SPEAKING FROM IGNORANCE 223 Table f and Percentages per Participant try,and not enou to succeed in reading the targe for Experiments la and Ib Disagreements Experiment la Experiment Ib ipated in Experim 142 edropped for not follo ing 7% comments indicating their igne 283%) 3.7% y (cg What ar e there were slight variation 90 ve affected answers 3.13 give a word that ght was "th wer that you can give,"effee tively repeating th Results mploved for this analysis.Consistent with our hypothesis.the umber of Manipulation and suspicion checks. Only one p nt ir gave ss al We retain d he n analyses since her d )137()1000.3 nts.Exp imen 2).When Discussion asked to dese indicate Parti are that es- ne EXp nd Ib The siz hat it fa It like to b 70 riment 1b) r that it was"frustrating”(or wrong answers effect in Asch-like ituati a(e.."looked for intelligible ette then supposed what ve expectation of agreement with eing and agre ers.The firs partic tions.E n normative expectati The for the two exp ith A and Be ansy ve deserihed them e than six time nany people to see what they would predict.Invariably.they vere introduced in later studies to provide more strineent asses e statistical procedure we a vard:Sh d the nat A and as an ov ersed ut likely to be in orrect?As ted,the mos umber ofesero-inflated negative binomial regression was common solution was always to agree with the answers of b bette
although not necessarily within the word that was to be identified. In short, there was just enough information to encourage participants at C to try, and not enough to succeed in, reading the target word. Participants and replication. Twenty-seven undergraduate students at a private liberal arts college participated in Experiment 1a, receiving extra credit in a course or $5 for their participation. The same was true in Experiment 1b, although three participants were dropped for not following instructions. Ten participants in Experiment 1a asked questions or offered spontaneous comments indicating their ignorance and/or uncertainty (e.g., “I can’t see. What am I supposed to do?”) during one of the early trials. Experimenters tried to answer these questions in a helpful, noncommittal way, but since there were slight variations in responding that might have affected participants’ answers, we replicated the experiment, standardizing the experimenter’s response. Thus, in Experiment 1b, if participants in Position C made spontaneous comments about their task or asked questions, the experimenter instructed them to give a single word that they thought was “the most correct answer that you can give,” effectively repeating the original instructions. Results Manipulation and suspicion checks. Only one participant in either experiment gave any evidence of suspecting that Participants A and B were confederates, and she agreed on all critical trials. We retained her in analyses since her responses did not provide support for our hypothesis. Participants believed that A and B had answered all or very nearly all of the questions correctly (M 17.6 out of 18) and that they themselves had answered many fewer questions correctly (M 7.8 for Experiment 1, M 6.3 for Experiment 2). When asked to describe their task in the experiment, participants indicated that that it was to give the correct answer, with two commenting on the difficulty of the task. Participants were well aware that they “could see little or nothing” at Position C (M 16.5 out of 18 items in Experiment 2). When asked what it felt like to be unable to see words clearly, 79% (Experiment 1b) wrote that it was “frustrating” (or something similar), often indicating their dependence on A and B. Others described straightforwardly the strategy they used to solve the dilemma (e.g., “I looked for intelligible letters, then supposed what the word might be”). Disagreeing and agreeing with correct answers. The first and second columns of Table 1 show the distribution of participants’ disagreeing answers for Experiments 1a and 1b, respectively. The distribution was similar for the two experiments: In both, approximately half of all participants always agreed with A and B, a quarter disagreed one to six times, and a quarter disagreed more than six times. To test for an SFI effect, we compared observed disagreements on the 12 critical items (i.e., those that A and B had answered correctly prior to C giving an answer) against an expected value of zero. In addition to this simple standard, experimental controls were introduced in later studies to provide more stringent assessments. The statistical procedure we adopted to test for the presence of an SFI effect was very conservative. Because number of disagreements was an overdispersed count variable with a large number of zeros, zero-inflated negative binomial regression was employed for this analysis. Consistent with our hypothesis, the number of disagreements on critical trials differed significantly from zero (in Experiment 1a: b 1.84, Wald 2 29.70, p .001; in Experiment 1b: b 1.79, Wald 2 50.41, p .001). Disagreeing answers were distributed relatively evenly across all 12 critical trials; there was no relation between order of trials and number of agreements, Experiment 1a: Q(11) 10.00, p .53; Experiment 1b: Q(11) 11.37, p .41. Discussion Two exploratory studies provided clear evidence for an SFI effect with 29% (94 out of 324) and 26% (75 out of 288) of responses disagreeing with the correct answers of better positioned others in Experiments 1a and 1b, respectively. The size of this effect is comparable to the average size of the agreeing-withwrong answers effect in Asch-like situations (about 25%), based on results of meta-analysis (Bond & Smith, 1996), although in Asch’s (1951) initial study, 34% answered incorrectly. This is true despite the fact that the normative expectation of agreement with correct answers in the SFI situation is stronger than is the expectation of disagreement with incorrect answers in the Asch-type situations. Evidence for this difference in normative expectations is found in the fact that over half of participants always agreed with A and B’s answers on critical trials, while about a quarter of Asch’s participants always disagreed. Prior to running these experiments, we described them carefully to many people to see what they would predict. Invariably, they thought the experiments hardly worth running because obviously people would always agree with their more knowledgeable peers. Intuitively, it does not seem like there should be any conflict in an SFI situation, unlike the Asch dilemma. Nevertheless, results indicated that participants found the situation frustrating and awkward: Should they repeat what A and B have said, or should they make up a word that is theirs, one unique to their position and perspective but likely to be incorrect? As expected, the most common solution was always to agree with the answers of better Table 1 Distribution of Disagreements and Percentages per Participant for Experiments 1a and 1b Disagreements Experiment 1a Experiment 1b 0 14 (51.9%) 12 (50.0%) 1 4 (14.8%) 3 (12.5%) 2 0 1 (4.2%) 30 0 4 1 (3.7%) 0 5 0 2 (8.3%) 6 1 (3.7%) 0 70 0 8 0 1 (4.2%) 9 0 1 (4.2%) 10 1 (3.7%) 1 (4.2%) 11 2 (7.4%) 3 (12.5%) 12 4 (14.8%) 0 M 3.48 3.13 Note. Means are calculated by dividing total number of disagreements by total number of critical trials and multiplying by the number of critical trials per participant (i.e., 12). This document is copyrighted by the American Psychological Association or one of its allied publishers. This article is intended solely for the personal use of the individual user and is not to be disseminated broadly. SPEAKING FROM IGNORANCE 223
224 HODGES.MEAGHER.NORTON.McBAIN.AND SROUBEK Whsitioncd others.Howevercable Sl efectoccd one epB Predictions First,in addition to hypotheses t sted earlier (i.e.an SFl effect Experiment 2 was designed to begin to address these issues. sat Position 6(Not see)than Position 5(See).If this Experiment 2 h nd,there evidence of the attr rious alternative hypoth two important not in placed not only in s.social solidarity.and pragmatics in ways that indicated n a p tion of knowled hich hee con for Participant A.B. or C (real p with C ment than motivations to act on me were not informed i of self- ort mea s that as o the dem ands of speakin and cally thar ing wi AeFree On the othe hand, we did o that ther oul 、D agreers do not care about other part allowed us to test sever al altemative hypot cial so tly.we mig H se they are speaking truthfullyfro d persp n eople want to distinguish th nd a pr d tha s of pe he fact that hird possibility is that those people who lisagree with corre thers.Fifth. 1981).A fourth pe sibility is that the in these term Furthermore,we expected that Agreer ackno hen that an ons (OmeWhit 2000)A fift ndicating anger about their position would not be correlated with d be nc ant s who then act in ith th pants are ry at the Method and byn Participant ght be was omitted from analyses due to a procedural led he and I anc stud &H e of th were addre ed b yincluding tw ware of the nc however.were addressed by comparing the number of disagrecing that they were not aware of how many people were expected at
positioned others. However, a sizeable SFI effect occurred, one which cannot be explained by a lack of trust in others or a misperception of one’s own (lack of) knowledge. Participants knowingly disagreed with correct answers. Although Experiments 1a and 1b indicated that an SFI effect can occur and that it is likely not due to mistrust or inattention, they did not provide positive evidence in favor of a values-pragmatics account, nor did they consider possible alternative explanations. Experiment 2 was designed to begin to address these issues. Experiment 2 The focus of Experiment 2 was to evaluate evidence that could provide clearer support for a values-pragmatics account of the SFI effect and that could begin to rule out various alternative hypotheses that might explain why some people some of the time choose to disagree with correct answers rather than conforming. The design of Experiment 2 included two important features not included in the earlier experiments. One was the addition of a new control condition: Participants were placed not only in a position of ignorance but also in a position of knowledge in which they could see, as well as hear, the correct answer. We set up six positions for Participant A, B, or C (real participant), with C alternating between Position 5, in which the participant could see clearly, and Position 6, in which he or she could not (see Figure 1). Participants were not informed in advance that they would be placed in both positions. A second feature was the addition of a series of self-report measures that assessed various possible motivations for how participants answered or other perceptions or feelings that might contribute to participants disagreeing with their peers. These two features allowed us to test several alternative hypotheses for why an SFI effect occurs. One possibility is that people do not want to look like copycats, merely repeating what others have said. That is, perhaps disagreeing answers are refusals to imitate others (Wheeler & Arrowood, 1966). Another possibility is that people want to distinguish themselves from others, and answering correctly does not do that in this context (Brewer & Roccas, 2001; Griskevicius, Goldstein, Mortensen, Cialdini, & Kenrick, 2006). A third possibility is that those people who disagree with correct answers are simply more independent than those who always agree (Perrin & Spencer, 1981). A fourth possibility is that the experimental situation somehow invites or demands that people give differing answers when they have been placed in an odd position and asked to answer questions (Orne & Whitehouse, 2000). A fifth possibility is that experimenters expect people to disagree and somehow subtly and nonconsciously signal this expectation to participants who then act in accordance with those signals (Hazelrigg, Cooper, & Strathman, 1991). A sixth possibility is that participants are angry at the experimenter for asking them impossible questions and they respond aggressively by giving incorrect answers (Brehm & Brehm, 1981). A seventh possibility, which might be framed in attribution terms, is that participants think the experimenter knows they cannot see and is testing them to see if they will acknowledge it (Ross, Bierbrauer, & Hoffman, 1976). Some of these alternatives were addressed by including relevant measures in the postidentification task (e.g., how angry did participants say they felt while in the position of ignorance?). Others, however, were addressed by comparing the number of disagreeing answers given in the position of ignorance with those given in the position of knowledge (e.g., are participants refusing to imitate?). Predictions First, in addition to hypotheses tested earlier (i.e., an SFI effect, despite trusting others’ answers, and experiencing the situation as a dilemma), we predicted that there would be significantly more disagreements at Position 6 (Not see) than Position 5 (See). If this occurs it would undercut alternative hypotheses mentioned earlier, such as refusing to imitate, experimenter expectancy, wanting to appear distinctive, and demand characteristics. Second, there should be some evidence of the attraction of truthfulness, social solidarity, and pragmatics on most if not all participants in a position of ignorance, not just those who choose to give disagreeing answers one or more times. Thus, we predicted that all participants, both those who always chose to agree with A and B (Agreers) and those who sometimes chose to disagree (Disagreers), would answer items directed at concerns for truthfulness, social solidarity, and pragmatics in ways that indicated their actions were constrained by values-realizing dynamics. Indexes of truthfulness, social solidarity, and pragmatics were expected to receive stronger endorsement than motivations to act agreeably or answer independently. Third, we expected that Disagreers would score higher on measures of truthfulness and pragmatics than Agreers. Disagreers, we believed, would likely show greater sensitivity in an SFI situation to the demands of speaking truthfully and pragmatically than Agreers. On the other hand, we did not expect that there would be a difference in expressions of social solidarity between Agreers and Disagreers. If Disagreers do not care about other participants and are simply acting independently, we might expect expressions of social solidarity to be less than for Agreers. However, if they disagree primarily because they are speaking truthfully from their assigned perspective, then there is no reason to expect that they would express significantly less social solidarity than Agreers. Fourth, we predicted that measures of agreement and independence would be relatively weaker than other motivations, despite the fact that participants would probably often choose to agree with others. Fifth, we expected that concerns to answer correctly would be expressed by virtually all participants since the task was defined in these terms. Furthermore, we expected that Agreers would be more likely than Disagreers to acknowledge correctness as a motivation for their actions. Finally, we predicted that an item indicating anger about their position would not be correlated with disagreeing answers at Position 6 and that there would be no difference between Disagreers and Agreers in this regard. Method Participants. Forty-seven undergraduate participants at a private liberal arts college completed the experiment for extra credit. One participant was omitted from analyses due to a procedural error, leaving 36 women and 10 men in the study. Procedure. In order to test for possible experimenter expectancy effects, two experimenters unaware of the previous studies, including their rationale and results, were used, while two others were aware. Participants were recruited by e-mail solicitations so that they were not aware of how many people were expected at This document is copyrighted by the American Psychological Association or one of its allied publishers. This article is intended solely for the personal use of the individual user and is not to be disseminated broadly. 224 HODGES, MEAGHER, NORTON, MCBAIN, AND SROUBEK