Talking frames : the assessment, use, and impact of message frames in conversational perspective

Talking frames : the assessment, use, and impact of message

frames in conversational perspective

Citation for published version (APA):

Buiten, van, M. (2007). Talking frames : the assessment, use, and impact of message frames in conversational perspective. Technische Universiteit Eindhoven. https://doi.org/10.6100/IR631570

DOI:

10.6100/IR631570

Document status and date: Published: 01/01/2007

Document Version:

Publisher’s PDF, also known as Version of Record (includes final page, issue and volume numbers)

Please check the document version of this publication:

• A submitted manuscript is the version of the article upon submission and before peer-review. There can be important differences between the submitted version and the official published version of record. People interested in the research are advised to contact the author for the final version of the publication, or visit the DOI to the publisher's website.

• The final author version and the galley proof are versions of the publication after peer review.

• The final published version features the final layout of the paper including the volume, issue and page numbers.

Link to publication

General rights

Copyright and moral rights for the publications made accessible in the public portal are retained by the authors and/or other copyright owners and it is a condition of accessing publications that users recognise and abide by the legal requirements associated with these rights. • Users may download and print one copy of any publication from the public portal for the purpose of private study or research. • You may not further distribute the material or use it for any profit-making activity or commercial gain

• You may freely distribute the URL identifying the publication in the public portal.

If the publication is distributed under the terms of Article 25fa of the Dutch Copyright Act, indicated by the “Taverne” license above, please follow below link for the End User Agreement:

www.tue.nl/taverne

Take down policy

If you believe that this document breaches copyright please contact us at: openaccess@tue.nl

providing details and we will investigate your claim.

T

F

: T

A

, U

,

I

Printing: Universiteitsdrukkerij Technische Universiteit Eindhoven

An electronic copy of this thesis is available from the site of the Eindhoven University Library in PDF format (http://www.tue.nl/bib).

© Marinus van Buiten, 2007

CIP-DATA LIBRARY TECHNISCHE UNIVERSITEIT EINDHOVEN Van Buiten, Marinus

Talking frames: the assessment, use, and impact of message frames in conversational per-spective/ by Marinus van Buiten. — Eindhoven: Technische Universiteit Eindhoven, 2007. — Proefschrift. —

ISBN 978-90-386-1177-8 NUR 778

T

F

: T

A

, U

,

I

M

F

C

P

PROEFSCHRIFT

ter verkrijging van de graad van doctor aan de Technische Universiteit Eindhoven, op gezag van de Rector Magnificus, prof.dr.ir. C.J. van Duijn, voor een

commissie aangewezen door het College voor Promoties in het openbaar te verdedigen op woensdag 19 december 2007 om 16.00 uur

door

Marinus van Buiten

prof.dr. G.B. Keren en

A

CKNOWLEDGEMENTS

This dissertation has greatly benefited from the contributions of numerous people. I would like to mention a few of them in particular.

First and foremost I thank my supervisors Gideon Keren and Karl Teigen. Over the past four years I had the great pleasure to work closely with both of them. Gideon, thank you for your unremitting creativity, dedication, guidance, and support. I have learned a lot. Karl, thank you for the many interesting insights and your helpful comments on my work. I thank Yaacov Schul, Chris Snijders, and Linda Moxey for taking the time to carefully read and reflect on my work. Your efforts are greatly appreciated.

I also thank my colleagues at the Human–Technology Interaction group at Eindhoven Uni-versity of Technology. In particular, I thank Stefan Gelissen, Iris van Rooij, Janneke Staaks, and Martijn Willemsen of the JDM research group. I have enjoyed the many discussions we had about science and various other topics often inspired by current events. A special thanks goes out to my fellow Ph.D. students Stefan and Janneke with whom I shared the joys and burdens of conducting experimental research.

Writing a dissertation is not without its setbacks. I owe many thanks to my good friend Arnold for lending an ear during the less enjoyable periods.

Finally, I thank my family. I am especially grateful to my parents and my brother Bas for their unconditional and invaluable support all along the way.

Thanks and all my love to my wife Tanja.

Marc

C

ONTENTS

Acknowledgements v

Contents ix

1 General Introduction 1

What’s in a frame? . . . 3

Research aim: framing in (conversational) perspective . . . 4

Thesis outline . . . 5

2 Speaker–Listener Incompatibility: Joint and Separate Processing in Risky Choice Framing 9 Speaker–Listener (In)compatibility . . . 11 Speaker–Listener Asymmetry . . . 12 Experiment 1 . . . 14 Experiment 2 . . . 15 Experiment 3 . . . 18 General Discussion . . . 20

Appendix Chapter 2: Instructions . . . 24

3 Speakers’ Choice of Frame in Binary Choice: Effects of Recommendation Mode and Option Attractiveness 27 Recommendation Mode: Encourage versus Discourage . . . 30

Experiment 1 . . . 31 Experiment 2 . . . 34 Experiment 3 . . . 36 Experiment 4 . . . 39 Experiment 5 . . . 41 General Discussion . . . 43

Appendix Chapter 3: Instructions . . . 46

4 The Impact of Message Framing on Breast Self–Examination: A Re–Examination 51 Experiment 1 . . . 54

Experiment 2 . . . 55 ix

Experiment 3 . . . 57

Experiment 4 . . . 58

Experiment 4a . . . 58

Experiment 4b . . . 59

General Discussion . . . 61

Appendix Chapter 4: Instructions . . . 66

5 A Framework for Framing Based on Joint–Separate Distinctions 71 Framing Typologies . . . 72

Reference–dependence and Evaluative Processes . . . 74

A Framework for Framing Based on Joint–Separate Distinctions . . . 75

Experiment 1 . . . 81

Experiment 2 . . . 82

Experiment 3 . . . 85

Experiment 4 . . . 87

General Discussion . . . 90

Appendix Chapter 5: Instructions . . . 92

Bibliography 97

Summary 105

Samenvatting 107

C

1

G

ENERAL

I

NTRODUCTION

The Dutch soccer legend Johan Cruijff has build quite a reputation with his remarkable, and according to some, brilliant comments on soccer. As a consequence, his comments have been often chronicled. A former Minister of Housing, Spatial Planning and the Environment even wrote a book in which he derives broader implications from Cruijff’s comments for e.g., leadership in organizations (Winsemius, 2004).

Another famous soccer expert, Louis van Gaal, is also known for his conspicuous remarks. Probably best–known is his cry–out in a moment of utter frustration and desperation about questions from journalists: “Am I that smart, or are you that stupid?”. Later, he cleverly cashed–in on the exposure that this outburst got, by appearing in a television commercial for an electronics shop. In this commercial he wondered whether this price–fighter is cheap, or whether its competitors are expensive.

Such a rephrasing from smart to stupid or from cheap to expensive may seem trivial. There is ample research, however, that says it is not. A research agenda introduced by Kahneman and Tversky, well over two decades ago, has spawned many interesting insights about the effects of, what is called, (message) framing on judgment and decision making.

In 2002, psychologist Daniel Kahneman won the Nobel Prize in Economics “for having in-tegrated insights from psychological research into economic science, especially concerning human judgment and decision–making under uncertainty” (The Sveriges Riksbank Prize in Economic Sciences in Memory of Alfred Nobel 2002). However, psychology and economics once had close interdisciplinary links but, due to various developments, started to drift apart well over a century ago. This separation has been fiercely defended by some, and at least as fiercely opposed by others ever since. Recent developments in economics, such as an increased appreciation of empirical and in particular experimental research, have restored some of its old ties to psychology. Increasingly, economists take notice of and participate in experimental research on (behavioral) decision making (e.g., Camerer & Loewenstein, 2003;

Kagel & Roth, 1995).

The last 50 years have witnessed converging experimental evidence that suggests choice behavior to be much more elusive than once thought. Traditionally, choice behavior has been conceived as the result of imposing a simple preference relation on the set of available choice alternatives (utility theory). More elaborate views in the same tradition, further distinguish between choice options’ constituent parts, namely the choice attributes. This view entails that an option’s value derives from assigning (subjective) values and weights (reflecting the associated time preference, importance, or risk) to the attributes of options, which are inte-grated by means of a combination rule (multi–attribute utility theory). Each decision (e.g., the choice between two used cars) is thus viewed as an assessment of the aspects of the choice options (e.g., mileage, condition) and the importance attached to these aspects.

It is said that many experimental findings on choice behavior violate two, mostly implicit, assumptions of the view explicated above, namely “procedure invariance” and “description invariance”. Procedure invariance entails an insensitivity of preferences to the particular elici-tation method that is employed. In this view, it should not matter whether preferences are measured in terms of direct choice, willingness–to–pay, or alternative ranking procedures. There is much research, however, that shows that our preferences are much more malleable than once thought and for one depends on the elicitation method (e.g., Lichtenstein & Slovic, 1971).

Arguably even more fundamental is the assumption of description invariance which re-quires preferences to be stable across alternate descriptions of choice alternatives and choice problems more generally. Violations of the latter postulate are collectively referred to as fra-ming effects. Frafra-ming effects pose a serious challenge to the traditional conception of rational choice behavior and have profound implications, as it suggests that the choice problem (op-tions, attributes, and so on) cannot be a theoretical primitive. They pose problems for most, if not all, state–of–the–art choice theories (e.g., Starmer, 2000), feed the discussions about the exact nature of preferences, i.e., whether they are revealed, constructed on–line, learned, or discovered (e.g., Fischhoff, 1991), and underscore the distinction between normative decision making (i.e., what people ought to do), and descriptive decision making (i.e., what people ac-tually do).

Many interesting examples of framing effects involve different yet logically equivalent de-scriptions of a situation. Consider the following two dede-scriptions of a choice problem (Tver-sky & Kahneman, 1986):

Description 1: Assume yourself richer by $300 than you are today. You have to choose between

a a sure gain of $100

WHAT’S IN A FRAME? 3

Description 2: Assume yourself richer by $500 than you are today. You have to choose between

a a sure loss of $100

b 50% chance to lose nothing and 50% chance to lose $200

Description 2 is merely a logically equivalent presentation of the choice problem in de-scription 1. Yet, if people are asked to make a decision based on the first dede-scription, most prefer the sure gain (alternative a), whereas under the second description most prefer the gamble (alternative b).

What’s in a frame?

Descriptions such as presented above are referred to as frames. There are many, mostly infor-mal, definitions of “frame” (e.g., Druckman, 2001b; Soman, 2004). All of them seem to capture the idea that a frame refers to a way of structuring a decision problem, or, more generally, situ-ations. The world around us is represented and interpreted through the use of frames. Just as a painting may be differently perceived in different picture–frames, so does the interpretation of a message depend on the linguistic frame employed to convey the message.

Structure is present in both the problem as it is presented to people, as in the mental model of the problem that people construct based on the stimulus information. The latter conception of “frame” is popular with many researchers. For example, Tversky and Kahne-man (1981) define decision frame as “the decision–maker’s conception of the acts, outcomes, and contingencies associated with a particular choice” (p. 453). More generally, Soman (2004) defines a frame as “a mental model of the decision problem that individuals use to solve the problem, and includes details about the elements of the decision problems (i.e., information) as well as context” (p. 380).

As noted, framing effects refer to the different behavioral effects supposedly caused by different frames. These framing effects encompass the trivial (e.g., the Dutch national rail-ways that announces trains arriving in x minutes rather than them being x minutes late), and the important (e.g., propaganda in warfare) and various attempts have been made to classify them (e.g., Druckman, 2001b; Levin, Schneider, & Gaeth, 1998; Soman, 2004). Most pertinent to the present thesis is the notion of valence framing (e.g., Levin et al., 1998). In valence fra-ming, the effects of positive (e.g., using words as “advantages” and “good”) and negative (e.g., using words as “disadvantages” and “bad”) frames, that are logically equivalent, are studied.

Levin et al. (1998) discerned three types of valence framing that play a recurring role in the present thesis. First, in risky choice framing, multiple choice options involving varying degrees of risk are described in terms of potential gains (e.g., probability to save lives) or losses (e.g., probability to lose lives). A relatively consistent finding is that people are risk seeking when choice option descriptions emphasize the chance to avoid losses, while they are risk averse when choice option descriptions emphasize the chance to realize gains (compare to the earlier example which asked you to imagine yourself richer than you are today with the subsequent possibility to gain or lose money for sure or probabilistically).

Second, in attribute framing some attribute of a choice option is described either posi-tively (e.g., in terms of a success rate) or negaposi-tively (e.g., in terms of a failure rate). People’s evaluations of options with positively framed attribute information are consistently more fa-vorable than those with negatively framed attribute information.

Finally, in goal framing actions or behaviors are linked to their consequences or implied goals. More specifically, goal frames link the presence or absence of a behavior to the at-tainment or non–atat-tainment of gains or losses. The persuasive impact of frames in terms of adoption rate of the desired behavior is measured and compared between different frames. Although empirical results are less consistent than in attribute framing, negative frames usu-ally turn out to be more persuasive than positive frames.

Research aim: framing in (conversational) perspective

The research on valence framing, and judgment and decision making more generally, has traditionally focused mainly on decision processes and behavior of decision makers. Most previous research provided information about choice alternatives without much reference to the social context, as if studying Robinson Crusoe. However, decision making, in particular concerning the more important decisions, typically does not take place in a social vacuum. This has been recognized (e.g., Payne, Bettman, & Johnson, 1993), but framing research has been somewhat slow to catch up and extend its focus.

In recent years, decision making research has embraced insights from research on the pragmatic implications of utterances made by speakers and directed at listeners (decision makers). Work on inferential pragmatics (e.g., Sperber & Wilson, 2005), offer thoughtful pro-positions about the role of intentions in communication. For example, Grice (1975) claims that messages from a speaker automatically raise expectations on the part of the listener that, in turn, determines which meaning the listener is likely to infer from it. These expectations are captured by the well–known cooperative principle (i.e., make your contribution such as is required, at the stage at which it occurs, by the accepted purpose or direction of the talk ex-change in which you are engaged), and four assumptions (maxims) which require the speaker to be informative, truthful, relevant, and clear. The interpretation that best satisfies these ex-pectations is the one a rational recipient should choose. Relevance theory (e.g., Wilson & Sperber, 2004) is a more concrete cognitive theory that builds on Gricean notions. In particu-lar, relevance theory specifies the determinants of relevance, and makes predictions that are experimentally testable.

Particularly relevant for the present thesis is the research by McKenzie and colleagues that draws from the literature on pragmatics. These investigators argue that framing effects should be understood by considering them from a conversational perspective. In particular, logi-cally equivalent frames may not necessarily be informationally equivalent (Sher & McKenzie, 2006). For example, describing a cup as “half full” or “half empty”, although logically equi-valent, leaks different information. A frame describing a cup as “half full” suggests that it was previously empty, whereas describing a cup as “half empty” suggests that it was previously full. This information leakage is used by speakers to convey information over and above the

THESIS OUTLINE 5

literal meaning of a message, that in turn is absorbed (i.e., inferred) by listeners to guide their behavior (McKenzie & Nelson, 2003; Sher & McKenzie, 2006).

Conceptually, communication can be thought of as the transfer of a message across some channel, from a source (e.g., speaker) to a recipient (e.g., listener), in a certain context: “who says what to whom (and with what effect?)” (e.g., Brock & Green, 2005; Kardes, 2002). A full understanding of communication requires an examination of all the five elements (i.e., source, recipient, message, channel, and context) involved. Following recent work on the conversational view of framing, the interest of the present research pertains to the interac-tion between speakers and listeners by means of alternative, logically equivalent, frames. The present research is confined to, what might be called, one–shot unidirectional communica-tion: speakers communicate information to listeners only once. There is no feedback flowing from listeners to speakers, nor is there opportunity for speakers to follow–up on their initial communication act.

The main focus is not on mutual understanding of speakers and listeners alone. Instead, the correspondence between speakers and listeners, implicit and explicit, is examined given that speakers have an underlying persuasion goal. The interest in the strategic use of frames derives from the observation that frames are selective representations that draw attention to different aspects of the presented information and thus suitable as strategic tools. The present research is aimed at investigating how speakers try to capitalize on the inherent potential of frames to be used strategically. Specifically, this thesis addresses (i) speakers’ assessment and selection of message frames, (ii) its corresponding persuasiveness as judged by listeners’ responses, and (iii) factors that influence (i) and (ii).

Thesis outline

Figure 1.1 presents a brief roadmap of the present thesis. The body of the thesis consists of four papers that address the role of framing in the communication between speakers and listeners in relation to three key factors, namely evaluation mode, option attractiveness, and recommendation mode. The three key factors are briefly discussed in turn below.

Evaluation mode

The importance of evaluating choice options in isolation (separate evaluation mode, or SE), or jointly with other choice options (joint evaluation mode, or JE) was probably first explicitly recognized by Hsee (e.g., Hsee, 1996; Hsee, Loewenstein, Blount, & Bazerman, 1999). For example, Hsee et al. (1999) discuss a seminal demonstration of a JE–SE preference reversal by Bazerman, Loewenstein, and White (1992). The case involved alternative resolutions to settle a dispute about a vacant lot between two neighbors’ houses. In particular, participants in the role of house–owner were asked to evaluate possible ways to split sales revenue from selling the vacant lot:

Option J: $600 for self and $800 for neighbor Option S: $500 for self and $500 for neighbor

In JE, participants evaluated the options jointly, and had to indicate which option they preferred. In SE, participants were exposed to the options one at a time, and were asked to provide a rating of the acceptability of each option. Remarkably, a large majority of partic-ipants preferred option J to option S in JE, whereas the majority rated option S to be more acceptable than option J in SE.

By extension, evaluation mode might potentially affect speakers’ assessment of alterna-tive (logically equivalent) frames. In particular, speakers are usually in JE mode (having to choose between several available formulations or frames), whereas listeners are usually in SE mode (exposed to the message as framed by the speaker). The influence of evaluation mode on speakers’ use of frames and decision makers’ choice behavior is primarily addressed in chapters 2, 4, and 5. Recommendation Mode (specifically chapter 3) Option Attractiveness (specifically chapter 3) Evaluation Mode (specifically chapters 2, 4, and 5) Assessment and Selection of Frames (Speakers) Persuasive Impact of Frames (Listeners) One-shot unidirectional persuasive communication

FIGURE1.1: Overview of the thesis

Option attractiveness

It is certainly not the case that negative terms are required to frame poor attribute values or unattractive options, nor are positive terms required to frame good attribute values and at-tractive options. To illustrate, consider Figure 1.2. Two choice options (A and B) are depicted in a two–dimensional space where each dimension refers to an attribute of the options. As-sume that higher attribute values are better. Suppose, for example, that options A and B re-fer to apartments where attribute 1 is apartment size and attribute 2 is the proximity of the apartment to local amenities. In this case, larger apartment size and greater proximity are

THESIS OUTLINE 7

preferred to smaller apartment size and smaller proximity. In addition to both options, two reference options (R1 and R2) are displayed. Suppose that R1 represents the default, salient option against which all other options are compared, i.e., the origin of the two–dimensional space. In this case, the valence of the attributes and choice options (A and B) as a whole is positive (+ Attr 1). That is, compared to R1, options A and B are very acceptable options with attribute values above that of R1. In contrast, moving from R1 to R2, changes the valence of attributes and options from positive to negative (- Attr 1). Compared to R2, both options are unattractive, because attribute values are below that of R2.

However, the valence of attributes and options is orthogonal to the valence of frames that can be used to convey (attribute) differences between options A and B. This is also depicted in Figure 1.2. A difference between the options on the first attribute may be framed either positively (+ Framea t t r 1) by anchoring on option A (e.g., “option A is better than option B on

attribute 1”), or negatively (- Framea t t r 1) by anchoring on option B (e.g., “option B is worse

than option B on attribute 1”).

R2

R1

A

- Attr1 - Frameattr1

“B is worse than A”

A tt ri b u te 2 Attribute 1 B R11 B1 A_{+ Attr1} 1 R21 + Frameattr1

“A is better than B”

FIGURE 1.2: Graphical illustration of valence of frames and valence of attributes. Higher attribute

values indicate better values.

To further illustrate the distinction between attribute and frame valence consider a physi-cian’s track record. Even the lousiest physiphysi-cian’s track record can be described in terms of his success–rate (positive) rather than the logically equivalent failure–rate (negative); conversely, the best physician’s history may be characterized in terms of his failures rather than his suc-cesses. The link between option attractiveness and speakers’ selection of frame is addressed in chapter 3.

Recommendation mode

Although “accepting a preferred alternative” and “rejecting a non–preferred alternative” are logically equivalent decision strategies in binary choice, decision–makers’ choice behavior is shown to be affected by this distinction. More generally, producing a shortlist of choice alternatives by eliminating prospects from an initial set of possible prospects is psychologi-cally distinct from generating the shortlist from scratch (e.g., Heller, Levin, & Goransson, 2002; Tversky, 1972; Yaniv, Schul, Raphaelli-Hirsch, & Maoz, 2002).

In the present thesis a closely related distinction is proposed between two general per-suasive orientations that speakers might adopt, i.e., “recommending for a preferred choice alternative” and “recommending against a non–preferred choice alternative”. This distinction is referred to as recommendation mode and it is primarily addressed in chapter 3.

In sum, this dissertation is “talking frames”. As subject of this thesis, framing will be dis-cussed at length. Specifically, framing will be addressed from a conversational viewpoint. It is concerned with speakers and listeners talking, and the role of message frames in these conversations. Moreover, figuratively speaking, frames “talk”. Frames are not neutral depic-tions of message content. Rather, different frames have different pragmatic implicadepic-tions, and different effects on choice behavior. As such, frames “speak” beyond the literal content of a message.

One of the main contributions of the present dissertation concerns the integration of two main themes in decision making, namely framing and evaluation mode, and tracing the im-plications thereof in a conversational perspective. Although some chapters focus more on this issue than others, it is an important main theme throughout.

C

2

S

PEAKER

–L

ISTENER

I

NCOMPATIBILITY

: J

OINT

AND

S

EPARATE

P

ROCESSING IN

R

ISKY

C

HOICE

F

RAMING

Abstract

Framing effects are considered in a conversational framework using the well–known Asian Disease problem (Tversky & Kahneman, 1981). Speakers’ preferred message framing is examined and its corresponding persuasiveness is assessed using listeners’ responses. The results show that speakers exhibit a marked and consistent preference for positive over negative framing (Experiment 1). Judged from listeners’ responses, this preference is effective for promoting riskless, but not risky options. The incompatibility between speakers and listeners may be resolved by noting that speakers can jointly (i.e., compar-atively) assess the information and the persuasive qualities of alternative frames. In con-trast, listeners are exposed only to one of these frames and, consequently, can only assess the information separately (i.e., non–comparatively). Experiments 2 and 3 demonstrate that no incompatibility exists when both speakers and listeners are either in separate, or in joint evaluation mode. Differences between risky choice and attribute framing (Levin et al., 1998) are briefly discussed.

Framing effects (e.g., Druckman, 2001b; Levin et al., 1998; Soman, 2004) may be viewed as dis-turbing examples of the volatility of preferences (e.g., Fischhoff, 1991; Lichtenstein & Slovic, 1971; Tversky & Simonson, 1993). This is especially true for what has been dubbed “equiva-lency framing” (Druckman, 2001b). Simply recasting information in different, yet logically equivalent, ways can dramatically affect preferences. These effects violate the normative prin-ciple of “description invariance” which states that preferences should be invariant to specific descriptions of the decision problem. The seminal Asian Disease problem (Tversky & Kahne-man, 1981) is a telling and robust case in point (e.g., Bless, Betsch, & Franzen, 1998; Druck-man, 2001a; Kühberger, 1998; but see Miller & Fagley, 1991). This framing effect demonstrates the dependence of people’s risk preferences on the exact framing of the choice alternatives. Confronted with an outbreak of the Asian disease that is expected to kill 600 persons, people (72%) predominantly choose a riskless program to combat the disease when both the riskless (program A below) and the risky program (program B below) are framed in terms of lives saved (i.e., positive frame):

Imagine that the U.S. is preparing for the outbreak of an unusual Asian disease, which is expected to kill 600 people. Two alternative programs to combat the disease have been proposed. Assume that the exact scientific estimate of the consequences of the programs are as follows:

If Program A is adopted, 200 people will be saved. If Program B is adopted, there is a one– third probability that 600 people will be saved and a two–thirds probability that no people will be saved.

Which of the two programs would you favor?

In contrast, when both programs are framed in terms of lives lost (i.e., negative frame), people (78%) tend to choose the risky program:

If Program C is adopted, 400 people will die. If Program D is adopted, there is a one–third probability that nobody will die and a two–thirds probability that 600 people will die.

The choice reversal, due to the framing manipulation, is usually explained in terms of the S–shaped value function of Prospect Theory (Kahneman & Tversky, 1979). According to Prospect Theory, people possess a diminishing marginal sensitivity to gains and losses. Peo-ple assess choice options along a concave value function in the domain of gains, and along a convex value function in the domain of losses. Generally, this leads them to be risk averse in the domain of gains (i.e., framing in terms of lives saved), and risk seeking in the domain of losses (i.e., framing in terms of lives lost)1_.

1_{Strictly, there is a four–fold pattern of risk attitudes (Tversky & Kahneman, 1992) which is supposedly due} to both (i) the S–shaped value function (characterized by diminishing sensitivity and loss aversion), as the (ii) probability weighting function (characterized by the overweighting of small probabilities and the underweight-ing of medium to large probabilities). As a result, risk aversion for gains and risk seekunderweight-ing for losses holds for medium to large probabilities. For small probabilities, people tend to be risk seeking for gains and risk averse for losses.

SPEAKER–LISTENER (IN)COMPATIBILITY 11

Speaker–Listener (In)compatibility

How can these seemingly incompatible choices be reconciled? Early research portrayed fra-ming effects mostly as violations of rational choice theory (e.g., Tversky & Kahneman, 1986). Later, however, alternative interpretations extended this narrow characterization of framing effects. For instance, work by McKenzie and colleagues (e.g., McKenzie & Nelson, 2003; Sher & McKenzie, 2006) stresses the importance of looking at framing effects from a conversational perspective. In particular, they argue that several framing phenomena, especially attribute framing effects (in which the characteristics of a single object or event are the focus of fra-ming; Levin et al., 1998) are in fact perfectly reasonable. Logically equivalent frames may not necessarily be informationally equivalent (Sher & McKenzie, 2006). Describing a cup as “half full” or “half empty”, although logically equivalent, leaks different information. A frame describing a cup as “half full” suggests that is was previously empty. Similarly, describing a cup as “half empty” suggests that it was previously full (McKenzie & Nelson, 2003; Sher & McKenzie, 2006). This information leakage is used by speakers to convey information over and above the literal meaning of a message, that in turn is absorbed (i.e., inferred) by listen-ers to guide their behavior (McKenzie, 2004; Sher & McKenzie, 2006). According to Sher and McKenzie, this research suggests that “information invariance” may be better suited as a nor-mative principle than “description invariance”. Further, logically equivalent, but information-ally non–equivalent frames may carry implicit recommendations (Sher & McKenzie, 2006; see also McKenzie, Liersch, & Finkelstein, 2006). Speakers and listeners can, and do, successfully capitalize on frames’ informational non–equivalence to implicitly transfer recommendations. Notwithstanding, the speaker–listener compatibility observed by McKenzie and Nelson (2003), and by Keren (2007), who exclusively employed attribute framing, may not generalize to all forms of equivalency framing. In the present paper the focus is on what has been termed risky choice framing as exemplified by the Asian Disease problem. The research is aimed at message framing by speakers who have an underlying goal of persuasion. We examine speak-ers’ perception of the persuasive impact of frames and the corresponding actual impact as measured by listeners’ responses. Consistent with earlier research, the present research is limited to what might be called one–shot unidirectional communication, i.e., speakers com-municate information to listeners only once. There is no feedback flowing from listeners to speakers, nor is there opportunity for speakers to follow–up on their initial communication act.

There are reasons to believe that in such situations, speakers will less successfully em-ploy frames than in the case of attribute framing. Our daily communication is imbued with positive wording of messages. Decision–makers, for example, predominantly describe choice options positively (Wang, 2004). This is just one manifestation of a more general “positivity bias” (e.g., Matlin & Stang, 1978; Peeters & Czapinski, 1990; Rozin & Royzman, 2001). Posi-tive words, posiPosi-tive experiences, and posiPosi-tive world views are all more frequent than their negative counterparts.

The positivity bias is also reflected in attribute framing research with positive and negative descriptions of attributes. For example, Keren (2007) used a scenario borrowed from Levin and Gaeth (1988), where participants in the role of speaker were asked if they would advise

a butcher to use a positive (i.e., “80% lean meat”) or a negative frame (i.e., “20% fat meat”) in order to maximize sales. Consistent with the positivity bias, most participants chose the positive frame.

Applied to the Asian Disease problem, the positivity bias implies that speakers will effec-tively influence decision–makers when promoting the riskless program, but will fail to do so when promoting the risky program. This prediction is based on the supposition that speak-ers’ responses reveal an a–priori positivity bias towards a cognitive organization having a fa-vorable content as much as possible (Peeters & Czapinski, 1990). On this account, positive frames that stress and signal the possibility of saving lives are preferred over equivalent ne-gative frames that stress the possibility to lose lives. Therefore, when speakers try to promote the risky program, they will (erroneously) presume the positive frame to be the most influen-tial frame when, in fact, listeners are more likely to choose the risky program when a negative frame is used. Pertaining to the Asian Disease problem in our speaker–listener framework, the following hypothesis is formulated:

HYPOTHESIS 1: Speakers tend to employ positive frames, regardless of the program being

promoted. This strategy is effective for promoting riskless, but not risky programs. Con-sequently, in case of risky programs, speakers’ conviction in the persuasive impact of posi-tive frames is incompatible with the actual impact as revealed by listeners’ choice behavior (role–incompatibility).

Speaker–Listener Asymmetry

An important distinction in decision making research is between joint and separate evalua-tion (e.g., Hsee, 1996). In joint evaluaevalua-tion mode, two or more opevalua-tions are presented simul-taneously, and can therefore be evaluated comparatively. In contrast, in separate evaluation mode, one option is presented and evaluated in isolation.

The joint–separate distinction has important implications. This is well demonstrated by the asymmetrical effect of evaluation mode on willingness–to–pay when choice alternatives have attributes that, by their nature, are hard to evaluate in separate mode (e.g., Hsee et al., 1999; Hsee, 1996). Difficult–to–evaluate attributes receive little weight in separate evaluation, yet their influence is elevated in joint evaluation in which one option serves as a reference point for the other. As a consequence, choices made in joint mode may not appear optimal when subsequently experienced in separate mode (Hsee & Zhang, 2004; Willemsen & Keren, 2004).

In a classic study, Hsee (1996) had participants state their willingness–to–pay for two mu-sic dictionaries. One dictionary was said to contain 20,000 entries and had a torn cover, the other supposedly contained 10,000 entries and looked entirely new. One group of participants stated their willingness–to–pay for both dictionaries (joint condition); two other groups eva-luated either one or the other dictionary (separate conditions). In joint evaluation mode, the dictionary high in number of entries, but with a torn cover, received a higher willingness–to–

SPEAKER–LISTENER ASYMMETRY 13

pay than the dictionary low in number of entries and no torn cover. In separate evaluation mode, however, the results were reversed.

By extension, evaluation mode may also have important consequences for speaker–liste-ner interaction, due to an evaluation mode asymmetry between speakers and listespeaker–liste-ners, or, more generally, between senders and receivers. Imagine, for example, gift giving and receiving (Teigen, Olsen, & Solas, 2005). Deciding on a gift, the giver may consider several alternatives, from which a gift is then selected. The receiver, however, is only exposed to the selected gift, and remains uninformed about the other alternatives.

Similarly, speakers have to decide which of the alternative frames to use, and are therefore in joint evaluation mode. Listeners, however, receive only one of the frames before making their decision and are thus in separate evaluation mode. This is in fact the conversational perspective originally studied by Tversky and Kahneman (1981).

We propose that removing the evaluation mode asymmetry will eliminate the incompati-bility conjectured in hypothesis 1. As argued, speakers in joint mode are inhibited in their strategic thinking by the positivity bias, resulting in the majority of speakers opting for the positive frame to exert influence (hypothesis 1). In contrast, when placed in listeners’ natu-ral mode, namely separate evaluation, speakers should be better able to form correct intu-itions about the impact of frames. It should be easier for speakers to imagine themselves in the listeners’ role (see for related work, e.g., Van Boven & Loewenstein, 2005). Consequently, they should be more likely to recognize the preference structure a frame yields for listeners. Specifically, in the positive domain, where programs are supposedly viewed as gains, program A seems to offer the larger gain and thus speakers promoting program A will rate the positive frame relatively favorably, while speakers promoting the seemingly smaller gain (i.e., program B) will rate the positive frame less favorably. In the negative domain, however, where pro-grams are supposedly viewed as losses, program B contains the seemingly smaller loss and thus speakers promoting program B will rate the negative frame relatively favorably, while speakers promoting the seemingly larger loss (i.e., program A) will rate the negative frames less favorably.

Further, we suggest that when listeners are placed in joint mode (i.e., the natural mode of speakers) and are asked to reflect on the persuasiveness of frames, they will, just as speakers in joint mode, have a bias towards the positive frame. When frames are evaluated side by side, the positive frame will appear more attractive and persuasive than the negative frame. This leads to the following hypothesis:

HYPOTHESIS2: Role–incompatibility disappears when controlling for the built–in

evalua-tion mode (joint vs. separate) asymmetry between speakers and listeners.

The two hypotheses were tested in three experiments. Hypothesis 1 was examined in Ex-periment 1. The second hypothesis was the subject of ExEx-periments 2 and 3. In order to gene-ralize our findings to other risky choice framing contexts, additional findings are presented in the general discussion.

Experiment 1

Method

Participants: One hundred thirty students from Fontys University of Professional Education in Eindhoven participated in the experiment. The experiment was part of a set of unrelated decision–making tasks that took about 20 minutes to complete. Participants received_€3 for their participation.

Design and procedure: This self–paced experiment was conducted on laptops. Participants were told that their task would require much attention and they were asked to read the in-structions carefully before answering. On the next screen participants received a scenario de-scribing the outbreak of an Asian Disease in a small village. This disease was expected to kill 600 inhabitants of a village. Two programs to combat the disease (with scientifically known consequences) were presented in both a positive (gain) as a negative (loss) frame. A note was included to point out the logical equivalence of both frames. Participants were placed in the role of speaker. Specifically, they were told that their task was to explain the programs to the city council and, to this end, had to select one of the two frames.

About half of the participants were instructed to assume a preference for the riskless pro-gram (propro-gram A), and the remainder was instructed to assume a preference for the risky program (program B). Participants were asked which frame they would use to convince the city council of their preference. To prevent mistakes, a box corresponding to the participant’s answer appeared around the frame.

Thus, the experiment consisted of a one–way between–subjects design with two condi-tions (promoting the riskless vs. the risky program). The dependent variable was the selec-tion of frame (positive vs. negative frame). The order in which frames were presented on the screen was counterbalanced. No order effects were found. The instructions of the experiment are presented in the appendix.

Results and discussion

The results of Experiment 1 are displayed in Table 2.1. Fifty–one out of 67 (76%) participants promoting the riskless program selected the positive frame; 53 out of 63 (84%) participants promoting the risky program selected the positive frame. Across conditions, and consistent with the positivity bias, the preference for positive framing expressed as the proportion of people framing positively (80%) was reliably above chance level (p < .001 as judged by a bi-nomial test). The difference in degree of positive framing between speakers promoting the riskless program and speakers promoting the risky program seems relatively small (e.g., Aber-son, 2002) and is not statistically reliable (C I0.95= .08 ± .14).

Thus, on the one hand, speakers’ perception of the persuasive impact of frames roughly matches the actual persuasive impact on decision–makers’ choices when the riskless pro-gram is being promoted (as is apparent from Tverksy and Kahneman’s 1981 results). On the other hand, speakers inaccurately anticipate the actual impact of the positive frame when the risky program is being promoted. Evidently, when the goal is to promote the risky program,

EXPERIMENT 2 15

the majority of people fail to select the negative frame which is the most persuasive frame. As-suming that our sample statistics and those of Tversky and Kahneman (1981) represent actual population parameters, the use of a loss frame would increase the hit–rate (i.e., the probabi-lity that the decision–maker would choose in accordance with the speakers’ suggestion) from approximately 36% (.84× .28 + .16 × .78) to approximately 78% (1 × .78)2_.

TABLE2.1: Number of participants (in the role of speaker) preferring the positive or negative frame as

a function of the program being promoted (riskless vs. risky) in Experiment 1.

Selected frame

Program being promoted Positive frame Negative frame Total

Riskless 51 16 67

Risky 53 10 63

Total 104 26 130

In the introduction a possible explanation for the discrepancy was posited. It was sug-gested that the identified incompatibility between speakers and listeners disappears when the evaluation–mode asymmetry of these conversational perspectives is removed. The fol-lowing two experiments were designed to test this prediction. In Experiment 2 the asym-metry is removed by changing speakers’ perspective from joint to separate evaluation mode, whereas in Experiment 3 it is removed by changing the listeners’ perspective from separate to joint evaluation mode.

Experiment 2

In order to select the most appropriate frame, a speaker (regardless of his goals) may want to go eyeball to eyeball with the listener, in order to penetrate the listener’s viewpoint. It may not be easy for speakers, placed in their natural joint mode, to simulate the listener perspective for a particular frame, because the other frame cannot be erased and may thus interfere. Placing the speaker in the separate mode, though by itself artificial (because speakers’ normal mode is joint), may facilitate simulating the listener’s grasp of a given frame. In other words, when forced into a separate mode, speakers should be able to better realize the effectiveness of a given frame in persuading the listener. We thus predict a higher level of congruency between speaker and listener when the speaker is placed in the separate mode, which is the listener’s default mode. Experiment 2 was designed to test this conjecture by placing participants in the speaker’s role yet exposing them to only one frame (separate mode).

2_{Strictly, there is also some room for improvement for speakers promoting the riskless program. Indeed,} if everyone, instead of 76% of the people, would select the gain frame, this would increase the hit–rate from approximately 60% to approximately 72%.

Method

Participants: One hundred thirty–three students from Radboud University of Nijmegen par-ticipated in the experiment. The experiment was part of a set of unrelated decision–making tasks that took about 25 minutes to complete. Participants received_€5 for their participation. Design and procedure: As in the previous experiment, participants were either required to promote the riskless or the risky program. In the present experiment, however, participants (initially) got to see only one of the two frames. The change from joint to separate evaluation mode required a change of the dependent variable. In Experiment 1, a simple dichotomous variable asking participants to select a frame was used. Here, a 6–point rating scale was used, ranging from “very inappropriate” (i.e., unconvincing) to “very appropriate” (i.e., convincing), to let participants rate the frame they got to evaluate in light of their goal of persuasion.

Subsequently, the remaining frame was shown. The frame that was rated before was shown again directly below it. The logical equivalence of the frames was pointed out and participants were asked to rate the remaining frame. In effect, participants progressed from separate (first rating task) to joint mode (second rating task).

Thus, the experiment had a 2 x 2 x 2 design with two between–subjects factors, i.e., pro-gram (promoting the riskless vs. the risky propro-gram), and order (positive vs. negative frame first), and one within–subjects factor, i.e., persuasiveness rating of the two frames. The order in which an option was presented within a frame was counterbalanced. No order effects were found.

Results and discussion

The results of Experiment 2 are portrayed in Table 2.2. The first rating task (which consti-tutes separate evaluation) is most important for our purposes. The average ratings of frames (positive and negative) by participants promoting the riskless program (M = 3.00,SD = 1.27) was higher than the average ratings given by participants promoting the risky program (M = 2.48,SD = 1.11, F (1,129) = 6.31,p = .013). More important, the main effect of program was qualified by a program by frame interaction (F(1,129) = 12.78,p < .001). Contrary to Ex-periment 1, speakers’ framing preference, as indicated by their rating, seems to be depen-dent on the program they promote. In fact, the pattern of results is consistent with the pat-tern of results reported by Tversky and Kahneman (1981) whose participants were placed in the listener’s role. Speakers promoting the riskless program judged the positive frame (M = 3.47,SD = 1.19) to be more persuasive (t (62) = 3.15,p = .002, one–sided) than the ne-gative frame (M = 2.53,SD = 1.19); speakers promoting the risky program judged the negative frame (M = 2.74,SD = 1.09) to be more persuasive (t (67) = 1.82,p = .037, one–sided) than the positive frame (M= 2.26,SD = 1.08).

The second rating refers to the rating of the remaining frame that was thus effectively made in joint evaluation mode. For example, the first row in Table 2.2 shows that participants promoting the riskless program gave a rating of 3.47 to the positive frame (separate mode), and subsequently gave a rating of 2.31 to the negative frame (joint mode). To understand the

EXPERIMENT 2 17

effect of the change of evaluation mode, difference scores (i.e., rating 2 minus rating 1) were calculated for each participant. Participants who first rated the negative frame and subse-quently the positive frame increased their rating on average (MD= 1.39,SD = 1.22), whereas

participants who first rated the positive frame and subsequently the negative frame slightly decreased their rating on average (MD = −.26,SD = 1.79). The difference between the two

was statistically reliable (F(1,129) = 47.17,p < .001). Evidently, the opportunity to compare frames leads, on average, to higher ratings if the remaining frame is positive than when the remaining frame is negative. These findings complement those of Experiment 1 that showed a preference for positive framing by speakers in joint mode. The analysis of the difference scores thus lends additional support to the notion of differential frame perception under joint and separate evaluation modes.

TABLE2.2: Mean persuasiveness ratings by participants (in the role of speaker) as a function of the

pro-gram being promoted (riskless vs. risky), and framing (positive vs. negative) in Experiment 2. Rating 1 (R1) refers to the rating of the frame shown first and constitutes “separate evaluation”. Rating 2 (R2) refers to the rating of the frame shown second and constitutes “joint evaluation”. Standard deviations are shown in parentheses.

Program being promoted Frame order n R1 R2 R2 minus R1

Riskless Positive frame first

(condition 1)

32 3.47 (1.19) 2.31 (1.20) -1.16 (1.59)

Negative frame first (condition 2)

32 2.53 (1.19) 4.16 (0.95) 1.62 (1.31)

Risky Positive frame first

(condition 3)

38 2.26 (1.08) 2.76 (1.30) 0.50 (1.61)

Negative frame first (condition 4)

31 2.74 (1.09) 3.87 (1.09) 1.13 (1.09)

The analysis of the difference scores further revealed that speakers promoting the risk-less program increased their ratings, on average, risk-less (MD= .23,SD = 2.01) than did speakers

promoting the risky program (MD= .78,SD = 1.42). This difference was statistically reliable

(F(1,129) = 5.46,p = .021). Both main effects were qualified by a statistically reliable pro-gram by frame order interaction (F(1,129) = 18.79,p < .001). Speakers promoting the riskless program substantially increased their ratings if the second rating task concerned the positive frame and substantially decreased their ratings if the second rating concerned the negative frame. Ratings of speakers promoting the risky program increased when the frame changed from negative to positive, but also increased slightly for second ratings regarding negative frames.

The first two experiments have shown that speakers’ perception of the persuasiveness of frames changes as evaluation mode changes. Positive frames are thought to be overall more persuasive than negative frames, whenever frames are considered jointly (Experiment 1 and second rating task of Experiment 2). If, however, frames are considered in isolation,

speak-ers promoting the risky program think negative frames to be more pspeak-ersuasive than positive frames, whereas speakers promoting the riskless program consider positive frames to be more persuasive. This pattern of results matches the pattern of results of listeners (Tversky & Kah-neman, 1981). Apparently, in separate evaluation mode, when speakers and listeners are in the same mode of reasoning, they agree about the persuasiveness of frames. Removing the natural conversational asymmetry in evaluation mode between conversational perspectives thus eliminates the incompatibility that was identified in Experiment 1.

In the present experiment, the evaluation mode asymmetry was removed by changing speakers’ evaluation mode from joint to separate. In the following experiment the asymmetry was removed by changing listeners’ evaluation mode from separate to joint.

Experiment 3

Eliciting listeners’ perception of the persuasive impact of frames is relatively simple and str-aightforward for listeners in separate evaluation mode. It amounts to measuring program preference for two groups, each of which receives only one frame. For listeners in joint eva-luation mode, however, matters are less clear–cut. The problem is how to disentangle the individual, and possibly combined, effects of both frames on final program preference. In the present experiment, participants were asked, after indicating their choice, to select which of the frames they believed to be the main determining factor of their choice. The answer served as a measure of perceived persuasive impact. This approach is adequate, under the assumption that choice proportions for both programs are approximately equal.

Method

Participants: One hundred forty–six students from the University of Tilburg, and one hun-dred thirty–five students from Radboud University of Nijmegen participated in the experi-ment. The experiment was part of a set of unrelated decision–making tasks that took about 25 minutes to complete. Participants received_€5 for their participation.

Design and procedure: Participants received the same scenario that was used in Experiment 1, except for small adjustments that were made to change it to a listener’s perspective. Specifi-cally, a supposedly reliable source provided information about two alternative ways to combat the Asian Disease. The source supposedly wanted to be as clear as possible about the conse-quences of both programs, and therefore presented the information in both a positive and a negative frame. As before, a note was included to point out the logical equivalence of both frames. Participants were asked to imagine that they were on the city council and responsible for the decision–making regarding the outbreak. Thus, they were placed in the listener role, and were asked to choose between the riskless and the risky program. Subsequently, they were asked to select the frame that they thought most influenced their decision. The order of frames and the order of options were counterbalanced.

EXPERIMENT 3 19

We also included a second condition in which the assessment of the programs and frames was reversed. Participants were first asked to select the frame that, in their view, most per-suasively promotes the riskless program and then the frame that most perper-suasively promotes the risky program. Finally, they had to choose their preferred program. Again, order of frames and options were counterbalanced, as was the order in which the persuasiveness of frames was judged.

Thus, the experiment had a one–way between–subjects design (program selection first vs. frame selection first). The dependent variables were the selection of program (riskless or risky program) and the selection of frame (positive vs. negative frame). None of the counterbalan-cing yielded order effects.

Results and discussion

The results are depicted in Table 2.3. We consider first the situation in which participants had to select a program first (panel A). The proportions of participants choosing the riskless program (42 out of 85, or 49%) and those choosing the risky program (43 out of 85, or 51%) are approximately equal. The fact that listeners can compare frames seems to cancel out the differences in program preference found with listeners in separate evaluation mode (Tversky & Kahneman, 1981). These results are compatible with those of Druckman (2001a) who ex-posed participants to a slightly different version of the Asian Disease problem also containing both frames.

Most important, collapsed across program preference, the vast majority of people (72 out of 85, or 85%) believed the positive frame to have influenced their choice most, whereas only a small minority (13 out of 85, or 15%) believed it to be the negative frame (p < 0.001 as judged by a binomial test). The proportion of people believing the positive frame to be most influential was 88% (37 out of 42) and 81% (35 out of 43) for those choosing the riskless and risky program respectively. The difference between these proportions was not statistically reliable (C I0.95= .07 ± .15).

The data from this and the first experiment were submitted to a log–linear analysis to as-sess the relation between the patterns of results of both experiments. The model fit with pro-gram selection, frame selection, and experiment as factors was compared to that of a model that did not include “experiment” as a factor. As expected, removing the “experiment” factor did not reliably reduce model fit (Breslow–Day test,χ2_{(1) = 1.87 with p = 0.17), providing}

sta-tistical evidence for compatibility of responses of speakers and listeners in joint evaluation mode. Placing both speakers and listeners in separate evaluation mode, also eliminated in-compatibility of responses (see Experiment 2). The more general conclusion is that removing the natural evaluation mode asymmetry between speakers and listeners, by placing both in the same mode, eliminates the incompatibility identified in Experiment 1.

The results for the condition in which programs were selected second are presented in panel B of Table 2.3. As can be seen, the response configurations in panel A and B are almost identical. The majority of listeners judged the positive frame to be more convincing of the two, for promoting the riskless (90%), and promoting the risky program (83%) respectively.

In fact, 75% of the participants selected the positive frame in both cases, whereas only 2% selected the negative frame in both cases (the remainder, 23%, selected each frame once).

TABLE2.3: Participants’ responses (in the role of listener) in Experiment 3. Panel A displays the number

of participants that judge the positive or negative frame to most persuasively affect their choice, as a function of the program being chosen (riskless vs. risky). Panel B displays the number of participants that judge the positive or negative frame to most persuasively promote both the riskless and risky program, as a function of the program being chosen.

Panel A: Program selection followed by frame selection–condition.

Frame deemed most persuasive

Chosen program Positive frame Negative frame Total

Riskless 37 5 42

Risky 35 8 43

Total 72 13 85

Panel B: Frame selection followed by program selection–condition.

Program

Riskless program Risky program

Frame deemed most persuasive Frame deemed most persuasive

Chosen program Positive frame Negative frame Total Chosen program Positive frame Negative frame Total Riskless 77 12 89 Riskless 74 15 89 Risky 99 8 107 Risky 89 18 107 Total 176 20 196 Total 163 33 196

General Discussion

When exchanging information in daily discourse, speakers and listeners are often in diffe-rent evaluation modes (e.g., Hsee, 1996). Speakers pass on information after considering and comparing different alternatives in terms of content and form. In contrast, listeners are only exposed to the single message transmitted by the speakers. Information frames that are as-sessed to be persuasive when compared with other alternatives may not seem equally persua-sive in isolation. This difference may have important implications in many real–life situations such as, for example, the communication between a doctor and patient, or between health organizations and patients more generally.

GENERAL DISCUSSION 21

The evaluation–mode–asymmetry can lead speakers to incorrectly assess the persuasive-ness of positive and negative frames (Experiment 1). Mutual understanding between speakers and listeners can be enhanced when the evaluation asymmetry is eliminated, either by chang-ing speakers’ perspective from joint to separate mode (Experiment 2), or by changchang-ing listen-ers’ perspective from separate to joint mode (Experiment 3). Specifically, speakers asked to promote the riskless program provided higher persuasiveness ratings when the positive frame was judged in isolation than when the negative frame was judged in isolation. Speakers that promoted the risky program, when placed in separate mode, rated the negative frame higher than the corresponding positive frame. Evidently, speakers’ perception of the persuasiveness of frames closely matches the actual impact on listeners’ choices, when both are in separate evaluation mode. In a similar vein, listeners exposed to both frames predominantly report thinking the positive frame to be most persuasive, independent of the program they chose. Evidently, there is also close correspondence between listeners and speakers, when both are in joint evaluation mode.

To test the robustness of our results, we replicated the reported experiments using a finan-cial context (borrowed from Roszkowski & Snelbecker, 1990) which is structurally identical to the Asian Disease problem. The scenario revolves around an investment in the stock mar-ket during a downturn in the economy and two strategies designed to preserve some of the investment. The same experimental procedures and similar student populations were em-ployed (see appendix for instructions).

The results were entirely consistent with the results of the Asian Disease experiments. As in Experiment 1, speakers in the joint mode, generally preferred to use the positive frame, irrespective of the strategy they were promoting3_{. A meta–analysis on the combined data of}

the Asian Disease and financial context shows that 75% of the participants selected the pos-itive frame when promoting the riskless program, and 89% selected the pospos-itive frame when promoting the risky program (82% across conditions; p < .001 as judged by a binomial test). When frames were evaluated separately (first rating in Experiment 2), persuasiveness ratings for the positive and negative frame depended on the strategy being promoted. Across the Asian Disease and financial context, the positive frame (M = 3.62,SD = 1.21) is judged more persuasive than the negative frame (M = 2.67,SD = 1.36) for promoting the riskless program (t(122) = 4.13,p < .001, one–sided), whereas the opposite (M = 2.49,SD = 1.23 vs. M = 2.92,SD = 1.14) holds for promoting the risky program (t (127) = 2.06,p = .02, one–sided). Finally, similar to Experiment 3, the majority of listeners exposed to the financial scenario judged the positive frame to be more persuasive than the negative frame when they could compare the two. This tendency was largely unaffected by strategy that was subsequently chosen. Across context, 79% of the participants that chose the riskless program thought the positive frame to be most persuasive, and 76% of those choosing the risky program found the positive frame to be most persuasive. Both reliably above chance level (p< .001 as judged by

3_{The present paper examines the framing behavior of speakers that try to exert persuasive influence through} a (default) strategy of recommendation. In binary choice, advising against the non–preferred option is a logically equivalent strategy. Interestingly, experiments that are yet unpublished show that this alternative strategy leads speakers to adopt a positive frame less often, sometimes even less than a negative frame. However, the results do not generally suggest a full preference reversal in frame selection.

a binomial test).

Mandel (2001) argued that the framing effect reported by Tversky and Kahneman (1981) may have been an artifact due to “asymmetries in the ambiguity of the sure and risky pros-pects and to the entanglement of two types of formulation manipulations: one having to do with the expected outcomes that are made explicit (positive vs. negative) and the other hav-ing to do with the descriptors used to convey the relevant expected outcomes (lives saved/not saved vs. lives lost/not lost)” (p. 56). Mandel’s analysis may be correct, yet it does not eradi-cate Kahneman and Tversky’s demonstration that people choose different programs depend-ing on how they are framed (Mandel’s main argument is that this demonstration is not neces-sary a violation of the requirement of description invariance). In our experiments we indeed employed the original formulation of Tversky and Kahneman. However, Mandel’s criticism, even if correct, does not apply to our experiments because our goal was not to test utility theory but rather the mutual comprehension and match between speakers and listeners.

Recent research (e.g., Keren, 2007; McKenzie & Nelson, 2003; Sher & McKenzie, 2006) has shown that speakers and listeners are able to achieve a high level of mutual understanding through the application and interpretation of frames that convey tacit information. In con-trast to these studies, the experiments described in this paper identified situations in which the correspondence between speakers and listeners is imperfect. The present research, how-ever, differs from that reported by McKenzie and Nelson, Sher and McKenzie, and Keren. Most important, the research of these investigators focused exclusively on attribute framing, whereas the present studies employed what is termed risky choice framing (Levin et al., 1998). Levin’s et al. framing typology, which does not consider conversational issues, distinguishes different kinds of valence (positive vs. negative) framing effects that supposedly tap different underlying processes. In risky choice framing, choice options involving varying degrees of risk are described in terms of potential gains (e.g., probability to save lives) or potential losses (e.g., probability to lose lives). Studies like the Asian Disease problem show, quite consistently, that people are risk seeking when the description of options draw attention to the chance of avoiding losses, and risk averse when descriptions focus on the chance to realize gains. In attribute framing, an attribute of a choice option is described either favorably (e.g., in terms of a success rate) or unfavorably (e.g., in terms of a failure rate). Positive framing consistently leads to more favorable evaluations than negative framing. A well–known example, referred to earlier, is that of Levin and Gaeth (1988)) who report that participants rated ground beef described in terms of its percentage lean meat as better tasting and less greasy than ground beef described in terms of its percentage fat meat.

The difference between attribute and risky choice framing in a conversational setting be-comes evident when considering the impact of evaluation mode on speakers’ and listeners’ responses. Supposedly, in attribute framing, compatibility is obtained whether or not evalu-ation modes of speakers and listeners are identical. For instance, using the above mentioned butcher scenario, Keren (2007) showed that speakers considering frames in joint mode, when instructed to maximize sales, tended to select the positive frame (“80% lean meat”). Listen-ers, in joint mode, instructed to choose between butchers described in either positive (“80% lean meat”) or negative (“20% fat meat”) terms, generally preferred the one who employed the positive frame. This compatibility is consistent with the information leakage account of Sher