A Case in Point: Social Facilitation

\label{sec:mere-presence}

Context: Studying the Mere Presence Effect in Social Facilitation

Background and Related Theories
In 1898, Norman Triplett \cite{Triplett1898} observed that cyclists pedal faster in presence of rivals than when they are alone. He studied this effect on children by using a fishing reel that they need to turn as quickly as possible and found the same effect, although a later analysis of his work by Stroebe \cite{Stroebe2012} showed that there was no significant difference in his findings. This effect has later been termed as ‘social facilitation’ by Allport \cite{Allport1924} to describe the increase in response due to the presence of others who are performing the same task. Social facilitation occurs by two types of condition: ‘co-action effects’ like Triplett’s examples, and ‘audience effects’, in which only the mere presence of an observer effects the performance of a person performing the task. In order to explain the audience effects, Zajonc \cite{Zajonc1965} proposed the drive theory, which states that the audience enhances the exhibition of dominant responses in a person. In the case of a well-mastered task (‘simple task’), the performance is facilitated, whereas, for the tasks that are new or require learning (‘complex tasks’), the performance is inhibited.
Many other theories underpin social facilitation: anticipation of evaluation \cite{Cottrell1972}, attentional conflict \cite{Sanders1975}, monitoring  \cite{Guerin1983}, self-awareness \cite{Duval1972} and self-presentation  \cite{Bond1982}.
Factors
A meta-analysis by Bond and Titus \cite{Bond1983} compares 202 published and 39 unpublished studies on social facilitation. They provide a list of 13 factors that might impact social facilitation (like the participants’ age, the number of observers, the role of the observers, the familiarity of the observers, etc.). The meta-study shows that the performance speed (‘quantity’) is increased for the simple tasks and the performance accuracy (‘quality’) is decreased for the complex tasks. The performance quantity is measured by the latency to respond, time it takes to complete a task and the number of responses per unit time. The performance quality is measured by the number of errors. The study also shows that the visibility of the observers to the subjects has a slightly bigger effect than the non-visibility, although the difference was not significant. Examples from the literature are use of a one-way mirror \cite{Ganzer1968, Criddle1971}, use of a video camera \cite{Geen1973,Terry1993}, or a desktop image on a computer screen \cite{Gardner2008}. On the other hand, Bernard Guerin \cite{Guerin1983} argues in a review that the majority of studies with social facilitation effects had observers or the experimenters watching the subject while the task was performed, that is, they were not busy with other tasks. However, he states that there should be some uncertainty if the observer will evaluate the subjects, which meant that there should be evaluation potential which the subject should not know beforehand. He also draws attention the ceiling and floor effects for the tasks. He states that the task should be difficult enough so that the measurements can vary between subjects and conditions.
Tasks
Following Zajonc, the literature on social facilitation distinguish between ‘simple tasks’ and ‘complex tasks’. Examples of simple tasks include letter cancellation, multiplication, and vowel cancellation; examples of complex tasks include concept formation, anagrams, digit span, and pursuit rotor tasks. Tasks such as letter-copying and paired associates can be either simple or complex depending on the task structure. McCaffrey et al. \cite{McCaffrey1996} also presented significance levels of each of these tasks in the literature. It shows that visual perception and construction tasks such as letter/word copying \cite{Guerin1989, Terry1993, Gardner2008} and motor tasks such as physical activities \cite{Strube1981} are good tasks in terms of significance as simple tasks, and memory or learning tasks such as paired associates \cite{Cottrell1967, Geen1973, Guerin1983} and visuomotor tasks as in rotary pursuit performance \cite{Lombardo1975, Miller1979} have higher significance for social facilitation as complex tasks.
Cheating as a reinforcing factor
Self-presentation theory \cite{Bond1982} also suggests conformity to normative behaviours to gain approval of another person. For example, in the case of an embarrassing situation such as cheating, this should prevent the subject from engaging in the cheating behaviour due to social pressure. There might be several factors that effect cheating behaviour as examined by several researchers, such as the importance of the task, risk of being caught, probability of success \cite{Vitro1972}, belief in free-will \cite{Vohs2008}, knowledge of peer performance \cite{Hill1969}, higher gain of money or grades, penalty \cite{Nagin2003} or conformity to cheating behaviour in peers \cite{Fosgaard2013}. In the study by Vohs and Schooler \cite{Vohs2008}, in order to observe the cheating behaviour, a computer-based mental arithmetic test was used. The participants were made aware that there is a glitch in the program which shows the correct answer to the problem but they could close the answer window by pressing a key after the problem appeared. They were also told that the experimenter would not know whether they pressed the bar, even though the number of presses was recorded, but they should try to solve the problems honestly. Their results revealed that those who were given an essay prior to the test that stated the lack of free-will cheated more frequently than others.
Social Facilitation In Robotics
The audience effect has been studied in HRI by Schermerhorn \cite{Schermerhorn2008} and Riether et al. \cite{Riether2012}. Schermerhorn compared the effect of the robot presence during easy and difficult arithmetic tasks with alone and robot-presence conditions. A significant two-way interaction between gender and robot was found, because the male subjects performed worse during the difficult task when the robot was present. Overall, a marginally significant effect of robot presence was found. Riether et al. on the other hand, compared alone, human-presence, and anthropomorphic robot-presence conditions with four different tasks with easy and complex conditions: anagram solving, numerical distance, finger tapping and motor reaction task. They observed that in the anagram solving, numerical distance and finger tapping tasks, there were significantly larger difference scores than the alone group for both robot and human conditions, but there was no significant difference between the robot and the human observer conditions. This suggests that people regard robots as social beings. After the experiment, the subjects were asked to fill a survey in which they gave higher observation impression scores for the robot condition than the human observer, implying that they thought someone else was watching through the eyes of the robot or due to novelty effects leading to distraction.
Following the findings from social facilitation literature, we decided to explore the mere presence of two robotic platforms (the Aldebaran NAO and Pepper) through a social facilitation task. We anticipated that there would be a difference between the two platforms due to their appearance that would be useful to make other HRI researchers aware of. While the studies aimed to compare the social facilitation of two different robots, it was important to establish two baselines first: one with no observers, and one with the social facilitation elicited by a human. This would essentially be a first step in replicating the finding from psychology. Assuming this replication was successful, the robots could then be tested to see how they compare both with each other and our psychology-inspired baselines.
We ran two distinct studies, with a total of three different tasks. Because no effect could be found between the alone condition and the human condition in any of our tasks, we did not actually pursue the studies to include the robots.

Social Facilitation: First Attempt

\label{sec:first}
The first study was run between-subjects with two conditions: alone condition against human-presence. Participants were recruited on the university campus and taken to a room in the campus library for the experiment. The experimenter would take the participant to the room and tell them to follow instructions on the tablet, then the experimenter would leave. In the human observer condition, a second experimenter would already be sitting in the room and would remain there for the duration of the experiment (as per Figure \ref{fig:setup}).