Experimental Design Details
Design: Two-arm between-subjects online RCT implemented in Qualtrics with individual-level random assignment (1:1). Participants are recruited online via Prolific and complete the study in a single session.
Stimuli: The treatment stimulus is a 57-second altruism-themed video selected based on prior laboratory validation using physiological responses (SCR and pulse rate) and subjective evaluations. The neutral control stimulus is a 57-second video matched in length and format; the specific control video will be selected based on a separate stimulus-validation pretest comparing candidate neutral clips (including a neutral segment from All the President’s Men and a low-content abstract-shapes clip) and a fear/suspense benchmark (The Shining).
Stimulus validation (separate online pretest): Prior to the main experiment, we conduct an online stimulus-validation pretest using 57-second clips to select the neutral control and to benchmark emotional arousal. Candidate clips include: (i) the altruism video, (ii) a neutral segment from All the President’s Men (1976), (iii) a fear/suspense clip from The Shining (1980), and (iv) a low-content neutral baseline (abstract shapes). The pretest measures (a) perceived prosocial content, (b) valence and arousal, and (c) state emotions. This pretest will work as a manipulation check for the treatment video, and the control video for the main experiment will be chosen as the candidate that is closest to the low-content baseline on prosocial perception and arousal/valence, while remaining clearly below the altruism video on perceived prosocial content.
Sequence: Consent/instructions → video (treatment vs control) → two instructional attention checks (“Please select Correct”; “Please type 20”) → state emotions measured immediately after the video on 0–100 sliders (anger, fear/worry, happiness, sadness, guilt; order randomized) → incentive rule screen → risk tolerance elicitation: (i) self-reported general willingness to take risks (0–10) and (ii) a five-question monetary staircase lottery module → demographics and personality traits.
Participants will be excluded if they: (1) do not consent, (2) have missing data for the outcomes, (3) are duplicate respondents or using bots (detected by Qualtrics), (4) fail one or both instructional attention checks, (5) complete the study implausibly fast (<1/3 of the median completion time), or (6) experience technical failures preventing video exposure or task completion.
Primary outcome: Composite risk preference index constructed as the mean of z-scored subjective risk (0–10) and z-scored staircase risk measure (z-scoring rule pre-specified in the analysis plan).
Incentives: Participants receive a fixed participation payment. Additionally, 10% of participants are randomly selected for a performance-based bonus linked to the staircase task. For selected participants, one staircase decision is randomly selected for payment; the bonus equals 1% of the realized monetary payoff from that decision. The incentive rule is displayed immediately before the risk task.
Analysis Plan
A) Pretest
A = Altruism video
B = Neutral candidate (All the President’s Men)
C = Fear benchmark (The Shining)
D = Low-content baseline (Abstract Shapes)
Hypotheses
H1 (Prosocial superiority): The altruism video is perceived as more prosocial than the other stimuli:
A > B, A > C, and A > D on the Prosocial Perception Index.
H2 (Neutrality): The neutral candidate video is neutral:
B is equivalent to D on the Prosocial Perception Index and on valence/arousal.
H3 (Fear benchmark validity): The fear video induces fear/arousal but is not prosocial:
C > D and B on fear/worry and arousal, and C is equivalent to D and B on the Prosocial Perception Index.
H4 (Specificity): Altruism is not only general emotionality:
A > C on the Prosocial Perception Index, even if C is more arousing.
In the pretest, we will compare the four arms (A altruism, B neutral candidate, C fear benchmark, D abstract-shapes baseline) on a Prosocial Perception Index (primary), and on valence, arousal, and state emotions (secondary). We will test whether the altruism video is rated as more prosocial than B, C, and D using mean comparisons (t-tests). To assess whether the neutral candidate is truly neutral, we will use equivalence testing (TOST) comparing B to D on prosocial perception and on valence/arousal. We will validate the fear benchmark by testing whether C increases fear and arousal relative to D while remaining equivalent to D on prosocial perception. The control video for the main experiment will be selected as the candidate that is closest to the abstract-shapes baseline on prosocial perception and arousal/valence, while remaining clearly below the altruism video on prosocial perception.
B) Main experiment
Hypotheses
H1 (direct effect). Participants exposed to an altruism prime will report lower risk tolerance than participants in a neutral control condition
H2 (Mediating role of emotions). The effect of altruistic priming on risk tolerance will be mediated by post-prime emotions. Specifically, to the extent that the altruism prime affects a given emotion, the corresponding indirect effect on risk tolerance is expected to be negative via fear and guilt, and positive via happiness and anger.
H3 (Net effect). The overall (total) effect of the altruism prime on risk tolerance is expected to be negative.
We will analyze the main experiment using Structural Equation Modeling (SEM) within an intention-to-treat framework. Treatment assignment is modeled as an exogenous predictor of post-video emotions (happiness, anger, fear, guilt), which in turn predict risk tolerance. the primary outcome is a composite risk preference index. The SEM estimates the total effect of treatment on risk tolerance, the direct effect conditional on emotions, and emotion-specific indirect effects (treatment → emotion → risk). Emotions are allowed to correlate. Inference for indirect effects will rely on bootstrap confidence intervals with robust standard errors, and we will report total, direct, and indirect effects along with standard fit statistics.