MBA Level Term Paper

Journal of Personality and Social Psychology 1990, Vol. 58, No. 4,593-609

Copyright 1990 by the American Psychological Association, Inc. 0022-3514/90/S00.75

Multiple Audience Problem: A Strategic Communication Perspective on Social Perception

John H. Fleming University of Minnesota

James L. Hilton University of Michigan

John M. parley Princeton University

Brian A. Kojetin University of Minnesota

Examined how communicators send mixed messages containing an explicit surface content and a covert hidden content. In Study 1, Ss wrote constrained essays presenting either an introverted or extraverted personality. Although authors reported manipulating essay credibility and readers re- ported relying on credibility to make their judgments, readers succumbed to correspondence bias. In Studies 2 and 3, Ss again prepared either constrained essays (Study 2) or constrained videotapes (Study 3) and included in them a hidden message that would be understood by only their friends but not by strangers. Observers then read these essays or watched these videotapes. Friends detected and decoded the hidden messages, whereas strangers did not. We discuss these findings in terms of social perception and strategic communication.

A large proportion of everyday social behavior is strategic: People tell lies, feign emotions, ingratiate powerful others, and take stands with which they do not agree. That is, people fre- quently attempt to manage the impressions and the information they present to others in order to further their interpersonal goals. Although people's strategic behavior may often accu- rately reflect their underlying personalities, beliefs, or values, people sometimes behave in ways that do not accurately reflect who they are or what they believe.

This behavior may result from a commitment on the part of the person to mislead the audience; the person intends to de- ceive. The medicine-show man, for example, seeks to convince his audience that the elixir he hawks is of great value. Similarly, the greedy gold digger seeks to convince a wealthy widower that her love is genuine. In contrast, at other times this behavior flows from constraints inherent in the situation in which people find themselves. A defense attorney for an accused pornographer, for example, must play the role of advocate for the client regard-

Portions of this research were presented at the meetings of the Social Psychology Winter Conference, Park City, Utah, January 9-15, 1988, and at the Midwestern Psychological Association, Chicago, Illinois, May 4-6, 1989.

This research was supported in part by a grant-in-aid from the Grad- uate School of the University of Minnesota to John H. Fleming, by a grant from the John D. and Catherine T. MacArthur Foundation to John M. Darley, and by a National Institute of Health Training grant (1-T32-MH18021-03).

We gratefully acknowledge the assistance of Marti Hope Gonzales, Debbie Prentice, Kelly Shaver, and Mark Zanna for their helpful com- ments on drafts of the manuscript. Additionally, we thank Linda Ginzel for her help with the procedure in Study 2, Kay Zenk for her assistance with the data collection in Study 2, and Katie Fridgen for her assistance with data collection in Study 3.

Correspondence concerning this article should be addressed to John H. Fleming, Department of Psychology, University of Minnesota, El- liott Hall, 75 East River Road, Minneapolis, Minnesota 55455.

less of the attorney's personal feelings toward pornography; de- baters who are constrained by a flip of a coin to support a par- ticular side of an issue do not necessarily hold the views they endorse during the debate. But how good are people at detecting misrepresentations that arise either from active attempts to de- ceive the audience or from the constraints inherent in the situa- tion? At least two literatures in social psychology address these issues. Both reach similar conclusions.

In the past 20 years, over 700 articles have been devoted to understanding the antecedents and consequences of intention- ally misleading deceptions (Hyman, 1989; see DePaulo, Stone, & Lassiter, 1985, DePaulo, Zuckerman, & Rosenthal, 1980a, 1980b,Ekman, 1985,Knapp&Comadena, 1979, Kraut, 1980, G. R. Miller & Burgoon, 1981, and Zuckerman, DePaulo, & Rosenthal, 1981, 1986 for reviews). These lie detection studies encompass two broad categories: (a) those that address aspects of actors' behaviors that differentiate lying from truthful behav- ior; and (b) those that examine the processes by which observers determine, first, whether a person is lying or telling the truth and, second, what the concealed truth is in a deceptive perfor- mance. The typical finding is that despite the fact that there are documented differences between truthful and deceptive behav- iors (for reviews, see Ekman, 1985; Krauss, 1981), people are not very good at perceiving those behaviors and are not very good at detecting lies. By and large, observers take a person's deceptive behavior at face value and detect deception at levels only slightly better than chance. This tendency to take behavior at face value is known in the lie detection literature as the truth- fulness bias (Zuckerman, Koestner, Colella, & Alton, 1984).

A similar tendency has emerged in the literature on the dispo- sitional attributions that are made when situational constraint is high. One of the major discoveries to emerge from the attribu- tion literature has been variously labeled the fundamental attri- bution error (Ross, 1977), observer bias (Jones & Nisbett, 1972), overattribution (Quattrone, 1982), or correspondence bias (Gilbert & Jones, 1986). All of these terms refer to a persis- tent tendency among observers to underestimate the effects of

593

594 FLEMING, DARLEY, HILTON, AND KOJETIN

situational constraints on behavior, and to attribute causes to a person's internal dispositions instead. Observers who have been asked to read pro- and anti-Castro essays, for example, have assumed that the authors endorse the positions advocated in the essays even when they have been told that the direction of the essays was assigned by the experimenter (Jones & Harris, 1967). Although originally unexpected, this finding has proven both robust (e.g., Jones, Worchel, Goethals, & Grumet, 1971) and generalizable across variations (A. G. Miller, Jones, & Hinkle, 1981; Schneider & Miller, 1975). It has also survived sugges- tions that it was produced by experimental artifact (Kelley, 1972; Snyder & Jones, 1974), and by now numerous studies have found evidence documenting the existence of correspon- dence bias (for reviews, see Fiske & Taylor, 1984, Jones, 1979, 1986, Nisbett& Ross, 1980, and Ross, 1977).

The similarities in the findings from the lie detection litera- ture and the attribution literature are not accidental. Both dem- onstrate the same fundamental tendency for observers to as- sume that people's behavior accurately reflects their attitudes or dispositions. It is possible, however, that these studies may underestimate communicators' abilities to send complex mes- sages and observers' abilities to interpret those messages, be- cause these studies have relied on a limited class of situations in which a person generates complex communications. What happens, for example, when one sees a captive American decry American imperialism and terrorism on a videotape made by the captors (ABC News, 1986, 1987)? Or, more mundanely, what happens when one sees teenagers who, in the presence of their peers, are forced by their parents to promise to walk the straight and narrow? In other words, what happens when actors must be (realistically) concerned that their behavior, con- strained by the presence of one audience, will result in unfortu- nate or undesirable judgments being made about them by an- other audience (e.g., the American public who sees the hostage videotape, or the teenager's friends)? We refer to these situations as multiple audience problems, because actors face the task of simultaneously sending two different messages to two different audiences receiving the same communication (also see Fleming &Darley, 1989).

Our goal in this research was to begin examining the strate- gies that communicators and observers adopt when they are confronted with these kinds of situations. We have attempted to address questions such as: What (if anything) do constrained actors attempt to communicate about their true personalities or opinions, and how do they attempt to communicate it when they are constrained to behave in personally discrepant ways? What attributes of the target audience (or audiences) determine the types of communications they will use? What strategies or cues do observers use to interpret communicators' messages and infer actors' dispositions, and how do these strategies paral- lel communicators' strategies for conveying the underlying truth in their constrained behavior? The answers to these ques- tions may not only reveal more about the nature of lie detection and correspondence bias but may also provide a foundation for a more general understanding of strategic interpersonal com- munication.

We designed three different studies to examine from a com- munication perspective how constrained actors attempt to send cues in messages prepared under conditions of high constraint to signal the actual nature of the message to observers. Likewise,

we examined the responses and reactions of the observers to see whether they detected those cues or focused on other aspects of the communications instead. Study 1 followed the typical per- sonality attribution/assigned essay paradigm used in many cor- respondence bias studies. In those studies, authors are assigned a position to present that may or may not contradict their own opinions or personalities. Observers are then given the essays to read with full knowledge that the direction of the essays was determined by the experimenter. The observers' task is then to determine the true attitudes or personalities of the authors. (For a more complete review of studies using this paradigm, see Jones, 1979.) With a few additions to be described later, we fol- lowed this general procedure. We also asked the essay authors to do something not normally done in a social perception study: We asked them to write a second essay, under the same con- straints as the first, with additional instructions to attempt to signal their own true personalities while keeping the surface content of the essay the same.

STUDY 1

Phase 1: Generating the Essays

Method

The 32 subjects contacted by telephone who served as essay authors were paid volunteers enrolled at Princeton University. After consenting to participate in a communication experiment, they heard descriptions of an introvert and an extravert containing adjectives that have been found to have high prototypicality ratings for introversion and extraver- sion (Cantor & Mischel, 1977). Authors then rated themselves on a 9- point introversion-extraversion scale. Equal numbers of subjects who rated themselves on either side of the midpoint of the scale agreed to come into the laboratory for an experiment lasting 45-60 min.

Standard Essays

We asked the authors to write an essay designed to convince another person that they were either an extreme introvert or an extreme extra- vert. Half of the authors were randomly assigned to present a trait corre- sponding to their self-ratings, and half were randomly assigned to pre- sent a trait opposite of their self-ratings. The instructions used in this experiment were previously used by A. G. Miller et al. (1981, p. 590). The experimenter pointed out to the authors that the set of instructions they each received (e.g., "Write about being introverted") was randomly determined and that the only restriction in writing the essays was that the words extravert and introvert could not be used. Finally, they were informed that the essays should be approximately one page long and were given 15 min to complete the task. At this point, if there were no questions, the experimenter left.

Communicate Essays

The experimenter returned 15 min later, collected the first essays, and then presented the written instructions for the second essay. These instructions initially made salient the fact that the instructional set for the first essay had been determined by chance and continued as follows:

What we want you to do now is to rewrite your first essay, only this time we want you to try to get across what you really are. Notice, however, that if you first wrote presenting yourself as an introvert you must still write presenting yourself as an introvert. Similarly, if you first wrote presenting yourself as an extravert you must still write as an extravert. At an overt level the essay topic must remain

MULTIPLE AUDIENCE PROBLEM 595

the same. The only difference is that this time we want you to try to get your true self across.

The authors were told that they could use as much or as little of their first essay as they could remember. They were then told that there were three constraints on the essay. First, the essay had to be similar in form to the first essay. Second, the words introvert and extrovert again could not be used. Third, phrases such as "I'm writing this essay because I have to" were not to be included. Again, they were given 15 min to complete the essay. After the experimenter collected the second essays, authors were asked to complete a two-page questionnaire that contained the dependent measures.1

Dependent Measures

We first asked the authors whether they would describe themselves as introverts or extraverts. (There were no shifts from previous self-rat- ings.) We then asked them which of these traits they had been asked to portray in the two essays; all accurately recalled their assigned traits. Next, we asked them to rate both of their essays on two 9-point scales, rating the extremity of their presentations of the trait first and the credi- bility of their presentations second. On a final 9-point scale, authors rated how concerned they were with communicating their true person- ality traits in each essay.2 When they completed the dependent mea- sures, authors were debriefed, thanked for their participation, and ex- cused.

Results

Manipulation Check: Ratings of Concern

To examine the authors' motivations to transmit their true personalities, a 2 (essay assignment: portray introvert vs. por- tray extravert) X 2 (actual trait: introvert vs. extravert) X 2 (es- say: first vs. second) mixed-model analysis of variance (ANOVA) with repeated measures on the last factor was performed on the authors' reports of concern for presenting their true personali- ties. Consistent with the instructions they received, a significant essay main effect emerged; authors expressed greater concern for presenting their true personalities in the second essay (M = 7.54) than in the first essay (M = 3.67), F(l, 20) = 71.04, p < .001. A significant Essay Assignment X Actual Trait interaction also emerged; authors who were assigned positions correspond- ing to their true personalities were more concerned about pre- senting their true personalities than were authors assigned to present a noncorresponding trait, F(l, 20) = 27.28, p < .001. The interpretations of both of these effects are qualified, how- ever, by the presence of a significant Essay Assignment X Actual Trait X Essay interaction indicating that these differences are confined primarily to the first essay, F( 1, 20) = 13.15, p < .01. As the top pair of graphs in Figure 1 indicates, the instructions for the second essay greatly increased the reported motivation of all subjects to convey their true personalities (and especially those who presented a false image of their personalities).

Ratings of Credibility

The credibility ratings provided some insight into how the authors attempted to convey their true personalities. A signifi- cant essay main effect emerged; authors rated their second essay as less credible (M = 6.12) than their first essay (M = 6.89), F( 1, 28) = 5.73, p < .03. A significant Essay Assignment X Actual Trait interaction also emerged; authors who wrote noncorre-

sponding essays rated them as less credible than authors who presented a corresponding trait, F(l, 28) = 23.75, p < .001. Once again, however, the interpretations of these effects are qualified by the presence of a significant Essay Assignment X Actual Trait X Essay interaction indicating that the credibility differences that emerged among authors who wrote noncorre- sponding essays were limited primarily to the second essay, F( 1, 28) = 16.45, p < .001 (see the middle pair of graphs in Fig- ure 1).

Ratings of Extremity

Analysis of the authors' perceptions of the extremity of their presentations of the assigned trait revealed that they rated their first essay as more extreme (M = 7.17) than their second essay (M = 6.42), F(l, 28) = 6.59, p < .02. A marginally significant effect for essay assignment emerged; authors who presented themselves as extraverts tended to rate their essays as more ex- treme (M = 7.19) than did authors who presented themselves as introverts (M = 6.40), F(\, 28) = 3.54, p < .08. A marginally significant Essay Assignment X Actual Trait interaction also emerged; authors who presented a corresponding personality tended to rate their essays as more extreme (Ms = 6.78 and 7.62 for introverts and extraverts, respectively) than did authors who presented a noncorresponding personality (Ms = 6.03 and 6.75 for introverts and extraverts, respectively), F(l,28) = 3.83, p < .07. Importantly, the three-way interaction was not significant in this analysis; the patterns of extremity ratings did not differ between the first and second essays (see the bottom pair of graphs in Figure 1).

Together, these results suggest that authors in the typical as- signed essay paradigm are not particularly motivated to com- municate their true personalities when assigned to describe their personalities in a noncorresponding way. When they were motivated to communicate their true personalities, however, they attempted to do so and reported greater concern with com- municating an accurate self-image. Apparently, as their de- briefing comments suggest, authors attempted to manipulate the credibility of their essays to signal their true personalities. Our next question naturally follows: If authors try to signal their true personalities by manipulating the credibility of their essays, do the readers of the essays notice those cues, and if so, how do those cues affect readers' judgments of the authors' personali- ties? We addressed this question in Phase 2.

Phase 2: Readers' Judgments

Method

The 16 subjects who served as essay readers were paid volunteers en- rolled at Princeton University and were contacted by telephone to par-

1 We gave the questionnaire measures about both essays after the sec- ond essay, because we did not want to run the risk of alerting subjects to dimensions useful in writing their second essays. This means, how- ever, that their ratings of their first essays were made with some possibil- ity of comparison with their second essays, and vice versa.

2 Because of experimenter oversight, 8 of the 32 subjects did not re- ceive this question, because we initially failed to recognize its impor- tance. The 24 subjects who did receive the question were distributed evenly across the four conditions.

596 FLEMING, DARLEY, HILTON, AND K.OJETIN

01 UJ 0.

cc a

tu >-

2£ IE

3 OC < o

9

8

7

6 '

5 -

4 -

3 '

2 -

1 -

9 -

8 -

7 -

6 '

5 -

4 -

3 -

2 -

1 -

9 1

a

7 •

INTROVERTED ESSAY EXTRA VERTED ESSAY

ACTUAL INTROVERT

ACTUAL EXTRAVERT

ACTUAL INTROVERT

ACTUAL EXTRAVERT

FIRST ESSAY SECOND ESSAY

Figure 1. Authors' perceptions of essays written under the two sets of instructions in Study 1.

ticipate in a communication experiment. All subjects were randomly assigned to conditions. On arriving readers read instructions that stated that in a previous study the author-subjects had written essays designed to convince another person that they were either an extreme introvert or an extreme extravert. Readers then read the same descriptions of an introvert and an extravert that the authors had seen and learned that the authors had been instructed not to use the words introvert and extravert in their essays. We emphasized that the instructional set each author received had been determined by chance. We told the readers that they would read 16 essays that had all been written under the assigned posi- tion instructions.3 They then read and rated the 16 essays; half of the subjects read essays under the assigned position instructions, and half read essays written under the instructions directing authors to convey their true personalities.

After reading each essay, the readers rated the author's true personal- ity trait on a 19-point scale with endpoints labeled extremely introverted (1) and extremely extroverted (19).4 They then indicated to which of the four experimental conditions the author of the essay had been assigned and, additionally, which personality trait the author had presented.

3 Recall that we had 32 essays written under the standard set of in- structions and 32 essays written under the communicate set of instruc- tions. In order to keep the observers' task within a manageable time framework, the essays written under each set of instructions were di- vided into two sets of 16 essays. Thus, each of the four sets of four ob- servers read one of the sets of 16 essays. Two sets of observers read essays written under the standard set of instructions, and two sets of observers read essays written under the communicate set of instructions.

4 Authors were first asked to indicate whether they were an introvert or an extravert and then were asked to indicate the extremity of their possession of the trait on a 9-point scale. Similarly, they were asked whether the essay they wrote was in the direction of introversion or ex- traversion and then were asked to rate the extremity of their presenta- tion of the trait. For observers and paired observers the 19-point scale incorporated both direction and extremity by ranging from extremely introverted (1) to extremely extroverted (19). Reported correlations in- volving extremity ratings were calculated separately for introverted and extraverted essays using recoded extremity scores that eliminated the direction of the rating.

MULTIPLE AUDIENCE PROBLEM 597

Results

We used readers' estimations of the true personality of each author (ratings of inferred personality) and readers' ratings of the extremity and credibility of each essay to examine three related questions. First, what information did readers use to in- fer the authors' personalities? More specifically, what was the relation between judged extremity and credibility of the essays and the resulting judgments of the authors' true personalities? Second, did our readers succumb to correspondence bias, or in the case of the second essays, did they instead detect the changes the authors reported making? Third, what were the relations between the authors' and readers' judgments of the authors' es- says?

Informational Bases of Readers'Judgments

As might be expected, readers' inferences about the essay au- thors' true personalities were strongly related to readers' judg- ments of the credibility of the essays and less strongly to the readers' judgments of the extremity of the essays; the average correlations between these variables were both significant— r(62) = .63, p < .001, and r(62) = .24, p = .05, for credibility and extremity, respectively.5 Credibility and extremity, however, were not correlated (r(62) = —.07). Because the correlation pat- tern did not differ significantly for readers of the first and second essays, these results were pooled across the two essays.

To determine the relative contribution of these variables to the readers'judgments of the authors' true personalities, we per- formed multiple regression analyses on the data from each reader. The averaged multiple regression equation predicting inferred personality from the credibility and extremity of the essays resulted in a correlation of .70 and accounted for 53% of the variance. The average beta weight for credibility in the predictor equations was generally high (.61) and was statistically reliable for all but 5 of the 16 readers. The average beta weight for extremity was significantly lower (.25); for 4 readers, how- ever, it was a statistically reliable predictor of judgments of the author's true personality. These results suggest, that most read- ers' personality judgments were related to the authors' estimates of the essay's credibility, although a few readers' judgments were also related to authors' estimates of extremity over and above credibility (also see Fleming & Kraut, 1988).

Accuracy of Readers'Judgments and Correspondence Bias

As we mentioned earlier, during debriefing, authors reported attempting to manipulate essay credibility to convey their true personalities in the second essay, but not in the first. Further- more, readers reported basing their judgments of the authors' personalities on their judgments of essay credibility. But it re- mains to assess the actual success of their joint efforts. This can be determined by examining the relation between the readers' ratings and the authors' self-reported personality traits for the two essays. The average correlations between authors' and read- ers' ratings of the essays are presented in columns 2 and 3 in the upper panel of Table 1, and as the table shows, these corre- lations for essays written under both sets of instructions were quite low; readers generally misjudged the authors' true person-

alities, even when authors tried to convey their true personali- ties.

If readers' judgments were not related to the authors' true personalities, then to what were they related? An examination of the average correlations between essay personality assign- ment (i.e., write presenting yourself as an introvert vs. extra- vert) and the readers' ratings of inferred personality under both sets of instructions provides an answer (see column 4 in the upper panel of Table 1). As Table 1 shows, readers tended to ascribe personality in line with the direction of essay assign- ment for both essays. In other words, readers tended to fall prey to correspondence bias even when they were reading essays in which authors attempted to convey their true personalities.

Relations Between Authors' and Readers' Judgments

To understand why readers apparently missed the cues au- thors reported placing in their essays (despite the fact that read- ers reported using them), we analyzed the correlational re- lations between readers' and authors' ratings of essay extremity, essay credibility, and author personality type. These corre- lations are presented in the left-hand portion of Table 2. As Ta- ble 2 shows, for essays written under both sets of instructions, the noise in the earlier analyses is evident here in the low average correlations; authors and readers did not seem to agree on what constitutes credibility and, to a lesser extent, extremity.

These results suggest that although authors and readers agreed that essay credibility and extremity were valid cues to signal authors' true personalities, the two groups seemed to dis- agree on exactly what constitutes those cues. As a result, readers did not detect the authors' credibility and extremity cues, and made correspondent judgments of the authors' personalities. Our next question naturally follows: If readers in the standard assigned essay paradigm (first essays) did not accurately detect the authors' cues signaling their true personalities, under what conditions could readers be made to do so? We addressed this question in Phase 3.

Phase 3: Paired Readers' Judgments

Method

Four additional subjects or paired readers read sets of the first and second essays written by the same author.6 These paired readers were paid undergraduate volunteers enrolled at Princeton University and re- ceived the same information as the readers in Phase 2. In addition,

5 The average correlations reported were computed by performing a Fisher Z' transformation on the r for each subject, averaging the Z' across subjects, and then transforming the resultant Z' to r. Following Cohen and Cohen (1983), the probability associated with each averaged correlation was determined by first performing Fisher's r to Z' transfor- mation for each subject and then transforming each Z' to Z using a hypothetical population value of r = 0. We then treated each Z score as an independent test of the hypothesis and combined them as suggested by Rosenthal (1978). Specifically, we summed the Zs and divided by the square root of JV. The associated probability of the combined Zs is reported.

6 In line with the logic of Footnote 3, the 32 pairs of essays were di- vided into two sets of 16 pairs. Thus, two paired observers read one set of 16 pairs and, the remaining two paired observers read the second set of 16 pairs.

598 FLEMING, DARLEY, HILTON, AND KOJETIN

Table 1 Average Correlations Between Readers' Ratings of Authors' Personalities and Selected Variables From Study 1

Reader ratings of authors' true personalities

Readers First essays Second essays

Paired readers Both essays

Authors' self-reported personality trait3

.07

.08

.20

Authors' ratings of actual personality11

.15

.14

.22

Authors' assigned personality0

.55**

.30*

.53**

Note. N = 32 ratings for readers, 64 ratings for paired readers. * Correlations reported are point-biserial correlations calculated by grouping authors as introverts or extra- verts on the basis of their self-reports of the trait they possessed. b Correlations were calculated using authors' self-reports of introversion-extraversion on 9-point scales. c Correlations reported are point-bise- rial correlations calculated using the trait that authors were instructed to present. *p<.0l. **/x.001.

paired readers were told of the existence of the essays written under the second instructional set and were told these instructions. We empha- sized that each pair of essays was written by the same author but that the order of presentation within pairs had been randomized. After read- ing each pair of essays, paired readers completed the same dependent measures as those previously completed by the original readers with appropriate changes for the task of rating pairs of essays.

Results

For paired readers we addressed the same three questions we asked of the original readers. First, what information did paired readers use to infer authors' personalities? Second, were paired readers able to detect the changes the authors reported making in their second essays, or did they again succumb to the corre- spondence bias? Third, what were the relations between au- thors' and paired readers' judgments of the essays?

Informational Bases of Paired Readers' Judgments

The average correlations between paired readers' ratings of inferred personality and essay credibility for essays written un- der both sets of instructions were significant; average—rs(30) = .46 and .63 for the first and second essays, respectively; both ps < .01. The correlations between essay extremity and essay credibility failed to reach significance, as did correlations be-

tween inferred personality and essay extremity. Paired readers' inferences of the authors' true personalities, like those of the original readers, were strongly related to paired readers' judg- ments of essay credibility. Once again, this pattern does not ap- pear to differ for essays written under the two instructional sets, and there is no clear relations between judgments of essay credi- bility and extremity for either set of essays.

Accuracy of Paired Readers' Judgments and Correspondence Bias

Apparently, the strategy that paired readers adopted to de- code the essays was quite similar to the strategy adopted by the original readers. Paired readers, however, had access to more information than the original readers, in that they knew of the existence of both instructional and could compare each au- thor's attempts at the two types of essays. We can now ask whether this additional information increased the accuracy of the paired readers' judgments and whether paired readers were less likely to succumb to correspondence bias. Again, we corre- lated the authors' true personalities with paired readers' ratings of inferred personality (see columns 2 and 3 in the lower panel of Table 1). The resulting correlations were quite low in com- parison; paired readers were also relatively unsuccessful in their attempts to infer the authors' true personalities. We again corre-

Table 2 Average Intercalations Among Selected Variables From Study 1

Readers' ratings Paired readers' ratings

Authors' ratings

Essay credibility First essays Second essays

Essay extremity First essays Second essays

Essay credibility

.14

.21*

Essay extremity

.28**

.39***

Essay credibility

-.02 .27*

Essay extremity

.27*

.27*

Note. N = 32 ratings. Dash means no data are available. *p<.05. **p<. 01. ***/)< .001.

MULTIPLE AUDIENCE PROBLEM 599

lated the essay personality assignment with the paired readers' ratings of ascribed personality. The resulting correlation again was quite high and positive (see column 4 in the lower panel of Table 1); paired readers tended to ascribe personality in line with the direction of essay assignment.

Relations Between Authors' and Paired Readers' Judgments

To understand why paired readers also apparently missed the cues authors reported placing in their essays (despite the fact that readers reported using them), we again analyzed the corre- lational relations between paired readers' and authors' ratings of essay extremity, essay credibility, and author personality type. These correlations are presented in the right-hand portion of Table 2. As Table 2 shows, the relations between paired read- ers' ratings and authors' ratings of essay credibility and essay extremity are quite small; apparently, authors and paired read- ers, like the original readers, did not agree on what constitutes credibility but did seem to agree somewhat on what constitutes extremity.

At this point, because even paired readers who were allowed to see both essays could not infer the authors' true personalities, we became concerned that authors, despite their claims to the contrary, had failed to imbed clues about their true personali- ties in the second essays. To investigate this possibility, the third author read all 32 pairs of essays with full knowledge of the equal distribution of introverts and extraverts across the four conditions, the order in which the essays were written, and with no time constraints. In other words, the only thing of which the third author was kept unaware was the actual assignment to conditions of each pair. To our relief, 26 of the 32 authors were correctly identified using this procedure (sign test, p < .01). Thus, authors apparently did indeed imbed clues in their sec- ond essays, but normally instructed readers failed to detect those clues.

of war during the Korean war, who were forced to state false beliefs during "self-criticism" sessions, were able to convince their captors that they held those beliefs while simultaneously conveying to other prisoners that they actually held attitudes in opposition to the ones they were publicly proclaiming (Schein, 1956; Schein, Schneier, & Barker, 1961). Similarly, in a recently released videotape, a Beirut hostage communicated the fact that his confession was made under duress and should not be believed (ABC News, 1987). But of course, from a communica- tion perspective, those prisoners had certain advantages over the subjects in our study: They shared a culture and a language that were alien to their captors. They could rely on shared rules of communication among their friends and acquaintances (or among fellow prisoners) to signal their true beliefs. Friends and acquaintances of the Beirut hostage were the first to detect the falsity of his videotaped communication. In contrast, the au- thors in our experiment knew nothing specific about the audi- ence to whom they were to convey their true personalities, noth- ing that might have given them clues about successful commu- nication techniques. They had no chance to spend time with the audience to develop a shared language or set of experiences that could provide the bases for reference that would cue the audience to the fact that the message was other than what it seemed.8

This discussion suggests that knowledge shared with the in- tended target audience enables a communicator to devise a communication strategy; such shared knowledge may be a criti- cal component of successful covert communication. For exam- ple, suppose that the essay readers had been the authors' friends. If the author wanted to make sure his or her friends disregarded the overt content of the essay, the author could use some com- mon point of reference or experience shared with his or her friends to signal the constrained nature of the essay, perhaps contradicting or lying about some thing or event that the friends would know to be true. In their work on overheard communica- tions, Clark and his colleagues (Clark, 1985; Clark & Schaefer,

Discussion

The main findings that emerged from Study 1 are clear: First, authors given the usual rationale for writing a constrained essay presenting a particular personality reported that they were not motivated to convey their true personalities. Not surprisingly, readers did not successfully detect the true personalities of those authors. More surprising, authors motivated to convey their true personalities did not seem to be able to communicate that information successfully. Despite the fact that authors and read- ers claimed to use the same strategies (i.e., manipulating and assessing credibility) and that authors reported being motivated to convey their true personalities in the second essay, they were not successful. Apparently, credibility is not easily manipulated in ways that are clearly understood by the readers.

The failure on the part of our readers to detect the clues that authors embedded in the essays stands in rather stark contrast to the success experienced by others in similar situations. Zuck- erman, DeFrank, Hall, Larrance, and Rosenthal (1979), for ex- ample, found that observers were able to detect posed deception (e.g., "Tell a lie, but communicate that you are lying") from the facial expressions and voices of videotaped targets at levels greater than chance.7 Along different lines, American prisoners

7 It should be noted, however, that there are also several important differences between the present studies and the Zuckerman et al. (1979) research. First, senders in the Zuckerman et al. research were instructed to lie and to pose deception by communicating through their facial ex- pressions that they were lying rather than to communicate a second message within a written essay (as in our research). In other words, those subjects were provided with a strategy to use; our subjects, on the other hand, were not. Second, those subjects were not instructed to attempt to communicate multiple messages to different audiences (i.e., that the surface content had to be believed by one set of observers but not by a different set).

8 Consistent with this argument, it is important to realize that the third author's successful decoding of the essay authors' true personali- ties depended, in large part, on the information that the author gained during the debriefings concerning the kinds of strategies they adopted in response to the task. Specifically, some authors reported in the de- briefing that they actually attempted to make their second essay less credible by including acts that were incompatible with the personalities they presented but that were instead associated with their true personal- ities. In rather stark contrast, others took a different approach and at- tempted to present the assigned trait in such an extreme manner that it became unbelievable. This suggests that authors attempted to alter the credibility of their essays in different (and opposite) ways: Some made their essays more extreme; others made their essays less extreme.

600 FLEMING, DARLEY, HILTON, AND KOJETIN

1987) have referred to these techniques collectively as private keys. We would expect that the larger the body of shared knowl- edge that exists between the communicator and the audience, the easier it will be to convey a subtle message to them while simultaneously conveying the demanded contradicting message to a different audience.9 In other words, the successful commu- nication of mixed messages may be less a function of the com- municator's motivation to convey the truth and more a function of the nature of the audiences to which the messages are di- rected. In Study 2, we examined the possibility that audience differentiation (which creates the possibility of differentiated shared knowledge) may serve as one mechanism through which constrained communicators can effectively communicate one message to one audience while simultaneously conveying a sec- ond, contradictory message to a different, second audience.

STUDY 2

Overview of Hypotheses and Design

Study 2 examined the multiple audience problem, which was described earlier, in two phases. In the first phase, 8 authors came to the laboratory to write four essays each. In each essay they attempted to persuade readers to select one object from a set of four objects. The authors were told that the essays would be read by two groups of readers: (a) a group of their friends (the friends audience), and (b) a group of strangers (the strangers audience). Operationally, the use of these audiences represents a manipulation of the amount of shared knowledge between the authors and the readers. The four essays comprised two truthful messages and two false messages; one of the false messages con- tained a hidden message directed at the friends. Authors were told that the communication would be successful if the two au- diences correctly received their intended messages; it was un- successful if either audience failed to detect their respective messages or if they detected the message intended for the other audience. In the second phase, readers in each of the audiences (friends and strangers) individually read and rated each of the essays.

Initially, we planned to replicate Study 1 by having authors communicate a potentially false message about their introver- sion or extraversion. This approach, however, would have cre- ated the artifactual possibility that friends would know or would be able to guess the author's true personality with consid- erable accuracy. Thus, we shifted the topic of the essay from a communication about personality type to one conveying which object from a set of four the reader should choose. We made one more procedural alteration that was intended to broaden the generality of the results if the predicted findings were ob- tained. In Study 1, authors described themselves in terms of a dichotomous trait dimension: introversion or extraversion. Consequently, if a reader detected that the overt communica- tion (e.g., "I'm an introvert") was untrue, he or she also knew that the author must be an extravert. In Study 2, we removed this possibility by having the author attempt to signal which of four objects was to be chosen.10

The hypotheses tested were first, that authors who were con- strained to make a deliberately false statement would attempt to send hidden messages within the constrained communica- tion and would do so by reference to a domain of shared knowl-

edge between themselves and their friends. Second, we pre- dicted that the authors would be successful: That is, essay read- ers who shared knowledge with the author (his or her friends) would be able to detect and translate a hidden message embed- ded in a false communication. They would use their shared knowledge to discount (or disregard) the surface content of the message and would correctly determine the concealed content of the message. In contrast, essay readers who did not possess shared knowledge (the strangers) would be significantly less able (or unable) to detect and decode the hidden message embedded in a false communication. Instead, they would receive (and be- lieve) the manifest or surface content of the communication.

Method

Subjects

Eight undergraduate students from the introductory psychology sub- ject pool at the University of Minnesota volunteered for an experiment on social communication and served as the essay authors. Each author- subject received course credit and $5 for participating in the experi- ment. We recruited the subjects for the strangers audience from the introductory psychology subject pool as well as other introductory psy- chology classes at the University of Minnesota; all subjects in the strang- ers audience received course credit for their participation. Thirty-two students served as readers in the strangers audience, 4 readers for each of the 8 authors. Each author also nominated four friends, the majority of whom also tended to be university students. The 31 friends who fi- nally agreed to participate received either course credit or $5 for the half-hour judging session."

Procedure

Essay Authors

Essay authors reported to the laboratory and prepared four short es- says ostensibly to be used in a study on lie detection. After completing a set of demographic and filler questionnaires, and reading a pamphlet describing the essay writing task, authors were told that the essays would be read by two different audiences: a group of friends whom they were

9 The concept of shared knowledge has been described by others as common knowledge (Lewis, 1969), common ground (Clark, 1985; Clark & Carlson, 1981, 1982a, 1982b; Clark & Marshall, 1981), and mutual knowledge (Schiffer, 1972).

10 On reflection, we realized that there were two possible lie/truth detection interrelations: In the simplest, if one detects that a lie is being told, deductively one knows the truth. If, for instance, a student claims to have aced an exam, and you detect that he or she is lying, then you know that he or she actually did not receive a perfect score. In more complex, or multialternative situations, the detection of a lie does not necessarily lead to a perception of the truth. For instance, if a student claims to have received an A on an exam, and you detect that he or she is lying, you still do not know which of the remaining possible letter grades (B, C, D, or F) he or she actually received; the discovery of the presence of the lie does not enable you to infer the truth. Our experi- mental tests generally involved the latter sort of situation.

11 One author nominated five friends, only three of whom agreed to participate. As a result, the data for this author are based on three friend readers. Where proportions were generated on the basis of the friends' ratings of the essays, however, we calculated them as though four friends had participated. We considered this a very conservative solution and one that actually works against the hypotheses by reducing the overall proportions of hits for that author.

MULTIPLE AUDIENCE PROBLEM 601

asked to nominate and a group of strangers from the introductory psy- chology subject pool.

The written instructions presented a role-playing scenario in which authors were to imagine themselves as members of an underground re- sistance movement who had been captured by the enemy. Authors were also instructed to imagine that they were being held hostage and that they were being forced by the enemy to write a set of essays that would later be read by their fellow resistance members (their friends). The es- says were to consist of a short persuasive argument that should convince readers in the two different audiences (friends and strangers) to choose one randomly predetermined box (the target) from a set of four boxes ostensibly containing a set of secret maps that would be of assistance to the resistance. We randomly preselected one of the remaining boxes as the decoy and informed the authors which box had been so designated. Each of the boxes (labeled A through D) was unique in color, size, and shape.

To mimic conditions in the typical assigned essay paradigm, the au- thors wrote four essays: two truthful messages (the truth trials) and two deceptive messages, one consisting of an outright falsehood (the lie trial) and one containing a hidden message designed to get the friends to select the target box (the multiple message trial). The actual content of the arguments for and against each of the four boxes could consist of any- thing the communicator desired, but in order to minimize the variabil- ity among the essays, we instructed authors to follow a prescribed for- mat for the presentation of their arguments. Specifically, they were to request the readers to select the target and then to give a reason (or reasons) for them to do so. Then, communicators were to identify the remaining three alternatives and provide reasons why they should not be selected. Authors were instructed to conclude their arguments with a reiteration of their requests to select the target. The order of trials was randomized for each author.

Truth trials. For the two truthful essays, authors were told that they should try to convince both audiences to select the preselected target box and not to select any of the other three boxes. In other words, au- thors were to lead both audiences to the location of the secret maps (ostensibly) hidden inside the target box. For example, if the maps were hidden inside Box A, then on the truth trials, authors were to get both audiences to select Box A.

Lie trial. For the lie trial, authors were told that they should try to convince both audiences to select the preselected decoy box and not to select any of the other three boxes. In other words, authors were in- structed to mislead both audiences as to the true location of the secret maps (the target). For example, if the maps were hidden inside Box A (the target) and Box D was identified as the decoy, then on the lie trial, authors were to get both audiences to select Box D. Thus, the difference between the truth trial and the lie trial was that on the lie trial, authors knew which box was the target and they were not directing any of the readers to it; this corresponds closely to other experimental conditions in which introverts write as extraverts with the author and the experi- menter being the only ones who know the truth. '2

Multiple message trial. Finally, in the multiple message trial, which was of primary interest in Study 2, authors again were provided with both a decoy and a target randomly preselected from among the four boxes. On this trial, authors were told that they should try to convince only the strangers to select the decoy box and not to select any of the other three boxes; however, they were also asked to simultaneously com- municate the identity of the target box only to their friends so that their friends would select that box. In other words, authors were to mislead the strangers as to the location of the secret maps while simultaneously communicating the true location to their friends. For example, if the maps were hidden inside Box A (the target) and Box D was identified as the decoy, then on the multiple message trial, authors were to get the strangers to select Box D and their friends to select Box A. Thus, in the multiple message trial, authors were aware not only that they were not telling the truth, but also that they were actually attempting to actively deceive the strangers and to send a hidden message to their friends.

The experimenter placed the boxes on the table in front of the authors and gave them as much time as they wanted to write their essays. The experimenter answered any questions but would not give any direct aid in deciding on a particular strategy to use to communicate the various messages. When they were finished, authors reviewed their essays and were allowed to rewrite any essay with which they were not satisfied. Finally, we arranged the four essays for each author in rating booklets, each in a different counterbalanced order.

Essay Readers

Readers arrived for a study on lie detection, completed a set of demo- graphic questionnaires, and were told that they would be reading a series of four essays consisting of a short persuasive argument to get them to choose one box from a set of four boxes. Each then received a booklet describing the essay reading task. For the friends, the written instruc- tions presented a role-playing scenario in which they were to imagine themselves as a member of an underground resistance movement desig- nated to receive secret communications from other resistance members. We also instructed readers to imagine that the essay author might have been captured by the enemy and there was reason to suspect the truth- fulness of the essays: Specifically, we told the readers that of the four essays they would read, two truthfully directed them to the secret maps and two directed them to incorrect locations. Strangers received identi- cal instructions, except that their affiliation was with the enemy rather than with the resistance. None of the readers was explicitly told about the existence of the hidden communication.

The experimenter placed the four boxes on the table in front of the readers. After reading each essay, all readers indicated which box they believed had been designated as the target and rated their confidence in that choice and the other alternatives on an 11-point scale with end- points labeled very confident the box was the correct choice (11) and very confident the box was not the correct choice (1) with an unmarked, neutral midpoint. We slightly restricted readers' responses on this scale such that their ratings of the chosen box had to be placed above the midpoint toward being confident that the box was correct. The other nonchosen alternatives had to be rated on the other side of the midpoint toward being confident that the box was not correct. In other words, readers rated all four boxes on the same scale, with the stipulation that their choice be placed on the was side of the midpoint and the other three alternatives be placed on the was not side. If they had simply guessed, we told them that they could put all of their choices on the scale midpoint. Finally, on an 11-point scale with endpoints labeled not at all(\) and very well (11), all participants indicated how well, prior to the experiment, they knew the essay author. Also, the participants wrote a brief explanation of what information or cues they had used to decide on the chosen box.

After completing their task, all subjects were told about the existence of the hidden message, debriefed about the actual purpose of the study, thanked for their participation, and excused.

Results

Ratings of How Well Readers Knew Author

In Study 2, we operationalized shared knowledge by using two groups of readers who differed in the extent to which they

12 Of course, we need to acknowledge that the box paradigm differs from the assigned essay/personality attribution paradigm in several ways. Specifically, in the personality attribution paradigm, authors pre- sumably have some investment in being accurately perceived; in the box paradigm, this investment is lower. We attempted to reduce this difference by making the communication task as involving as possible for the authors.

602 FLEMING, DARLEY, HILTON, AND KOJETIN

knew the author. As a check on this manipulation, readers' rat- ings of how well they knew the author were submitted to a 2 (reader audience: friends vs. strangers) X 8 (author) ANOVA. This analysis revealed a significant main effect for the reader audience, F( 1,47) = 295.25, p < .0001. Friends reported know- ing the author significantly better (M = 11.48) than did the strangers (M = 2.44). Neither the author main effect nor the Reader Audience X Author interaction was significant on this measure, both Fs(7,47) < 1.9.

Accuracy at Selecting Target Box

Because we were interested in differences between our two reader audiences in their ability to correctly select the target, our primary dependent measure was the number of hits (correct selections of the target from each set of boxes on each message trial) registered by members of the two reader audiences. In other words, our main dependent measure was the number of correct guesses on each message trial.

With categorical data such as these, it is generally appropriate to use one of the more recently developed nonparametric ana- lytic techniques such as loglinear models, logit or probit regres- sions, or categorical modeling programs. Because the large number of structural zeros in the present design would render such procedures too unreliable, however, we decided to use a different strategy. Specifically, within each audience, we calcu- lated the proportion of hits among the readers for each author on each message trial (thereby collapsing the data across each author's readers). Then, following procedures recommended by Bishop, Feinberg, and Holland (1975, pp. 367-372), Cohen and Cohen (1983, pp. 265-270), and Kenny (1985, pp. 502-503), we transformed those proportions using an arc sine (angular) transformation; this transformation strategy yields data that as closely as possible approximate parametric data that can then be analyzed using conventional ANOVA procedures. Thus, we submitted the transformed percentages (as well as the raw pro- portions) to separate 2 (reader audience: friends vs. strangers) X 3 (message trial: truth, lie, and multiple message) fixed effects ANOVAS, with repeated measures on the second factor.13 Be- cause the results of the analyses of the transformed and raw proportions were virtually identical, for ease of explication we will report only the raw proportion results.

We expected both reader audiences to achieve approximately equal numbers of hits on both the truth and the lie trials. On the multiple message trial, however, we expected the friends to achieve a significantly larger proportion of hits than the strang- ers. In other words, we predicted a Reader Audience X Message Trial interaction. As can be seen in Table 3, our predictions were supported. Specifically, a significant Reader Audience X Message Trial interaction emerged with all means in the antici- pated directions, F(2,28) = 6.64, p < .01. A simple effects anal- ysis revealed no significant differences between the reader audi- ences in the proportion of hits on the truth trials or the lie trial, Fs(l, 41) = 1.33 and 1.45, respectively. The predicted simple effect difference did emerge, however, on the multiple message trial. Friends registered a significantly higher proportion of hits (correctly guessing that the actual target location was the cor- rect choice on that trial) than did the strangers, F(l, 41) = 13.08, p<. 001.

In addition, a significant repeated-measures effect emerged

Table 3 Cell and Marginal Mean Proportions for Truth Detection, Confidence, and Lie Detection Analyses, and z Score Analysis From Study 2

Message trial

Reader audience

Friends Strangers Marginal

Truth

53.12a 41.15. 47.14A

Lie

Truth detection3

6.25b 18.75b 12.50B

Multiple message

56.25a 18.75b 37.50A

Marginal

38.54A 26.22A

Confidence15

Friends Strangers Marginal

8.70a 7.20a 7.95A

3.69b 5.86C 4.78B

9.07a 5.89C 7.48A

7.15A 6.32A

—

Friends Strangers Marginal

46.87. 58.85a 52.86A

Lie detection0

40.62, 46.87a 43.75A

75.00b 50.00, 62.50A

54.17A 51.91A

Corresponding z scores for lie detection analyses

Friends Strangers Marginal

-0.35 1.00 0.46

-1.06 -0.35 -1.00

2.83* 0.00 2.00*

0.82 0.37

Note. Cell n - 8. Except for the lie detection means, cell means within a column or row that share a lowercase subscript, and marginal means that share an uppercase subscript, do not differ at p < .05 by Newman- Keuls test. Dash means no data are available. a Entries are proportions of correct selections of the target box. b Entries are ratings made on 15-point scales of each reader's confidence in the box that was the target on each trial. c Entries are proportions of responses in which readers selected a box other than the one the authors requested them to select (i.e., readers suspected a lie and did not use the surface content of the message to make their choices). Cell means within a column or row that do not share a lowercase subscript differ at p < .07 by Newman-Keuls test. *p<.05. **p<.01.

for the message trial variable, F(2,28) = 13.58, p < .001. Across the audiences, readers were significantly more accurate on the truth trials than they were on the lie trial or multiple message trial. The reader audience main effect was marginal in this anal- ysis, F(\, 14) = 3.40, p < .09. Across all message trials, friends registered a marginally higher proportion of hits than did the strangers, reflecting that the friends had significantly greater ac- curacy on the multiple message trial.

Readers' Confidence in Their Choices

Readers' ratings of their confidence that the target box was or was not the correct choice were submitted to a 2 (reader audi-

13 To balance the number of truthful and deceptive trials each ob- server would watch, we originally used two truth trials. Because these two truth trials were identical in every respect, we pooled observers' responses on these two trials into a single combined truth trial in which the mean number of hits across the truth trials was used.

MULTIPLE AUDIENCE PROBLEM 603

ence) X 8 (author) X 3 (message trial) ANOVA, with repeated measures on the last factor. In other words, in this analysis we used readers' ratings of confidence for the target box on each trial. We expected the results of this analysis to be identical to those from our main dependent measure, and they were. Spe- cifically, a significant Reader Audience X Message Trial interac- tion emerged, F(2,94) = 8.74, p < .001. Simple effects analysis revealed no significant differences between the two reader audi- ences in their confidence that the actual target box was the cor- rect choice on the truth trials, F(l, 135) = 2.23.'" Strangers re- ported significantly more confidence, however, that the actual target box was the correct choice on the lie trial, F(l, 135) = 4.71, p < .05. Finally, the predicted simple effect difference emerged on the multiple message trial, /^l, 135) = 10.12, p < .002. Friends expressed significantly greater confidence that the actual target box was the correct choice on that trial than did the strangers.

In addition, a significant repeated-measures effect emerged across the message trial variable, F(2, 94) = 13.71, p < .001. Readers were significantly more confident that the target was the correct choice on the truth trials and the multiple message trial than they were on the lie trial. Confidence ratings for the truth trials and the multiple message trial did not differ reliably. Finally, the reader audience main effect was not significant, F( 1, 47)= 1.62.

Lie Detection Versus Truth Detection

The first major question addressed by Study 2 was whether readers who shared common knowledge with an author were more accurate than those who did not in determining the un- derlying truth in a potentially false communication. The results suggest that they were more accurate. A second and equally im- portant question, however, is whether this increased accuracy was obtained at the cost of being discovered sending hidden messages by the second audience. In other words, did authors sacrifice secrecy in order to convey their hidden messages, or were they able to communicate them without being discovered? Multiple audience problems differ from situations in which one can be overheard (e.g., Clark & Schaefer, 1987) in that commu- nicators caught in multiple audience problems must not only communicate a different message to one of the present audi- ences but must also do so without being detected by the other audience.

The data from this experiment also allowed us to address this issue. Specifically, within each reader audience, we receded the original truth detection data and calculated the proportion of responses in which the readers selected a box other than the one the author told them to select (i.e., selected a box other than the target box on the truth trials or the decoy box on the lie trial and multiple message trial) for each author on each message trial. In other words, in this analysis we used the proportion of responses in which readers were not convinced by the surface content of the essay and suspected the author of lying. As with the truth detection analyses, we transformed these proportions using an arc sine transformation and then submitted these data sets to separate 2 (reader audience) X 3 (message trial) ANOVAS, with repeated measures on the second factor. Again, because the patterns of significance in the analyses of the transformed

and raw proportions were identical, we report only the raw pro- portion results.

We reasoned that if authors were communicating their hid- den messages only to their friends without betraying to the strangers the fact that they were sending hidden messages, then no differences should emerge across message trials in the extent to which the strangers suspected a lie (i.e., they selected a box other than the one the author requested them to select). More- over, we expected that the friends would not differ reliably on this measure either, with the exception that their demonstrated enhanced accuracy on the multiple message trial would also translate into better lie detection on that trial. Consistent with this interpretation, results of this analysis revealed no signifi- cant main or interaction effects (all Fs < 2.14, all ps > . 13; see the third panel of Table 3). Across all message trials, neither audience selected a box other than the one the communicator requested them to select more frequently, F( 1,14) < 1; nor were the combined audiences better able to detect a lie from among the three message trials, F( 1,14) = 1.90, p <. 11. We also com- pared each of these marginal means with the chance guessing rate of 50%. As the bottom panel in Table 3 shows, with the exception of the multiple message trial mean, none of the mar- ginal means differed significantly from chance levels. The guess- ing rate for the multiple message trial did exceed the chance level as a result of the friends' significantly better hit rate on that trial. Thus, in general, it appears that readers in both audiences adhered to our original baseline parameters for lying and truth telling: Two of the trials were true, and two were false. More important, these data suggest that authors successfully commu- nicated their hidden messages without arousing the suspicions of the strangers. Consequently, this demonstrates that the au- thors successfully executed their strategic communication task and effectively eliminates a number of guessing strategy expla- nations for the previously demonstrated effects.

Authors' Hidden Message Strategies

While completing the essay writing task, all 8 authors re- ported that the task seemed extremely difficult. Nonetheless, all senders were able to complete the task. Without help or guid- ance from the experimenter, 7 of the 8 authors arrived at essen- tially similar strategies for communicating their hidden mes- sages: They all used the falsification of some self-relevant infor- mation that only their friends would know about them. For example, one author who had a love for Wales and for chocolate, but not for a local department store, communicated the hidden message by talking about the possibility of traveling to Wales while describing the box she wanted her friends to select, and by mentioning the department store while praising the box she wanted the strangers to choose. Another author mentioned the names of several potential dating partners while presenting rea- sons why the target box should not be chosen. The eighth author used a different (and substantially less effective) strategy; the author capitalized every word in the essay that began with the letter B to communicate that Box B should be chosen. Compar-

14 Again, observers' responses on the two truth trials were pooled. Specifically, we used the mean confidence rating across both truth trials for each observer.

604 FLEMING, DARLEY, HILTON, AND KOJETIN

isons of the readers' explanations for their strategies, coupled with measures of communication success, suggested that al- though the hidden communications were, in general, quite suc- cessful, some (such as the capitalization ruse) were less success- ful than others. Moreover, communicators who were unsuccess- ful usually failed not because their friends could not understand their messages, but instead because the strangers were able to detect and decode them. That is, failures were due to using a strategy that had too large a scope, strategies that did not effec- tively differentiate among the audiences.

STUDY 3

In Study 2, the communicators prepared written essays to communicate the various messages to the readers in the strate- gic communication task we gave them. Obviously, readers had access only to the verbal content of the messages in making their judgments; written essays do not incorporate the numerous nonverbal and paralinguistic cues that not only may facilitate the communication of hidden messages but also may lead to their increased detection by unintended audiences (see Zucker- man et al, 1979). Consequently, the procedure we used in Study 2 may not have provided a complete picture of strategic communication in multiple audience situations. Moreover, readers in Study 2 were informed about the task in which the authors had engaged. They knew, for example, about the role- playing scenario, and this additional situational information may have contributed to the friends' demonstrated abilities to detect and decode the hidden messages. In Study 3, then, we set about to replicate and extend the general findings from Study 2 by examining strategic messages delivered over a broader com- munication channel. To this end our senders made videotaped (rather than written) messages that were later shown to the vari- ous audiences, who were not given any information about the task in which the senders had engaged. In addition, we added an audience of strangers whose performance we hoped would increase the generality of the findings.

Thus, in Study 3, individuals made a set of four videotapes, in one of which they were asked to communicate a false message to two different audiences while simultaneously conveying a true and contradictory message to a third audience. The (stu- dent) senders knew that the first two audiences would be fellow students who did not know them, whereas the third audience would be students whom they had identified as their friends.

Method

Subjects

We recruited 10 senders to participate in Study 3: 8 graduate students (4 women, 4 men) from six different specialty areas within the psychol- ogy department at the University of Minnesota, 1 male undergraduate research assistant, and 1 male undergraduate office worker. The mean age of the senders was 24 years. Each received $ 10 for the 2-hr videotap- ing session. Each sender nominated four friends to serve as observers, the majority of whom tended to be University of Minnesota social sci- ence graduate students. The 40 friends (20 women, 20 men, mean age = 27 years) were paid $5 for the half-hour judging session.

We also recruited a group of strangers who were matched to the friends on several variables, including age and academic status. The 40 matched strangers (24 women, 16 men, mean age = 27 years), 4 for each of the 10 senders, were also University of Minnesota social science

graduate student volunteers and advanced psychology undergraduate students; each student received $5 for their participation. Finally, the unmatched strangers in Study 3 were recruited from the introductory psychology subject pool as well as other introductory psychology classes at the University of Minnesota. These subjects either received course credit or $5 for their participation. A total of 40 students (22 women, 18 men, mean age = 21 years) served as unmatched strangers, 4 for each of the 10 senders.

Procedure

Senders

With the exceptions described later, the procedure for Study 3 was identical to that used in Study 2. In the first phase, the 10 senders came to the laboratory to make four videotaped messages following the proce- dures previously described. In each of the videotapes, they attempted to persuade observers to choose one randomly predetermined song from a set of four popular songs. To minimize variability in the content and style of the songs, we used four popular songs from the same artist or group for each videotaped message.15 The senders were told that the videotapes would be viewed by two groups of observers: (a) a group of their close friends (the friends audience), and (b) two groups of persons whom they did not know (the matched and unmatched strangers audi- ences). Again, the use of these groups represented a manipulation of the amount of shared knowledge between the senders and the observers. As in Study 2, the four videotaped trials comprised two truthful messages and two false messages; one of the false trials contained a hidden mes- sage directed at the friends.

Before making the videotapes, senders were allowed to listen to a 1- min segment of each of the songs and were given as much time as they wanted to think, make notes, or write an argument verbatim. The exper- imenter answered questions but would not give senders any direct aid in deciding on a particular strategy to use. When senders were ready, they were videotaped individually presenting their arguments. The senders reviewed their videotapes and could remake any tape with which they were not satisfied; only two senders decided to remake one trial each. Finally, we transferred the four videotapes for each sender to a master tape, each arranged in a different counterbalanced order.

Observers

Observers arrived for a study on lie detection and completed a brief demographic information sheet. We then told them that they would be watching a series of videotapes of a person delivering a short persuasive argument to get them to choose one song from a set of four popular songs. Unlike Study 2, however, the role-playing task that the senders had completed was not described to the observers. Instead, observers were told only that in two of the four tapes they would see, the person was telling the truth and that in the other two, the person was lying. Observers were asked to carefully watch the videotapes and to try to decide whether the person was lying or telling the truth, and to guess which of the four songs had been preselected as the target song. The observers were also allowed to listen to 1-min segments from each of the songs before watching the videotapes.

After viewing each tape, all observers indicated which song they be- lieved was the correct choice (had been designated as the target) and completed the same dependent measures used in Study 2. After com- pleting their task, all subjects were told about the existence of the hidden

15 Specifically, we used four songs by Whitney Houston, four by Billy Joel, four by R.E.M., and four by the Beatles. We used four choice alter- natives (rather than two) to ensure that simply detecting a lie would not necessarily lead to the correct selection of the target song by default (see Footnote 10).

MULTIPLE AUDIENCE PROBLEM 605

message, debriefed about the actual purpose of the study, thanked for their participation, and excused.

Results

Ratings of Hew Well Observers Knew Sender

As in Study 2, observers' ratings of how well they knew the sender were submitted to a 3 (observer audience: friends, matched strangers, and unmatched strangers) X 10 (sender) AN- OVA. This analysis revealed a significant main effect for ob- server audience, F(2, 90) = 87.39, p < .001. Friends reported knowing the sender significantly better (M - 8.59) than the un- matched strangers (M = 4.45), who in turn reported knowing the sender significantly better than the matched strangers (M = 3.04, all comparisons significant at p < .05 by Newman-Keuls test). There was also a significant main effect for how well each of the 10 senders was known, F(9, 90) = 2.69, p < .01. A New- man-Keuls analysis revealed that the differences between the best known sender (M = 7.25) and the three least known senders (Ms = 4.08,4.46, and 4.71, respectively) were significant at p < .05. Finally, the Observer Audience X Sender interaction was not significant on this measure, F( 18,90) = 1.34.

Accuracy at Selecting Target Song

Overview of Main Analysis

Again, because we were interested in differences among our three observer audiences in their ability to detect the target song, our primary dependent measure was the number of hits registered by members of the three observer audiences. Within each observer audience, we calculated and transformed the pro- portion of hits for each sender on each message trial as in Study 2 (thus collapsing the data across each sender's observers). These data sets were then submitted to separate 3 (observer au- dience) X 3 (message trial: truth, lie, and multiple message) fixed-effects ANOVAS, with repeated measures on the second factor. Again, only the raw proportion results are reported; these appear in Table 4.

We expected that all three observer audiences would achieve approximately equal numbers of hits on both the truth trials and the lie trial. On the multiple message trial, however, we ex- pected the friends to achieve a significantly larger proportion of hits than either group of strangers. In other words, we predicted an Observer Audience X Message Trial interaction. As can be seen in Table 4, our predictions were again supported. Specifi- cally, a significant Observer Audience X Message Trial interac- tion emerged with all means in the anticipated directions, F(4, 54) = 2.67, p < .05. A simple effects analysis revealed no sig- nificant differences among the observer audiences in the pro- portion of hits on the truth trials and the lie trial, both Fs(2, 80) < 1. The predicted simple effect difference did emerge, how- ever, on the multiple message trial, F(2, 80) = 5.29, p < .01. Friends registered a significantly higher proportion of hits on that trial (correctly guessing that the actual target song was the correct choice) than did the matched or unmatched strangers, who did not differ significantly. The observer audience main effect was not significant, F(2, 27) = 1.42.

In addition, a significant repeated measures effect emerged for the message trial variable, F(2, 54) = 31.54, p < .001. Across

Table 4 Cell and Marginal Mean Proportions for Truth Detection, Confidence, and Lie Detection Analyses, and z Score Analysis From Study 3

Message trial

Observer audience

Friends Matched

strangers Unmatched

strangers Marginal

Friends Matched

strangers Unmatched

strangers Marginal

Friends Matched

strangers Unmatched

strangers Marginal

Truth

57.50a

51.25,

48.75a 52.50A

6.66,

6.02,

6.24, 6.30A

42.50a

48.75,

51.25, 47.50A

Lie

Truth detection"

10.00,,

20.00b

17.50b 15.83C

Confidence"

4.29b

4.56b

4.14b 4.33C

Lie detection0

47.50.

52.50a

62.50a 54.17A

Multiple message

42.50a

17.50b

20.00b 26.67B

6.0 la

4.55b

4.26b 4.94B

67.50a

52.50,

52.50a 57.50A

Marginal

36.67A

29.58A

28.75A

5.65A

5.05A

4.88A

52.50A

54.58A

52.08A

Corresponding z scores for lie detection analyses

Friends Matched

strangers Unmatched

strangers Marginal

-0.95

0.16

-0.16 -0.55

-0.32

0.32

1.58 0.91

2.21**

0.32

0.32 1.64*

0.55

0.46

1.00

Note. Cell n = 10. Cell means within a column or row that share a lowercase subscript, and marginal means that share an uppercase sub- script, do not differ at p < .05 by Newman-Keuls test. Dash means no data are available. 1 Entries are proportions of correct selections of the target song. b Entries are ratings made on 15-point scales of observers' confidence in the song that was the target on each trial. c Entries are proportions of responses in which observers selected a song other than the one send- ers requested them to select (i.e., observers suspected a lie and did not use the surface content of the message to make their choices). *p = .10. **p<.01.

the three audiences, observers were most accurate on the truth trials, significantly less accurate on the multiple message trial, and even less accurate on the lie trial (see Table 4 for specific mean comparisons).

Observers'Confidence in Their Choices

Because no significant main or interaction effects emerged from preliminary analyses that included the senders variable, this variable was not included in subsequent analyses. Conse- quently, we submitted observers' confidence ratings to a 3 (ob-

606 FLEMING, DARLEY, HILTON, AND KOJETIN

server audience) X 3 (message trial) ANOVA, with repeated mea- sures on the last factor.

We expected the results of this analysis to be identical to those from our main dependent measure, and as the middle panel of Table 4 shows, they were. Specifically, a significant Observer Audience X Message Trial interaction emerged, F(4, 234) = 2.54, p < .05. No significant differences emerged among the ob- server audiences in their confidence that the actual target song was the correct choice on the truth trials or the lie trial, both /s(2, 329) < 1. Again, however, the predicted simple effect difference emerged on the multiple message trial, F(2, 329) = 6.77, p < .001. Friends expressed significantly greater confi- dence that the actual target song was the correct choice on that trial than did either the matched or unmatched strangers.

Finally, a significant repeated-measures effect emerged across the message trial variable, F(2,234) = 28.79, p < .001. Observ- ers were most confident that the target was the correct choice on the truth trials, significantly less confident that the target was the correct choice on the multiple message trial, and least confident that the target was the correct choice on the lie trial. A marginal main effect also emerged among the three observer audiences, F(2, 117) = 2.79, p < .07. Friends were marginally more confident in their ratings of the actual target than were the matched or unmatched strangers, which were not reliably different.

Lie Detection Versus Truth Detection

Again, one of the major questions addressed by Study 3 was whether observers who shared common knowledge with a sender were more accurate than those who did not in determin- ing the underlying truth in a potentially false communication. The results demonstrated that even when the communication channel is broad (i.e., using videotaped rather than written es- says) and even when observers are not told about the senders' task, observers are more accurate. As in Study 2, a second and equally important question was whether this increased accu- racy was obtained at the cost of being discovered sending hid- den messages by other observers. It is possible that increasing the information available to all observers by using a broad com- munication channel could actually lead to higher lie detection rates by all three audiences in Study 3. The same lie detection analyses that we performed in Study 2 can also be applied here; the cell and marginal means from these analyses are presented in the lower panel of Table 4.

Results of the lie detection analysis revealed no significant main or interaction effects (all Fs < 1.06, ps > .35). As Table 4 shows, across all message trials, none of the audiences selected a song other than the one the sender requested them to select more frequently; nor were the combined audiences better able to detect a lie from among the three message trials. Again, we compared each of the marginal and cell means with the chance guessing rate of 50%. As in Study 2, with the exception of the multiple message trial mean, none of the marginal means differed significantly from chance levels. The guessing rate for the multiple message trial did marginally exceed the chance level (significant at p = .10) as a result of the friends' signifi- cantly better hit rate on that trial. Thus it appears that observers in the three audiences in Study 3 also adhered to our original baseline parameters for lying and truthtelling: Two of the trials

were true, and two were false. More important, these data sug- gest that senders successfully communicated their hidden mes- sages without arousing the suspicions of the strangers, even though the communication channel had been broadened. Con- sequently, this supports the major findings of this research: Not only were the senders able to communicate different messages to the different witnessing audiences, but they managed to com- municate a message to their friends that was undetected by the strangers.

Senders' Hidden Message Strategies

While completing the videotaping task, all 10 senders re- ported that the task seemed extremely difficult. (Nonetheless, all senders were again able to complete the task.) Without help or guidance from the experimenter, all 10 senders arrived at es- sentially similar strategies for communicating their hidden mes- sages: Eight senders used the falsification of some self-relevant information that only their friends would know about them, and 2 senders used a set of uncharacteristic (or characteristic but inappropriate) facial expressions while making the video- tapes. For example, 1 sender who had recently lost a great deal of weight communicated his hidden message while describing the target song by talking about how he was presently gaining weight. Another sender mentioned the name of her fiance while presenting reasons why the target song should not be chosen. A different sender smirked and raised her eyebrows when telling the observers not to select the target song—a characteristic fa- cial expression that to her friends meant she was being sarcastic.

Discussion

The present results provide strong evidence for a multiple au- dience effect. Communicators in both Studies 2 and 3 spontane- ously generated techniques to conceal cues to their friends that the overt content of the messages they were communicating to strangers was sometimes false. These techniques involved the use of private keys—references to knowledge that is shared be- tween the communicator and the friends (but not the communi- cator and the strangers)—to alert the friends both to the fact that this was one of the trials on which deception was taking place and to what the correct selection should be. In Studies 2 and 3, the most frequent variant of this strategy involved refer- ring to some personal information accessible to friends but not to strangers. In addition, senders in Study 3 used another vari- ant of the private keying strategy. This second strategy involved using facial expressions that because of the history surrounding these expressions in interactions with friends, apparently had meanings that were known and highly likely to be recognized within the friendship group but were not likely to be so idiosyn- cratic or out of place as to give away their meanings to strangers.

These strategies were generally successful, in that friends, who were alerted to the existence of a certain frequency of de- ceptive messages within the set of four essays or videotapes, were often able to infer the correct answer in the one message in which the communicators attempted to encode the truth in a way that would be detectable by the friends. Although equally alerted to the existence of a certain frequency of deceptive mes- sages within the set of four messages, strangers were not gener-

MULTIPLE AUDIENCE PROBLEM 607

ally able to detect or decode the true content of the hidden mes- sages.

In addition, it is worth noting that none of the observers in Study 3 were aware of the possibility that hidden messages were being sent when the task began. Instead, we simply told them that the sender had lied on half of the trials. Interestingly, this created a predicament for the observers. Although many of the friends detected and correctly interpreted the meaning of the hidden signals, some who detected them did not interpret them correctly. Using an unusually Byzantine logic, they reasoned that their friends (the senders) used the clues they detected to throw them further off the correct track. That is, because they knew that the sender was instructed to lie on some trials, they reasoned that the senders were using the hidden messages to deceive them more effectively rather than to provide correct in- formation. What is clear is that the task we gave the senders and the observers was by no means a simple one. These observations underscore at least one way in which covert communication at- tempts can go wrong.

Because we used multiple, rather than two, choice alterna- tives within each communication and because the communica- tions were conveyed both in written form and on videotape, we are able to conclude that the major path to the preceding result was not solely through the detection that the communicator was lying. Nor, more specifically, was it through the increased likeli- hood that the communicator's friends were better able to recog- nize the emission of the traditional lying cues of deception cues and leakage (Ekman & Friesen, 1969, 1974; cf. Zuckerman et al., 1979). Rather, friends were able to recognize the meanings of the clues that the senders planted in the communication di- rected solely at them and were able to decode the meaning of those clues. For instance, one sender in Study 3 pointed out that the presence of religious symbolism in the target song greatly annoyed him; he claimed to have a tremendous disdain for reli- gion and stressed his belief that religion had no place in popular music. In fact, he held very deep and strong religious convic- tions. His friends reported that this passage alerted them to the falseness of the communication and signaled the actual location of the target song. Strangers, who did not have this shared piece of common knowledge, were not clued in by it.

Finally, because we matched one stranger audience with the friends audience on several important variables (specifically age, method of payment for participation, and academic status), we are also able to conclude that the major path to the differen- tial detection abilities on the multiple message trial was not ex- clusively a result of these characteristics of the observer audi- ences. The two groups of strangers did not differ from one an- other in terms of their performance, despite the fact that one audience was matched to the friends audience. If any of these audience variables were responsible for the results we obtained, we would have expected the friends and the matched strangers to perform more similarly, and they did not.

GENERAL DISCUSSION

Taken together, the results from the current studies demon- strate that behavior is not always taken at face value. Under certain circumstances, people are capable of conveying accu- rate information (or possibly an accurate self-image) even when they are constrained. Whether they are successful depends on

several factors. First, the communicator must be motivated to convey an accurate image of his or her true personality, opinion, or information. As we have demonstrated, essay authors in the assigned position paradigm are not particularly motivated to present their true personalities in the essays they write, and readers of the essays fall prey to the correspondence bias. Sim- ilarly, numerous studies in the lie detection literature demon- strate that when people seek to conceal their true personalities or opinions, they can be quite successful (see Ekman, 1985).

Second, the audiences witness to the behavior must have some reason to doubt the veracity of the presentation (Ekman & Friesen, 1969; Zuckerman et al., 1984). Most of the time, audiences operate in a conspiracy mode (Kraut, 1978) in which they not only honor a person's claim to a specific role but also actively help support that claim (see also Goffman, 1959). As a result, and assuming that their criteria for assessing the validity of a performance are met (e.g., our readers' assessments of essay credibility and extremity in Study 1), observers will take behav- ior at face value and infer a corresponding disposition. Some- times, however, audiences become suspicious and begin ques- tioning the communicator's intent. In some cases, expectations about the person or situation can arouse general suspicions on the part of an observer. It is the wealth of the widower, for exam- ple, that leads us to question the intentions of his pursuer (Fein, Hilton, & Miller, in press). Similarly, when observers are told by an experimenter that a sender may be lying, they will be sus- picious (e.g., Toris & DePaulo, 1984). In other cases, suspicion may be aroused by some aspect of the display that seems unex- pected, untoward, or unnatural (see Ekman & Friesen, 1969, 1974; Krauss, 1981; Zuckerman, Amidon, Bishop, & Pomer- antz, 1982). If observers suspect that a communicator is being untruthful, they may move into an agonistic or sparring mode (Kraut, 1978) and begin searching for diagnostic information in the presentation (Fein et al., in press; Hilton, Miller, Fein, & Darley, 1989; Pyszczynski & Greenberg, 1981). As we have shown, however, unless the cues observers seek out are unequiv- ocal or particularly informative, they are not likely to arrive at an accurate impression of the communicator.

We have pointed to the existence of a class of situations, mul- tiple audience problems, in which individuals attempt to con- vey different messages to different, but simultaneously present, audiences. In one large subclass of these situations, communica- tors want the larger audience to accept the surface content of their messages, while desiring that some targeted subset of that audience receives a more complex message that is hidden from the larger audience. For this complex task to be carried out suc- cessfully, we suggest that the communicator and the targeted subaudience must have some shared knowledge that creates the possibility of a communication strategy. By referencing this shared knowledge, constrained communicators can both signal the presence of constraint, and reveal to that subaudience the true message, while masking its content and even its existence from the larger audience. Friends in Studies 2 and 3 detected and correctly interpreted the clues that communicators put into their messages and arrived at an accurate appraisal of the under- lying message. Strangers who did not possess the requisite shared knowledge, on the other hand, took the communicators' messages at face value.

At a more general level, this research is consistent with a growing research tradition that attempts to take into account

608 FLEMING, DARLEY, HILTON, AND KOJETIN

the strategic nature of our social lives in our experimentation. Impression management theory (Schlenker, 1985; Tedeschi, 1981), for example, emphasizes people's motives to convey pos- itive impressions of themselves to others. Likewise, researchers interested in strategic self-presentation stress the instrumental component of presenting a desired self-image (Jones & Pittman, 1982). Similarly, recent conceptions of intelligence have taken multidimensional approaches to understanding intelligence that frequently include concerns of a strategic nature (e.g., Can- tor & Kihlstrom, 1987; Sternberg, 1986). Still others have ar- gued that human communication is itself fundamentally instru- mental and that the goals of the participants in the communica- tion game must be taken into account in order to reach a comprehensive understanding of social behavior (Higgins, 1981; Kraut &Higgins, 1984).

Finally, the results presented here suggest that a communica- tion perspective may help us to understand certain aspects of the social perception process. Our typical experimental meth- ods for studying social perception tend to eliminate the possibil- ities of developing shared understandings of their task between observers and targets. In the everyday world, however, those in- teractions are often possible, and as Goffman (1959, 1961, 1969) has noted, both actors and observers are capable of strate- gic interaction behaviors that facilitate decoding quite complex social messages. We have demonstrated that given the existence of shared knowledge with a targeted subaudience, social actors spontaneously generate and use complicated communication strategies to convey mixed messages, and the targeted observers can detect and interpret those strategies and arrive at accurate perceptions of the communicators' messages and dispositions. If conditions that allow for these possibilities are included in future social perception research, a more complex empirical picture may emerge of the conditions under which a person's true opinions or attitudes do or do not become apparent to the perceiver.

References

ABC News. (1986, July 27). World news tonight with Peter Jennings. New York: American Broadcasting Corporation.

ABC News. (1987, August 18). Nightline. Washington, DC: American Broadcasting Corporation.

Bishop, Y. M. M., Feinberg, S. E., & Holland, P. W. (1975). Discrete multivariate analysis. Cambridge, MA: MIT Press.

Cantor, N., & Kihlstrom, J. (1987). Personality and social intelligence. Englewood Cliffs, NJ: Prentice-Hall.

Cantor, N., & Mischel, W. (1977). Traits as prototypes: Effects on recog- nition memory. Journal of Personality and Social Psychology, 35, 38- 48.

Clark, H. H. (1985). Language use and language users. In G. Lindzey & E. Aronson (Eds.), The handbook of social psychology (3rd ed., Vol. 2, pp. 179-231). New York: Random House.

Clark, H. H., & Carlson, T. B. (1981). Context for comprehension. In J. Long & A. D. Baddeley (Eds.), Attention and performance IX (pp. 313-330). Hillsdale, NJ: Erlbaum.

Clark, H. H., & Carlson, T. B. (1982a). Hearers and speech acts. Lan- guage, 58, 332-373.

Clark, H. H., & Carlson, T. B. (1982b). Speech acts and hearers' beliefs. In N. V. Smith (Ed.), Mutual knowledge (pp. 1-36). Orlando, FL: Academic Press.

Clark, H. H., & Marshall, C. R. (1981). Definite reference and mutual knowledge. In A. K. Joshi, B. L. Webber, & I. A. Sag (Eds.), Elements

of discourse and understanding (pp. 10-63). Cambridge, England: Cambridge University Press.

Clark, H. H., & Schaefer, E. F. (1987). Concealing one's meaning from overhearers. Journal of Memory and Language, 26, 209-225.

Cohen, J., & Cohen, P. (1983). Applied multiple regression/correlation analysis for the behavioral sciences (2nd ed.). Hillsdale, NJ: Erlbaum.

DePaulo, B. M., Stone, J. I., & Lassiter, G. D. (1985). Deceiving and detecting deceit. In B. R. Schlenker (Ed.), The self and social life (pp. 323-370). New York: McGraw-Hill.

DePaulo, B. M., Zuckerman, M., & Rosenthal, R. (1980a). Humans as \iz detectors. Journal of Communication, 30, 129-139.

DePaulo, B. M., Zuckerman, M., & Rosenthal, R. (1980b). Detecting deception: Modality effects. In L. Wheeler (Ed.), Review of personal- ity and social psychology (Vol. 1, pp. 125-162). Beverly Hills, CA: Sage.

Ekman, P. (1985). Telling lies: Clues to deceit in the marketplace, poli- tics, and marriage. New York: Norton.

Ekman, P., & Friesen, W. V. (1969). Nonverbal leakage and clues to deception. Psychiatry, 32. 88-106.

Ekman, P., & Friesen, W. V. (1974). Detecting deception from the body and the face. Journal of Personality and Social Psychology, 29, 288- 298.

Fein, S., Hilton, J. L., & Miller, D. T. (in press). Suspicion of ulterior motivation and correspondence bias. Journal of Personality and So- cial Psychology.

Fiske, S. T, & Taylor, S. E. (1984). Social cognition. Reading, MA: Ad- dison-Wesley.

Fleming, J. H., & Darley, J. M. (1989). Mixed messages: The multiple audience problem and strategic social communication. Unpublished manuscript, University of Minnesota, Minneapolis.

Fleming, J. H., & Kraut, R. E. (1988). Not at face value: Suspicion- induced discounting and correspondence bias in interpreting facial displays of emotion. Unpublished manuscript, University of Minne- sota, Minneapolis.

Gilbert, D. T, & Jones, E. E. (1986). Perceiver-induced constraint: In- terpretations of self-generated reality. Journal of Personality and So- cial Psychology, 50, 269-280.

Goffman, E. (1959). The presentation of self in everyday life. Garden City, NY: Doubleday.

Goffman, E. (1961). Encounters. Indianapolis, IN: Bobbs-Merrill. Goffman, E. (1969). Strategic interaction. Philadelphia: University of

Pennsylvania Press. Higgins, E. T. (1981). The "communication game": Implications for

social cognition and persuasion. In E. T. Higgins, C. P. Herman, & M. P. Zanna (Eds.), Social cognition: The Ontario symposium (Vol. 1, pp. 343-392). Hillsdale, NJ: Erlbaum.

Hilton, J. L., Miller, D. T, Fein, S., & Darley, J. M. (1989). When dispo- sitional inferences are suspended: Diagnosing and calibrating traits. Unpublished manuscript, University of Michigan, Ann Arbor.

Hyman, R. (1989). The psychology of deception. Annual Review oj Psy- chology, 40, 133-154.

Jones, E. E. (1979). The rocky road from acts to dispositions. American Psychologist, 34, 107-117.

Jones, E. E. (1986). Interpreting interpersonal behavior: The effects of expectancies. Science, 234, 41-46.

Jones, E. E., & Harris, V. A. (1967). The attribution of attitudes. Jour- nal of Experimental Social Psychology, 3, 1-24.

Jones, E. E., & Nisbett, R. E. (1972). The actor and the observer: Diver- gent perceptions of the causes of behavior. In E. E. Jones, D. E. Ka- nouse, H. H. Kelley, R. E. Nisbett, S. Valins, & B. Weiner (Eds.), Attribution: Perceiving the causes of behavior (pp. 79-94). Morris- town, NJ: General Learning Press.

Jones, E. E., & Pittman, T. S. (1982). Toward a general theory of strate- gic self-presentation. In J. Suls (Ed.), Psychological perspectives on the self (Vol. 1, pp. 231-262). Hillsdale, NJ: Erlbaum.

Jones, E. E., Worchel, S., Goethals, G. R., & Grumet, J. (1971). Prior

MULTIPLE AUDIENCE PROBLEM 609

expectancy and behavioral extremity as determinants of attitude at- tribution. Journal of Experimental Social Psychology, 7, 59-80.

Kelley, H. H. (1972). Attribution in social interaction. In E. E. Jones, D. E. Kanouse, H. H. Kelley, R. E., Nisbett, S. Valins, & B. Weiner (Eds.), Attribution: Perceiving the causes of behavior (pp. 1 -26). Mor- ristown, NJ: General Learning Press.

Kenny, D. A. (1985). Quantitative methods for social psychology. In G. Lindzey & E. Aronson (Eds.), Handbook of social psychology (Vol. 1, pp. 487-508). New York: Random House.

Knapp, M. L., & Comadena, M. E. (1979). Telling it like it isn't: A review of theory and research on deceptive communications. Human Communication Research, 5, 270-285.

Krauss, R. M. (1981). Impression formation, impression management, and nonverbal behaviors. In E. T. Higgins, C. P. Herman, & M. P. Zanna (Eds.), Social cognition: The Ontario symposium (Vol. 1, pp. 323-341). Hillsdale, NJ: Erlbaum.

Kraut, R. E. (1978). Verbal and nonverbal cues in the perception of lying. Journal of Personality and Social Psychology, 36, 380-391.

Kraut, R. E. (1980). Humans as lie detectors: Some second thoughts. Journal of Communication, 8, 209-216.

Kraut, R. E., & Higgins, E. T. (1984). Communication and social cogni- tion. In R. S. Wyer, Jr., & T. K. Srull (Eds.), Handbook of social cogni- tion (Vol. 3, pp. 87-127). Hillsdale, NJ: Erlbaum.

Lewis, D. K. (1969). Convention. Cambridge, MA: Harvard University Press.

Miller, A. G., Jones, E. E., & Hinkle, S. (1981). A robust attribution error in the personality domain. Journal of Experimental Social Psy- chology, 17, 587-600.

Miller, G. R., & Burgoon, J. K. (1981). Factors affecting assessments of witness credibility. In R. Bray & N. Kerr (Eds.), The psychology of the courtroom (pp. 169-194). Orlando, FL: Academic Press.

Nisbett, R. E., & Ross, L. (1980). Human inference: Strategies and shortcomings of social judgment. Englewood Cliffs, NJ: Prentice- Hall.

Pyszczynski, T. A., & Greenberg, J. (1981). Role of disconfirmed expec- tancies in the instigation of attributional processing. Journal of Per- sonality and Social Psychology, 40, 31-38.

Quattrone, G. A. (1982). Overattribution and unit formation: When behavior engulfs the person. Journal of Personality and Social Psy- chology, 42, 593-607.

Rosenthal, R. (1978). Combining results of independent studies. Psy- chological Bulletin, 85, 185-193.

Ross, L. (1977). The intuitive psychologist and his shortcomings. In L.

Berkowitz (Ed.), Advances in experimental social psychology (Vol. 10, pp. 173-220). Orlando, FL: Academic Press.

Schein, E. H. (1956). The Chinese indoctrination program for prisoners of war: A study of attempted "brainwashing." Psychiatry, 19, 149- 172.

Schein, E. H., Schneier, I., & Barker, C. H. (1961). Coercive persua- sion—A socio-psychological analysis of the "brainwashing" of Ameri- can civilian prisoners by Chinese Communists. New York: Norton.

Schiffer, S. (1972). Meaning. Oxford, England: Clarendon Press. Schlenker, B. R. (1985). The self and social life. New York: McGraw-

Hill. Schneider, K. J., & Miller, R. S. (1975). The effects of enthusiasm and

quality of arguments on attitude attribution. Journal of Personality, 43, 693-708.

Snyder, M. L., & Jones, E. E. (1974). Attitude attribution when behavior is constrained. Journal of Experimental Social Psychology, 10, 585- 600.

Sternberg, R. J. (1986). Beyond IQ: A triarchic theory of human intelli- gence. New York: Cambridge University Press.

Tedeschi, J. T. (1981). Impression management theory and social psy- chological research. Orlando, FL: Academic Press.

Toris, C., & DePaulo, B. M. (1984). Effects of actual deception and sus- piciousness of deception on interpersonal perceptions. Journal of Per- sonality and Social Psychology, 47, 1063-1073.

Zuckerman, M., Amidon, M. D., Bishop, S. E., & Pomerantz, S. D. (1982). Face and tone of voice in the communication of deception. Journal of Personality and Social Psychology, 43, 347-357.

Zuckerman, M., DeFrank, R. S., Hall, J. A., Larrance, D. T, & Rosen- thai, R. (1979). Facial and vocal cues of deception and honesty. Jour- nal of Experimental Social Psychology, 15, 378-396.

Zuckerman, M., DePaulo, B. M., & Rosenthal, R. (1981). Verbal and nonverbal communication of deception. In L. Berkowitz (Ed.), Ad- vances in experimental social psychology (Vol. 14, pp. 1-59). Or- lando, FL: Academic Press.

Zuckerman, M., DePaulo, B. M., & Rosenthal, R. (1986). Humans as deceivers and lie detectors. In P. D. Blanck, R. Buck, & R. Rosenthal (Eds.), Nonverbal communication in the clinical context (pp. 13-35). University Park: Pennsylvania State University Press.

Zuckerman, M., Koestner, R., Colella, M. J., & Alton, A. O. (1984). Anchoring in the detection of deception and leakage. Journal of Per- sonality and Social Psychology, 4 7, 301 -311.

Received June 27, 1984 Revision received November 15, 1989

Accepted December 5,1989 •