Measuring the Effects of Gender on Online Social Conformity
This post summarises our research paper titled “Measuring the Effects of Gender on Online Social Conformity” by Senuri Wijenayake, Niels van Berkel, Vassilis Kostakos, and Jorge Goncalves. We study how indications of peer gender may trigger stereotypical conformity in online group settings, setting forth design recommendations for future online settings accounting for conformity effects. The paper will be presented at the ACM Conference on Computer-Supported Cooperative Work and Social Computing 2019 (CSCW) on November 11th 2019.
“Going with the crowd” or adjusting one’s individual behaviour in order to align with the majority’s sentiment is known as social conformity. From online learning management systems where students conform to majority’s answers in quizzes, to adjusting personal opinions to reflect the crowd’s sentiment in political issues on Twitter, social conformity is a common phenomenon in online group settings. In uncertain situations (such as in a quiz or a political vote), individuals tend to perceive the crowd to have a better chance at being “correct” and adopt their behaviour with the intention of being “right” (known as informational influences leading to conformity).
In addition to contextual and personal determinants of conformity (such as the size of the group’s majority and an individual’s self-confidence), perceptions of peer competency is a crucial determinant of our likelihood to conform. The gender of the group members is one such factor influencing perceptions of competency in group settings. Previous work shows that individuals tend to perceive a certain gender group to be more competent than others in certain stereotypical tasks. However, the effect of such stereotypical perceptions on conformity behaviour is yet to be explored in online group settings.
Gender-based Stereotypes
We investigate how indications of peer gender in online group settings may trigger stereotypical perceptions of competency and result in social conformity. First, we explore how different gender group compositions may affect online conformity behaviour of participants (e.g., a majority with more women or men, or with an equal number of women and men). Since the experiment is related to an online setting, we utilise commonly used stereotypical gendered representations (i.e., stereotypical masculine and feminine avatars and names) to illustrate different gender compositions. Second, we compare these two stereotypical gendered representations in terms of triggering gender-related stereotypes and gender-biased conformity. Finally, we also investigate whether the self-disclosed gender identity of participants affects their susceptibility to such gender-biased stereotypes leading to social conformity. We limit our study design to binary genders (men and women) in order to capture the stereotypical notions assigned to these genders and to investigate the effects of the stereotypical designs common in online contexts (e.g., avatars).
We deployed an online quiz containing multiple-choice questions (illustrated in Figure 1) equally distributed among topics stereotypically seen as being of masculine (sports), feminine (fashion), and neutral (general knowledge) nature. While participants were instructed that they would join seven others (i.e., the size of the majority and the minority excluding the participant sums up to seven) in completing the quiz, there was in fact only one participant per session. Peer answers were simulated by our custom-developed software.
Participants first answer each question privately while also providing their self-reported confidence level (from 0-100) on the selected answer (Step 1). Next, our software displays a fabricated distribution of peer answers denoting a clear majority, while placing the participant in either the minority or majority (Step 2). Fabricating the peer answers allowed us to test a variety of gender compositions in both the group majority and minority. The example provided in Figure 1 demonstrates a situation in which the participant was placed in a women-only minority against a majority consisting of more men than women. Subsequent to displaying the fabricated peer answers, participants are given the opportunity to change both their initial answer and the level of confidence in their answer (Step 3). We consider a change to the participant’s answer to be a sign of conformity when the change is in line with the majority’s opinion.
To assess the impact of gender cues on triggering stereotypical perceptions among participants, we introduce a total of three conditions, with varying levels of peer representations:
- Control: Peers were represented using a grey square (no gender cues).
- Names: Peers were represented using stereotypical masculine and feminine names.
- Avatars: Peers were represented using stereotypical masculine and feminine silhouettes.
Following the completion of the quiz, participants were interviewed, and discussed which factors led them to change their initial answers during the quiz, and whether they sought answers from peers from a particular gender group. Participants were also asked to compare the effect of avatars and names used, on triggering stereotypical perceptions in online settings.
Pilot
Prior to the experiment, we conducted a pilot study with 10 men and 10 women, who independently answered the same set of multiple-choice questions. The responses received from the pilot study were used to determine the arrangement of the majority and minority groups when fabricating peer answer diagrams, ensuring that groups were placed in reasonable answer options regardless of being correct or incorrect. Additionally, the results of the pilot study (with a 38% overall correctness and no statistically significant relationships between gender and perceived question type) were useful to disprove the popular notion that men are more knowledgeable in stereotypical masculine topics and women in stereotypical feminine topics.
Findings
Conforming to the majority’s answers was visible in a significant 39% of the responses (across the three conditions) where participants were placed in the group minority. We then used three separate generalised linear mixed models (GLMM) for the three conditions, to identify statistically significant determinants of conformity in each scenario.
Across all three conditions, the group size difference (i.e., the difference between the majority and the minority) and the initial self-reported confidence of the participants had significant main effects. In the interviews, participants highlighted how low self-confidence and larger majorities made them rethink their personal answers, suggesting the effect of informational influences on online conformity behaviour.
P43: “When you see a significant majority, you start second-guessing. If it was something I knew 100% I would not change it. But if it was something I was very confident, but was not 100% sure, it made me second guess. When there were lots of people on the opposing majority, it made me feel that if that many selected the answer, it could be right.”
In conditions with gender cues, participants were seen to actively perceive peer gender through the available cues. Participants also identified sports related questions as masculine and fashion related questions as feminine expertise areas, suggesting the prevalence of gender-stereotypical perceptions of peer competency in online group settings.
P36: “I think generally in our society women would care more about fashion and the knowledge that comes with that. Men are more interested in a lot of different sports. ”
Moreover, such gender-stereotypical perceptions were seen to influence participants’ conformity behaviour during the quiz. We note that both men and women were more inclined to conform to a majority with more stereotypical masculine names or avatars in sports related questions, and similarly with stereotypical feminine names or avatars in fashion related questions. These results suggest that in the presence of gender cues, conformity behaviour was influenced by gender-stereotypes in addition to the usual informational influences.
P40: “If it was a sports-related question, I will feel most comfortable not to change my answer into a group with women. And vice versa. But if the question was about geography or flags, it won’t make any difference to me.”
Furthermore, this effect was larger in the presence of silhouette avatars than in names, implying that graphical user representations are more likely to trigger such stereotypes than names.
The use of objective questions enabled us to quantify the effects of conformity driven by gender-stereotypes on answer correctness. We note that the introduction of gender cues resulted in more incorrect answers for both stereotypical masculine and feminine questions, implying that gender cues encourage individuals to conform to incorrect answers more frequently, especially in topics that can trigger a biased response.
Implications for the Design
The findings of our study establish that, despite the reduced social presence in online settings, individuals stereotypically perceive others’ competency based on available gender cues, resulting in gender-biased conformity behaviour. Thus, to minimise the adverse effects of gender-biased conformity in online group settings, where user decisions may be negatively influenced by others (e.g., online learning platforms), we present the following design recommendations:
- We suggest carefully considering whether displaying gender and other user cues is relevant and value-adding from the perspective of the end-user.
- We recommend against the use of gender cues such as binary-gendered avatars, especially in situations in which group members could perceive the competency of others based on gender.
- We support using alternatives devoid of gender cues such as identicons, avatars with user initials, or site specific avatars, to ensure unbiased discussion and decision making.
It is worth noting that this study is premised on a traditional gender binary model, in which gendered senses of self fall into two clearly discernible categories, however gender is far more complex than this. In addition, there are other stereotypes beyond gender (e.g., age-based) which were not considered in this study and are worthy of further investigation.
Paper citation: S. Wijenayake, N. van Berkel, V. Kostakos, J. Goncalves, “Measuring the Effects of Gender on Online Social Conformity”, Proceedings of the ACM on Human-Computer Interaction — CSCW, vol. 3, 2019, 145:1–145:24. doi: 10.1145/3359247.
