significant, which leads them to be willing to pay a moderate amount more for the safer product. Fig. 2b illustrates the extreme case where attention is devoted solely to the foreground.

To the extent that this mechanism is correct, it extends the work of Stone et al. in two ways. First, it suggests that the previously documented effectiveness of graphical displays occurs due to their effect on the perceived risk reduction. It is worth emphasizing that there are other potential mechanisms for the graphical effect that do not imply that the risk reduction is perceived as being greater when graphical versus numerical displays are used. As Stone et al. (1997) discussed, the extent of the risk reduction is only one of many considerations that would determine how much a person would be willing to pay for a safer product, and some of these considerations will favor low prices for the safer product, such as the equivalence of the two toothpastes on other factors and additional uses for which the money could be spent. It is plausible that, by calling attention to the risk reduction, graphical displays increase the importance associated with the risk reduction in relation to these other considerations. If that is the case, then participants should see the risk reduction as being more important with the graphical display, but not see the risk reduction itself as being any larger.

Second, our proposed foreground:background salience explanation suggests that the previously named graphical effect is not driven by a graphical mode of presentation per se, but instead by the particular aspect of the graphical formats used by Stone et al., whereby the foreground, but not the background, information was highlighted via the graphical mode of presentation. In other words, this explanation suggests that the greater effectiveness of graphical formats will not hold for all graphical modes of presentation. The logic behind the present research was to modify the graphical formats so that they did not highlight the foreground information to a greater extent than the background information, and determine whether or not the effectiveness of graphical formats in inducing risk avoidance remained after that change.

Section snippets

Experiment 1

The purpose of our first study was to test the previously discussed implications of our proposed foreground:background salience mechanism for the graphical effect documented by Stone et al. First, we needed to construct a graphical presentation mode that emphasized the background information to the same extent as the foreground information. One graphical format that meets these requirements is a pie chart. This mode of presentation does display the risk level graphically, but, unlike the

Experiment 2

The basic approach taken in Experiment 2 was the same as in the first study. However, to provide tighter control, the asterisks condition was replaced with a bar graph condition, similar to the one used in the Stone et al. study, and the pie charts condition was replaced by a stacked bar chart, which displayed both the foreground and the background information graphically (see Fig. 4). This change accomplished two goals. First, it allowed us to test the generalizability of the results found in

General discussion

The goal of the present research was to determine why the graphical techniques used by Stone et al. (1997) were effective at modifying risk-relevant behavior. Our experiments suggest the presence of a foreground:background salience effect, whereby the key factor is whether attention is called to the number of people at risk of harm (background information), or whether the focus is on the number of people harmed (foreground information). In support of this conclusion, two graphical formats not

Acknowledgements

Partial support for this work was provided by the University of Michigan Business School. Portions of this research were previously presented at the November 1998 and November 2001 annual meetings of the Judgment and Decision Making Society. We are grateful for the assistance of Hannah Faye Chua and Laith Alattar in carrying out the work described here, and to Andrew Parker and Isaac Lipkus for helpful discussions regarding the ideas incorporated in this paper.

References (32)

  • B. Fischhoff

    Risk perception and communication unplugged: Twenty years of process

    Risk Analysis

    (1995)
  • A. Fisher

    Risk communication challenges

    Risk Analysis

    (1991)
  • A. Fisher et al.

    Communicating risk under Title III of SARA: Strategies for explaining very small risks in a community context

    Journal of Air Pollution Control Association

    (1989)
  • L. Gordis

    Epidemiology

    (1996)
  • D.F. Halpern et al.

    Using statistical risk information to assess oral contraceptive safety

    Applied Cognitive Psychology

    (1989)
  • D.C. Howell

    Statistical methods for psychology

    (1992)
  • Cited by (142)

    • How Dispersive Opinions Affect Consumer Decisions: Endowment Effect Guides Attributional Inferences

      2021, Journal of Retailing
      Citation Excerpt :

      This link is referred to as improving monitoring accuracy by reducing overconfidence—a metacognitive bias that relies on intuition to process information, and can be detrimental to self-regulated judgement (e.g. Dunlosky and Rawson 2012). Therefore, the use of two different ways of cognitive processes can depends on the perceived fluency or disfluency in a given cognitive task (Stone et al. 2003). EWOM dispersion in previous studies has usually been depicted with a “horizontal bar chart”.

    View all citing articles on Scopus
    View full text