Peer Influence on Conformity and Confidence in a Perceptual Judgment Task

Undergraduate college students were presented with two arrays of dots varying in numerosity on a computer screen and asked to indicate if the arrays differed in number. They also rated their level of confidence in their responses. Trials varied in difficulty based on the size of the arrays. On half of the trials, participants were shown the ostensible responses of confederates to test the effect of peer influence on numerosity judgments and participant confidence. On the other half of the trials, participants received no information about the responses of the confederates to provide a measure of baseline performance. Higher levels of conformity were observed for the difficult trials, on which participants were both less accurate and less confident. However, confidence ratings were influenced by peer judgments for easy trials but not for difficult trials. These data suggest that task difficulty influences conformity when making perceptual judgments.

PSIHOLOGIJA, 2020, Vol. 53(1), 101-113 that social conformity is pervasive and has implications for a wide variety of behaviors (e.g., purchasing expensive products, using illegal drugs; see Cialdini & Goldstein, 2004, for a review). The present study examined the influence of peer pressure on participants' perceptual judgments and confidence in their judgments. Our goal was to demonstrate that pressure to conform resulting from the presentation of incorrect information impacts cognition. For example, peer influence may affect task performance by increasing the uncertainty of both decision-making and perceptual functioning. What follows is a review of the history of conformity research with special emphasis on a variety of perceptual and cognitive tasks that have been used to show the effects of conformity. Sherif (1937) performed one of the first studies on perception and attitudes in relation to social influences. The participants' task was to perceive how many inches a dot on the wall had moved from its original location. The stimulus was a small spot of light presented at a constant distance in a completely dark room. The lack of visual reference created the illusory experience of autokinesis (Adams, 1912) whereby the observer typically reported the light as moving rather than being stationary. The confederates in each experiment knew the norm and the range, and tried to influence the naïve participant toward that norm or another norm that was arbitrarily larger or smaller. The results of that study revealed that participants displayed a tendency to answer in accordance with group norms. A few decades later Asch (1951) found that one or two confederates had relatively little influence on conformity whereas three or more had the power to lead participants to conform. Each participant was placed in a group and asked to name out loud which of three comparison lines was the same length as a standard line. The other members of the group were confederates in the experiment and often gave incorrect responses in an attempt to influence the participant to conform. Asch discovered that these attempts were successful about one third of the time.
Conformity is a multifaceted construct that appears in at least two forms: informational and normative conformity (Cialdini & Goldstein, 2004;Deutsch & Gerard, 1955). Informational conformity arises when perceivers base their judgments on evidence provided by others. It is important to note that, in these situations, the evidence is credible, but not necessarily verified (e.g., estimation tasks where the correct response is not divulged). Normative conformity is based on the desire of the individual to gain acceptance into a group. Sherif's (1937) experiment represents an example of informational conformity as elicited by the task used in the study (participants estimate, but do not know the correct response), while Asch's (1951) study demonstrates pure effects of peer pressure (as it is trivially easy to answer which line is longer/shorter). In addition to the distinction between normative and informational conformity, there are numerous variables that moderate the level of conformity, such as achievement motivation and affiliative motivation (Sistrunk & McDavid, 1965). The distinctions between different types of conformity and motivation were not the focus of the present study; therefore, we refer to social influence as conformity throughout the paper. PSIHOLOGIJA, 2020, Vol. 53(1), 101-113 Berns et al. (2005) extended Asch's studies by using a mental rotation task (Shepard & Metzler, 1971) and adding functional magnetic resonance imaging (fMRI) to measure brain activity. Conformity was induced either by a peer group or by a computer. Under both conditions, the influence of peers was shown to activate various parts of the brain involved in perception and decision making. Berns et al. proposed that, not only would individuals sometimes conform when they were aware that their peers had provided incorrect information, but that the act of conforming may also cause a genuine change in conscious perception.
Confidence judgments -defined as the conscious assessment of one's own task competence -were used to complement the standard measure of conformity in the present study (i.e., agreement with incorrect responses by confederates). Confidence judgments have typically been employed in perception and action research to distinguish between cognitive and perceptual modes of apprehension (Runeson & Andersson, 2004;Runeson, Juslin, & Olsson, 2000), indicate task difficulty, or identify transition points between action modes (Kinsella-Shaw, Shaw, & Turvey, 1992;Warren, 1984), such as the transition from walking to climbing up a steep hill, or the transition from walking to running. Runes on et al. demonstrated that two types of errors exist in judgments: Thurstonian errors that refer to "inconsistent use of perfect information" and Brunswikian errors that emerge from "consistent use of information with incomplete validity" (Runeson et al., 2000, p. 532). Thurstonian errors are considered to be perceptual in nature (relating to "neural noise"), whereas Brunswikian errors are due to the consistent use of incorrect heuristics or rules. The latter mode is characteristic of higher-level cognitive decision making; the former is more typical of sensory processes. In addition, Runeson's theory predicts (and empirically confirms) that underconfidence (being less confident than the accuracy of the performance would dictate) is associated with being in a perceptual mode, and overconfidence (being more confident than is warranted by performance) is associated with cognitive mode of apprehension. In the context of the current study, levels of confidence would indicate whether one engages the task at the cognitive level (decision making and rule-based heuristics) or at the perceptual level (focused on fine sensory discrimination of stimuli). This way, researchers would be able to discover which mode of apprehension is more prone to the effects of conformity.
Confidence has been used as a measure of conformity based on manipulating beliefs about the quality of one's memory (Gabbert, Memon, & Wright, 2007), a high level cognitive process, and also in online studies (Rosander & Eriksson, 2012), but no study to date has combined using a visual perceptual task with online fake confederates. In addition, Rosander and Eriksson predicted that conformity would increase with task difficulty in both perceptual and cognitive tasks. In the current study we addressed the effects of task difficulty on conformity and confidence measures. Consistent with the results of previous studies (e.g., Baron et al., 1996), we hypothesized that people should be more inclined to conform to the responses of confederates on more difficult tasks. In addition, they should be more confident in their own judgments and responses when they are not unduly influenced by the incorrect responses of peers and when the task is less difficult. PSIHOLOGIJA, 2020, Vol. 53(1), 101-113 The inclusion of confidence judgments is important because such judgments can indicate whether participants truly believe the confederates (i.e., acceptance) or are simply conforming even when they are certain their response is wrong (i.e., compliance). Following the predictions of Runeson et al. (2000), those who are more confident in their perceptual judgment should be more susceptible to normative influence, whereas those who are less confident should be more susceptible to informational influence.
Task difficulty can be easily manipulated in numerical discrimination tasks by varying quantifiable dimensions of the task such as the number of stimuli. Numerosity has been manipulated in conformity tasks by Wolosin, Sherman, and Mynatt (1972) who asked participants to make numerosity judgments in response to auditory tones. Participants were shown to conform to bogus responses by confederates when their results were compared to those of the control group. This study offered early evidence that performance in a numerosity task is affected by conformity. Our experiment employed a visual version of the auditory numerosity paradigm used by Wolosin et al. (1972). The visual numerosity task (see Durgin, 1995 for additional information) is well suited for the measurement of conformity because participants typically consider it to be a cognitive or perceptual task devoid of any obvious social component. As such, it is less likely to raise awareness of the conformity manipulation precisely because participants typically believe that their task performance is the direct result of perceptual and cognitive factors rather than social ones. We predicted that incorrect peer responses would induce conformity and that this effect would be modulated by changes in task difficulty. Conformity has not been measured previously in a visual numerosity task, with task difficulty as a factor in performance, and ours is one of a small number of studies that demonstrate the influence of virtual confederates (e.g., Guadagno, Blascovich, Bailenson, & McCall, 2007;Reysen, 2005).

Method
Participants made perceptual judgments and evaluated their own confidence levels in a perceptual numerosity judgment task. On each trial, participants had to compare the number of dots that appeared in two arrays and provide a response as to whether the number of dots in the arrays were the same or different. They were led to believe that other students were participating in the experiment at the same time and that the responses of the other participants would be visible to them on a portion of the trials. We expected naïve participants to conform to responses of the confederates and that this conformity would increase as the perceptual task became more difficult. We also expected the confidence levels of the participants to decrease when peers provided incorrect responses and the trials became more difficult.

Participants
Participants for this study were 43 (7 males, 36 females) undergraduate psychology students (average age = 21.5 years, SD = 7.1 years) at a large public university in the United States. Participants were enrolled in psychology courses and participation counted toward fulfilling a course requirement. Participants signed an informed consent form before participation. All procedures were approved by the local Institutional Review Board ensuring ethical principles were followed during the conduct of the study. PSIHOLOGIJA, 2020, Vol. 53(1), 101-113

Materials
The experiment was programmed in Inquisit (Seattle, WA: Millisecond Software) and presented on an IBM desktop computer. Stimuli consisted of 30 arrays of white dots on a black background as described in Table 1. A trial was categorized as difficult if there were 10 or more dots in either panel (trials were categorized as easy if both panels contained fewer than 10 dots). We had 12 easy trials and 18 difficult trials per block. Within each type of trial (easy vs. difficult), half of the trials were conformity trials and the other half were not conformity trials (see Table 1). We based our determination of task difficulty on Durgin (1995) who claimed that, after a trivial subitization range (up to 6 dots), the numerosity task becomes more difficult due to limited cognitive resources available to count stimuli (Burr, Turi, & Anobile, 2010;Jevons, 1870;Kaufman, Lord, Reese, & Volkmann, 1949). Our a priori categorization based on past research was confirmed by rank ordering each trial based on accuracy. This analysis revealed that the most accurate 11 trial types (exceeding 90% accuracy) contained less than 10 dots in both panels, except for one trial (5 versus 5 dots).

Procedure
An elaborate cover story was constructed concerning the ostensible goals of the study. The participants were told that they were participating in a distance communications study with other participants who were located in several different laboratories across campus. The other "participants" were in fact photographs drawn from a database that reflected the demographic characteristics of students at the university. These images were displayed in a single column on the right side of the computer screen with the actual participant always occupying the sixth position out of the seven available positions. The participant did not see their own picture but saw a box containing the word "YOU" to indicate where the other participants would purportedly see their picture. This setup is shown in Figure 1. The participants were told that they would see the responses of the first five participants as they answered each question on some trials but not others. That is, the actual participant was under the impression that he or she was seeing the responses of the other participants on only half of the trials.
Trials on which confederates' responses appeared (Conformity condition) occurred within a 30-trial block, whereas trials on which confederate responses did not appear (Control condition) occurred within a different 30-trial block. The order in which Conformity and Control condition trials were presented was counterbalanced across participants such that half of the participants received the Conformity condition first and half received the Control condition first. The Control condition allowed us to determine the unbiased accuracy of the dot array comparisons and compare the confidence judgments that were provided either with or without peer pressure.
All confederates provided the same wrong response on any given conformity trial which created the circumstances conducive to conformity. It is important to note that 50% of the trials for which participants were able to see the responses of the confederates were correct in order to reduce the suspicion of the participants (see Berns et al., 2005 for a similar approach). Only the trials showing the incorrect answer from the confederates were used in the calculation of conformity levels. It was not possible to use the trials with the correct answer given for this purpose because there was no way to determine whether the participant simply conformed to the responses provided by the confederates or provided the correct answer because they actually knew it was the correct answer.
The program displayed 60 pairs of dot arrays that were paired in such a way as to systematically manipulate task difficulty in terms of the difference in numerosity between the arrays, as well as the absolute number of dots depicted in the arrays. An example of a dot array is displayed in Figure 1. Each pair of dot arrays was displayed for one second. After each pair of arrays was displayed, the participant was asked to determine whether the two arrays contained the same number of dots or not. After selecting "yes" or "no", they rated their confidence on a scale from 0% (not confident at all) to 100% (absolutely confident). Figure 1. Screenshots of the experimental sequence including the dot arrays and the list of alleged participants with photographs. The word "YOU" represents the actual participant.

Experimental Design and Data Analysis
The dependent measures in the perceptual task were accuracy, confidence level, and level of agreement with other people's answers (expressed as a proportion of times an agreement was observed). The difficulty of the perceptual task was defined as easy (both arrays contained less than 10 dots each) or difficult (at least one array contained 10 or more dots). Our rationale for this distinction was that the difficulty of the task should increase as the arrays become larger (more dots) and the difference between the arrays becomes smaller (fewer dots separating the two arrays). There is evidence to support both elements of our rationale separately (Durgin, 1995;Green & Bavelier, 2003) but a systematic empirical test of the independent effects of both parameters in a single study awaits future research. Repeated measures analyses of variance (ANOVA) were conducted on confidence levels and accuracy PSIHOLOGIJA, 2020, Vol. 53(1), 101-113 as a function of Task Difficulty (easy vs. difficult) and Conformity Condition (incorrect peer response vs. correct peer response vs. no response). Conformity trials were defined as those trials on which participants saw incorrect peer responses. A related samples t-test was used to compare conformity levels in easy versus difficult trials.
Debriefing took place at or after the end of the experiment. The debriefing included explanations about why deception was necessary for conducting this experiment. None of the participants reported being aware of the conformity manipulation.

Accuracy
The 2 (Task Difficulty: easy vs. difficult) × 3 (Conformity Condition: incorrect response vs. correct response vs. no response) repeated measures ANOVA for accuracy resulted in a main effect of Task Difficulty, F(1, 42) = 146.2, p < .001, η p 2 = .78, revealing that participants were more accurate on easy trials (M = 0.83, SD = 0.22) than difficult trials (M = 0.57, SD = 0.18). The main effect of Conformity Condition was also statistically significant, F(2, 84) = 23.08, p < .001, η p 2 =.36. Post hoc comparisons using a Bonferroni correction revealed that participants were significantly less accurate when provided with incorrect peer responses than when provided with correct or no peer responses. The post hoc analysis did not detect any difference in accuracy between the no peer response and correct peer response conditions. The Task Difficulty × Conformity Condition interaction was significant, F(2, 84) = 9.95, p < .001, η p 2 = .19, revealing that the biggest difference between easy and difficult trials occurred in the no peer response condition. Post hoc tests also showed that correct and incorrect peer response did not distinguish performance on easy versus difficult trials. The results are presented in Figure 2.

Proportion of Conforming Responses
Because only incorrect peer response trials were included in the calculation of conformity, we used a repeated measures t-test to analyze differences in proportion of conformed responses among easy versus difficult trials. A dependent samples t-test for the proportion of conforming responses was statistically significant, t(42) = 12.74, p < .001, Cohen's d = 1.91, revealing that participants showed less conformity on easy trials (M = 0.10, SD = 0.17) than on difficult trials (M = 0.49, SD = 0.20). The results are presented in Figure 3.

Confidence Judgments
The 2 (Task Difficulty: easy vs. difficult) × 3 (Conformity Condition: incorrect response vs. correct response vs. no response) repeated measures ANOVA for confidence judgments resulted in a main effect of Task Difficulty, F(1, 42) = 88.17, p < .001, η p 2 = .68, revealing that participants were more confident on easy trials (M = 91.3%, SD = 10.4%) than on difficult trials (M = 61.6%, SD = 22.4%). This effect was qualified by a Task Difficulty × Conformity Condition interaction, F(2, 84) = 4.5, p < .014, η p 2 = .10. Post hoc comparisons showed that the difference between confidence on easy and difficult trials was significantly greater in the no peer response condition (ΔM = 31.7%, SD = 20.2%) as compared to the incorrect peer response condition (ΔM = 27.2%, SD = 20.9%). No other comparisons or main effects were significant. The interaction is presented in Figure 4. PSIHOLOGIJA, 2020, Vol. 53(1), 101-113 Figure 4. Confidence judgments expressed as a function of Task Difficulty and Conformity Condition. Confidence level was measured as a percentage between 0% and 100%. Error bars indicate +/-1 standard error around the means.

Discussion
Following Asch's seminal studies, we set out to further investigate the phenomenon of conformity. Our experiment used the empirical paradigm of numerosity perception, which is an established topic of interest in perceptual psychology. We also included confidence judgments as an additional dependent measure. Confidence has been used in past research as an indicator of the level of uncertainty of perceptual responses and has usually been found to be positively correlated with accuracy. Our results have shown great agreement with Asch's basic findings and extended those seminal findings in important ways. We found that factually incorrect -but unanimous -peer responses lower the accuracy of perception in both easy and difficult perceptual tasks. Conformity was observed more frequently when the numerosity task was difficult. Not surprisingly, confidence judgments followed the same general trend: the more difficult the task became, the lower the reported confidence of participants. Interestingly, confidence was not lowered further by the influence of peer responses. In fact, confidence during incorrect peer response trials was lower than during correct peer response trials only for the easy trials. A possible explanation for this may be that having peer responses available bolstered confidence, rather than inhibited it, if they believed peers were providing correct responses, only under conditions in which they were truly uncertain of the correct response. On easier trials, it may be easier to notice discrepancies between the correct response and the confederates' responses, thus lowering confidence in one's own judgments.
The no peer response condition revealed the largest difference in both confidence levels and accuracy between easy and difficult trials. It seems PSIHOLOGIJA, 2020, Vol. 53(1), 101-113 that the availability of peer responses decreases the perceptual discrimination between easy and difficult trials. This effect may occur due to a combination of statistical reasons (e.g., regression to the mean) as well as perceptual, social, and cognitive factors. Because the visual quality of the stimulus was not manipulated, we suspect that perceptual factors may not have contributed to this effect. Unanimous feedback (be it correct or incorrect), on the other hand, suggests the presence of a strong social influence on the results.
One could argue that the task used in this study elicits both normative and informational conformity, but the problem is that this distinction may overlap with task difficulty. For the easy trials, the correct response was fairly obvious, but that was not the case for the difficult trials. In fact, one could argue that the difficult trials were similar in some ways to Sherif's task. In other words, the difficult trials may have potentially elicited a different process. Teasing apart the effects of different types of conformity will be an important extension of the current empirical work in future research.
Future studies should investigate the effects of peer group size and the uniformity of peer responses on conformity and perceived confidence, as these parameters were fixed in the present experiment. A more sophisticated operational definition of task difficulty also awaits future analysis. Comparison of two dot arrays depends on the total number of elements, and the difference in the number of elements between arrays. These two parameters may interact in a nonlinear fashion to produce a continuous variable that describes task difficulty (Agrillo, Piffer, Bisazza, & Butterworth, 2015). Future studies should simultaneously manipulate both factors in a more controlled manner, which would require a larger number of trials than we wished to present in our within subjects experimental design. Our present categorization into easy and difficult tasks serves as a simple, but admittedly crude approximation that is in need of further methodological refinement. Finally, there is an urgent need for including personality measures as explanatory factors that hold the promise of providing the ultimate answer to the age-old puzzle: which individuals succumb to conformity pressure more easily and why?