'The Helsinki Summit : A multiround dictator game.'
Created: 09/05/2018 03:59 AM (PT)
This is an anonymized version of the pre-registration. It was created by the author(s) to use during peer-review.
A non-anonymized version (containing author names) should be made available by the authors when the work it supports is made public.
1) Have any data been collected for this study already?
It's complicated. We have already collected some data but explain in Question 8 why readers may consider this a valid pre-registration nevertheless.2) What's the main question being asked or hypothesis being tested in this study?
1. High trait paranoia be associated with increased attribution of harmful intent to partners across all trials (but trait paranoia will not be associated with variation in attributions of self-interest).
2. Attribution of harmful intent to different dictators will follow a dose-response relationship (fair < partially fair < unfair) across all ranges of paranoia. However, those will high paranoia will have a higher baseline of average harmful intent. There will be no interaction between trait paranoia and dictator fairness on attribution of harmful intent.
3) Describe the key dependent variable(s) specifying how they will be measured.
(1) Paranoia score (sum of scores for social reference and persecutory delusions subscales) converted to an ordinal categorical variable of 9-11 levels.
(2) Average Self-interest: ordinal categorical variable (< 11 levels) subject inference that dictator was motivated by desire to earn more averaged over each block of 6 trials.
(3) Average Harmful-intent: ordinal categorical variable (< 11 levels) subject inference that dictator was motivated by desire to reduce their bonus averaged over each block of 6 trials.
For each variable, the number of levels will depend on the distribution of the data. All levels will have > 20 observations
4) How many and which conditions will participants be assigned to?
All subjects will take part in the ‘receiver’ role in an 18 trial multi-round Dictator Game (DG) and will be asked to rate (2 x scales, 0-100) the extent to which they believe the Dictator decision was motivated by (i) self-interest and (ii) harmful intent after each transaction ('state' measures).
Participants will play sequentially against 3 dictators (order counter-balanced) and will interact with each dictator in 6 consecutive trials. After the 6th trial with each dictator, participants will be asked to attribute overall self-interest and harmful intent to each dictator ('trait' measures). This is so ‘state’ and ‘trait’ interpretations of the dictators can be dissociated.
Dictators are preprogramed to either make a fair (50:50) or unfair (100:0) split.
Subjects will be assigned to (i) fair (Dictators always make fair decisions) (ii) partially fair (50% of the time will make unfair decisions) and (iii) unfair (Dictators always make unfair decisions) Dictators across 18 trials.
Finally, all subjects will also make 6 Dictator decisions – these data are only collected so that we can truthfully inform subjects that they were paired with a real partner and these data will not be analysed.
5) Specify exactly which analyses you will conduct to examine the main question/hypothesis.
For all analyses, a model selection approach with model averaging will be used. Input variables will be standardised / centred as appropriate. Zero-averaging will be used.
We will use cumulative link models (clm) cumulative link mixed models (clmm, Christensen, 2015) which allow the dependent variable to be expressed as an ordinal categorical variable. Mixed models allow random terms to be included to account for repeated measures.
Model (i) will be a clm specified as follows: paranoia ~ gender + age
Model (ii) will be a clmm specified as follows:
harmful intent (average) ~ paranoia (continuous) + dictator (fair/unfair/partially fair) + dictator * paranoia + dictator-order (whether fair dictator is presented first, second etc.) + age + gender + comprehension-questions + (1|ParticipantID)
Model (iii) will be a clmm with identical explanatory terms except that 'self-interest (average) will replace 'harmful intent (average)'
Models (ii) and (iii): we plan to use clmms as we have 3 measures of harmful intent and self-interest attributions for each participant. However, with such a small number of repeated measures per participant, we may struggle to fit the model adequately. If the model fails to fit, we will instead run three separate models (clm) exploring responses to (i) fair; (ii) partially fair and (iii) unfair dictators - with the same explanatory terms as outlined above but without the 'participant id' random effect.6) Describe exactly how outliers will be defined and handled, and your precise rule(s) for excluding observations.
We will include participants: - Aged between 18-65 - UK residents - Fluent in english - No previous history of mental health diagnoses7) How many observations will be collected or what will determine sample size?
No need to justify decision, but be precise about exactly how the number will be determined.
We have already recruited 1000 participants to take the initial GPTS survey. These participants will be called back after a minimum interval of 7 days to take part in the Helsinki Summit (a multi-round dictator game) task. We aim at recalling as many of the original 1000 subjects as possible, but as a minimum we will aim for 500 subjects. To recall participants, we will first send an email to all participants who are eligible. Once the responses have slowed down to be < 10 per day, we will send another request email to the remaining subjects who have not responded. Once responses are < 2 per day, we will stop data collection.8) Anything else you would like to pre-register?
(e.g., secondary analyses, variables collected for exploratory purposes, unusual analyses planned?)
We have already collected the initial data for the GPTS scores from 1000 participants on prolific.ac. We have not yet submitted the Helsinki Summit to this population or collected any data from them for harmful intent or self-interest attributions.
We would like to explore a secondary aim:
High trait paranoia will be associated with reaching a peak in harmful intent attribution (defined as a score of 60 or more) in fewer trials when analysing each participant but no difference in attribution of self-interest.
We will analyse data from each dictator type separately.
We will split participants into high (scores => 61 on the GTPS), medium (scores 36 =< x =< 60) and low paranoia (scores =< 35 on the GTPS) types (Raihani & Bell, 2017). We will also conduct the same analysis with a more conservative high paranoia threshold if we have sufficient power (scores > 101.9 on the GTPS) to reflect clinical means, creating four groups of low (=< 35), medium (36 =< x =< 60) high (61 =< x =< 101.8), and very high (101.9 =< x =<160).
For each participant who attributes a score of 60 or more to the partner for HI, we will record the earliest trial number where this attribution was made. We will then perform a Kruskal-Wallace test with Dunn post-hoc analysis to analyse the difference in trial number between the high, medium, and low paranoia populations. If participants make intransitive responses, we will also perform a Kruskal-Wallace test with Dunn post-hoc analysis where intransitive responders are excluded.
For SI, we will conduct the same non-parametric analysis.
We would also like to explore average intention attributions (harmful intent and self-interest) across participants within trials for each dictator type, as a function of whether participants score high or low for paranoia (defined as above). We will use regression curves for both intransitive and non-intransitive trial sets to identify whether paranoia has an effect on the time to reach peak attributions of harmful intent and self-interest.
We would also like to explore an assumption: while trait attributions of harmful intent and self-interest are not used in the analysis, we will perform paired t-test to determine whether trait attributions differ from the average of state attributions. This will test our assumptions of using average state attributions as a measure of intent over time.