Presentation order but not duration affects binary risky choice

Felix Molter & Peter N. C. Mohr

Preprint at PsyArXiv · Preregistration on OSF

Abstract

Risky choice behaviour often deviates from the predictions of normative models. The information search process has been suggested as a source of some reported "biases". Specifically, gaze-dependent evidence accumulation models, where unfixated alternatives' signals are discounted, propose a mechanistic account of observed associations between eye movements, choices and response times, with longer fixated alternatives being chosen more frequently. It remains debated, however, whether gaze causally influences the choice process, or rather reflects emerging preferences. Furthermore, other aspects the information search process, like the order in which information is inspected, can be confounded with gaze duration, complicating the identification of their causal influences. In our preregistered study 179 participants made repeated incentivized choices between two sequentially presented risky gambles, allowing the experimental control of presentation duration, order, and format (i.e., alternative-wise or attribute-wise). Across presentation formats, we find evidence against an influence of presentation duration on choice. The order in which participants were shown stimulus information, however, causally affected choices, with alternatives shown last being chosen more frequently. Notably, while gaze-dependent accumulation models generally capture effects of gaze duration, causal effects of stimulus order are only predicted by some models, identifying potential for future theory development.

Project Organization

├── LICENSE
├── Makefile           <- Makefile with targets to run preprocessing, analyses and visualization
├── README.md          <- This README file.
├── data
│   ├── processed      <- The final, canonical data sets for modeling.
│   └── raw            <- The original, immutable data dump obtained from jsPsych.│
│
├── models             <- Results from statistical analyses and behavioral modeling.
│
├── reports            <- Reports and manuscripts associated with the project.
│   └── figures        <- Generated graphics and figures to be used in reporting
│
├── requirements.txt   <- The requirements file for reproducing the analysis environment, e.g.
│                         generated with `pip freeze > requirements.txt`
│
├── setup.py           <- makes project pip installable (pip install -e .) so src can be imported
├── src                <- Source code for use in this project.
│   ├── __init__.py    <- Makes src a Python module
│   │
│   ├── data           <- Scripts to download or generate data
│   │   ├── build_conditions.py
│   │   ├── make_data_overview.py
│   │   └── process_choice_data.py
│   │
│   ├── models         <- Scripts to perform statistical analyses and behavioral modeling
│   │   ├── choice_analyses.py      <- Runs behavioral analyses of choice data (GLM, BayesFactor t-tests, BEST)
│   │   ├── ddm_analyses.py         <- Runs BayesFactor t-test, BEST of relative model fit between presentation formats
│   │   ├── ddm_fitting.py          <- Fits the two DDMs to all trials, and separately for each presentation format
│   │   ├── ddm_recovery.py         <- Runs parameter and model recoveries of the DDMs
│   │   ├── model_comparison.py     <- Performs model comparison and selection
│   │   ├── static_models_fitting.py  <- Performs parameter estimation and recovery of additional, exploratory softmax models
│   │   ├── staticmodels.py         <- Contains model classes of additional exploratory softmax models
│   │   ├── utils.py                <- Contains helper functions of static models
│   │   └── ddms                    <- Contains DDM classes and functions for fitting
│   │
│   └── visualization  <- Contains scripts to create figures which are saved to reports/figures
└── task               <- Contains jsPsych code of the behavioral task, including instructions, etc.

---

Data description

The processed choice data file data/processed/main/choices.csv contains one line per trial with the following variables:

Variable name	Type	Description
subject_id	int	Subject ID
block	int	Experiment block (0 or 1)
trial	int	Trial number
condition	str	Unique condition (trial) identifier matching task/stimuli/conditions.csv
choice	int	The participant's choice. In experimental trials 0 = Hp, 1 = Hm. In catch trials 0 = dominated, 1 = dominant
rt	float	Response time (in ms)
p0	float	Alternative 0 probability
p1	float	Alternative 1 probability
m0	float	Alternative 0 winning amount
m1	float	Alternative 1 winning amount
label0	str	Text label for alternative 0
label1	str	Text label for alternative 1
higher_p	int	Indicating which alternative has a higher probability
higher_m	int	Indicating which alternative has a higher winning amount
presentation	str	Presentation format: Either by alternatives or attributes
left_alternative	int	Indicating which alternative is left
ev0	float	Alternative 0 expected value
ev1	float	Alternative 1 expected value
delta_ev	float	Difference in expected value (alt. 0 - alt. 1)
delta_ev_z	float	z-scored Difference in expected value (alt. 0 - alt. 1)
g0	float	Relative presentation duration for alternative 0
g1	float	Relative presentation duration for alternative 1
gp	float	Relative presentation duration for probability attribute
gm	float	Relative presentation duration for amount attribute
duration_favours	int	Indicates which alternative is favoured by presentation duration
duration_favours_str	int	Text label for duration_favours
last_stage_favours	int	Indicates which alternative is favoured by the last presentation stage
last_stage_favours_str	int	Text label for last_stage_favours
choose_higher_p	int	Indicates whether Hp was chosen
sequence	json	Object containing information over presentation sequence
webgazer_data	json	Webcam-based eyetracking data recorded with webgazer.js with jsPsych extension
screen_width	int	Participant screen width (in pixel)
screen_height	int	Participant screen height (in pixel)

The file data/processed/main/subject_summary.csv contains a participant-level overview. It contains the following variables:

Variable name	Type	Description
subject_id	int	Subject ID
run_id	int	disregard, dataset ID on server
exclude	bool	Whether the participant is excluded
exclusion_reason	str	Reason for which participant is excluded
gender	str	Gender (Female, Male, Other)
age	int	Age in years
n_records	int	Number of recorded trials
n_choose_nan	int	Number of timed-out responses
n_choose_dominated	int	Number of choices of dominated alternatives
n_choose_higher_p	int	Number of Hp choices
chosen_trial	str	ID of randomly determined trial for bonus payment
lucky_number	float	Random number generated to determine bonus win
won_amount	float	Bonus amount (GBP) won by participant
rg_blind	bool	Self-reported red-green blindness
rg_difficult	bool	Self-reported difficulty in distinguishing colors
serious	bool	Self-reported seriousness of participation
self_report	str	Self-reported decision strategy
comment	str	Comments by the participant (optional)

---

How to reproduce analyses

Analyses can be reproduced by installing required packages described in requirements.txt, installing the source package (pip install -e .) and using the included Makefile. The meta-target all includes all registered analyses and is automatically run with the make command.

The src folder additionally contains a copy of my myfuncs package used for statistical tests and plotting setup.

Target	Description
all	Runs all preprocessing, planned analyses and produces all figures.
conditions	Re-constructs trial conditions (output in task/stimuli/conditions.csv) and makes the corresponding figure.
data	Runs preprocessing pipeline. Creates choices.csv DataFrame and makes a summary of participants.
choice_analyses	Runs planned analyses of choice data.
choice_analyses_figures	Makes corresponding figures.
ddm_fitting	Fits DDMs to data. Accepts optional arguments: N_CORES (int, number of cores to use), N_RUNS_DDM (int, number of estimation runs) , OVERWRITE (True or False, whether to overwrite estimation results)
ddm_analyses	Runs analysis of differential relative DDM fit between presentation formats.
ddm_comparison	Runs model comparison and selection procedure.
ddm_comparison_figure	Makes the coresponding figure.
ddm_recovery	Runs recovery of the DDMs based on empirical estimates.
ddm_recovery_figures	Visualizes results from DDM recovery.

---

Project based on the cookiecutter data science project template. #cookiecutterdatascience