Hypotheses
Based on Zajonc’s drive theory \cite{Zajonc1965}, our hypotheses were the following:
Protocol & Data Collection
\label{sec:meth-data}
We recruited 45 participants after exclusion (25 for the alone condition and 20 for the human condition, 16 males, 29 females, balanced across conditions) on campus. The participants age was M=20.36 (SD=2.53). We ensured that all participants who enrolled were not colour-blind (due to the necessity of seeing colour accurately for the shape matching task) and that they were native English speakers (to prevent comprehension issues due to language in the story task).
Participants were first given information sheets describing the experiment (simply entitled “Learning with a touchscreen”, so as not to disclose the role of the mere presence of the observers). They then gave consent to participate, compliant with the university ethics committee rules. Participants were told in writing and verbally that whether or not they decided to withdraw early from the study, they would receive compensation equivalent to EUR6 (as an Amazon voucher). We made this point explicit to make sure the participants knew that, even if they quit the shape matching game early (i.e., between rounds 75 and 200), they would still receive the full compensation amount.
Results
\label{sec:study1-results}
We did not observe any difference between the two conditions concerning the number of time required to reach the limit of the 75 shapes, average reaction time, number of shape completed, ratio of correct matching or recall performance (cf. Table \ref{tab:results1}).
This means that we do not observe any social facilitation effect in this study. H1 is not supported: the presence of a human by-stander does not impact the performance for the simple task (no significance difference between number of mistakes and reaction time). Neither is H2: the ’Give-up’ button is similarly not used in both cases. Similarly, the presence or absence of a human did not impact the estimation ratio (ratio of perceived shape matching done by the real number) nor the recall of facts, invalidating both H3 and H4.
\label{tab:results1}Results for the shape matching task. No significance has been observed on any of the six metrics reported: time to reach 75 shapes, average reaction time, number of shapes completed, ratio of correct shape, ratio of perceived matching and recall performance.
Metric Alone condition \(M(STD)\) Human condition \(M(STD)\) \(p\)-\(value\)
Time to 75 shapes (s) \(117.7(30.01)\) \(110.63(17.25)\) \(.349\)
Average reaction time (s) \(1.70(0.47)\) \(1.58(0.26)\) \(.305\)
Number of shapes completed \(196(11.5)\) \(198(7.8)\) \(.522\)
Ratio of correct shapes \(0.98(0.02)\) \(0.99(0.01)\) \(.082\)
Ratio of perceived matching \(0.52(0.34)\) \(0.50(0.32)\) \(.88\)
Recall performance \(4.81(1.27)\) \(5.11(1.49)\) \(.473\)
Explain how our design decision could have impacted the results maybe later

Second Attempt

\label{sec:second}
Reflecting on the lack of effect observed in our first attempt, we designed a second experiment to address the possible failures of the first one.
Specifically, we chose to have the human observer closer to the participant (aiming for greater human influence), a stronger moral component (aiming for a greater influence of the human presence), a more difficult task (stronger incentive for behavioural differences – i.e., cheating – between conditions), money reward dependent on performance (stronger, clearer incentive for behavioural differences between conditions) and finally, regarding the methodology, we decided to move away from primarily using reaction times as metric, so as to avoid any natural performance limit.
Task
Based on these constraints, we designed a new task involving mental arithmetic. Participants were required to calculate the result of a set of non-trivial mental additions. The additions each had exactly three 2-digit numbers to sum, one carry (a digit that is transferred from one column of digits to another), and their results ranged from 100 to 200. Participants had 5 minutes to perform as many additions as possible. Each correct answer would earn them a small financial reward of £0.20 (Figure \ref{fig:sums}).
Critically, following the design of Vohs and Schooler \cite{Vohs2008}, a supposed ‘glitch’ was showing a pop-up dialogue before each addition. This dialogue was designed to look like a spurious debug dialogue and contained the expected answer. The participants were explicitly shown by the experimenter that the correct answer was erroneously displayed in the dialogue. They were instructed to ignore the dialogue and to dismiss it. This ‘bug’ was explained to the participant as being caused by a new operating system on the laptops used for the test (“Our previous computers did not have this issue”). The bug made it practically easy for participants to cheat: by briefly glimpsing at the debug dialogue before dismissing it, they could immediately know the correct answer, and earn money faster.
The dialogue could be dismissed by pressing ‘enter’ on the keyboard. ‘Enter’ was also the key used to move to the next question. As such, a double-press would move to the next question and close the dialogue before it could be seen. Through this mechanism, it was possible to measure how long it took participants to close the dialogue, and infer whether they had cheated.