This experiment examines people's aversion to harmful actions in real moral choice behaviour. It tests how harm aversion is affected when the harmful action performed has beneficial outcomes for the good of others in comparison to for the good of oneself. If pro-social outcomes are more competitive than self-interested outcomes against an emotionally salient cause, it is expected that people will be less harm averse when benefiting others than when benefiting themselves. But should self-interest out compete pro-sociality in this situation, people will actually be more harm averse when benefiting others. Participants: We expect to recruit 800 participants (200 per condition) from the US using the online labour market Amazon Mechanical Turk (MTurk). This will provide a more representative population sample than the commonly used student sample. Participants will be paid $0.50 and have the possibility to earn up to $1 more (either for themselves or charity) depending on the task. They will reserve complete anonymity, only identified by their MTurk identification number. In accordance with MTurk regulations, workers will be a minimum age of 18 years. Participants will be excluded from the analysis should they fail check tasks (see Method) or answer the main task faster than possible to properly read and comprehend (faster than 8 seconds, based on a pilot study involving 20 participants). Method (see file for survey material): We will use a 2 x 2 between-subjects design that varies the proposer role (self vs. agent) and their distribution action (split vs. take) in a one-shot dictator game. In the self condition, participants will be asked as proposers to distribute $1 between themselves and an orphan (responder). The distribution will be framed using the perceived initial ownership of the $1. In the split condition, initial ownership will be neutral. In the take condition, initial ownership will be with the orphan. The agent condition will be identical apart from that participants no longer propose a distribution between the orphan and themselves, but between the orphan and a charity. The dependent variable will be the amount remaining with the orphan once the participant has completed the task. Before undergoing the task, participants will be assured anonymity and asked to give consent. It will have been made clear that they are free to withdraw from the task at any point. After agreement they will each be randomly assigned to one of four groups (split-self, split-agent, take-self, take-agent) to play a particular variant of a dictator game. They will be reminded that all money involved is real and will be distributed according to their decision. Response times will be measured. Should any participant respond before 8 seconds, it is expected that they have not fully attended to the task and they will be excluded (based on the pilot study). A question will be displayed asking the participants how they would like to distribute 100 cents, the exact wording of which will vary by condition. It will be explained that we (the experimenter) have given them this amount to distribute as they wish. Below this question will be a photo of an orphan below which their name is written along with a brief biography and a statement stating that 100 cents can provide her with four meals. Below this (phrases in italics indicate variations between the self and agent conditions) will be a generic avatar/the charity logo, below which lies the word “YOU”/the charity name, a brief description of its work. To avoid copyright issues, the charity will be fictional but made realistic by basing its description on a real charity and having a logo constructed using a logo creation software. Adjacent to each of the two donation options will be a text box where the participant can type how much they wish to donate. There will also be a TOTAL text box displaying the sum of the two donations. The participant will only be able to progress if this displays 100. The order each entity appears will be randomised. For participants in the split condition, both text boxes will initially be set at 0 cents. The participants will be told that they have been given 100 cents and can divide it between themselves/charity and the child as they please. They will be asked to indicate how they would like to allocate the endowment by typing in the relevant text boxes. In the take condition, the orphan’s text box will initially be set to 100 cents while the participant’s/charity’s text box will initially be set to 0 cents. The participant will be told that the orphan has been given a donation of 100 cents and that they can take any amount of this away from her for themselves/charity. The participant will be asked to indicate how much they would like to take away from the orphan by typing in the relevant text boxes. The main task will be succeeded by an understanding and attention check question involving basic arithmetic. Then the same screen they had for the dictator game will be presented to them, but this time they will be asked to indicate what they think the average distribution behaviour of others given the same task will be. After a manual text transcribing task to filter out automated responses, the participant will be given a post-experimental questionnaire that will include a demographics survey. Finally, they will be debriefed. It will be explained that the orphan and charity are fictional and in reality any money attributed to them will be donated to a children's charity. Analysis: Analysis will primarily involve a 2 x 2 ANOVA. It will only be conducted once all data has been collected. In the take condition, the remaining orphan endowment will give a proxy for harm aversion (more remaining indicates greater aversion). The split condition will be used as a control. We will first see if there is a difference between the take and split conditions, which will confirm that the take action is viewed as more immoral than the action of split. Then we will see if there are any interaction effects between the self and agent conditions, which will help shed light upon the effects of self-beneficial and other-beneficial outcomes upon harm aversion.