Affect Misattribution Procedure (AMP) experiment for oTree

November 2019, Markus Konrad markus.konrad@wzb.eu / Berlin Social Science Center

Introduction

This repository contains an application for oTree (Chen et al. 2016) which implements the Affect Misattribution Procedure (AMP) experiment (Payne et al. 2005, Payne & Lundberg 2014, Teige-Mocigemba et al. 2017).

Images of targets (Chinese characters) in directory _static/amp/targets were published by Keith Payne at http://bkpayne.web.unc.edu/research-materials/.

Makes use of the otreeutils package (Konrad 2018).

Screenshot of a trial:

Features and limitations

• precise timing of prime and target exposures in milliseconds
• prime and target images are pre-loaded before first display to prevent download delay
• precise measurement of responses in milliseconds
• progress bar showing advancement of trials for each participant
• each measurement is stored individually in the database
• live view of measurements during experiment in oTree's data view
• easily adjustable (see configuration)
• requires keyboard for responses but may be extended to work on mobile devices as well
• results are transferred to server at the end of each round (each round or block consists of several trials), not after each trial

Requirements

• Python 3.5 or higher (tested with Python 3.6)
• otree 2.1.41
• otreeutils 0.9.1

You can install the exact requirements using pip: pip install -r requirements.txt

Configuration

Prime and target images

For your own experiment, you probably want to exchange the prime images (the repository contains example images of cats and dogs) and optionally the target images. Both are located in _static/amp/primes and _static/amp/targets respectively.

The repository comes with example images for two prime classes prime_a and prime_b inside the primes folder. The images are automatically loaded if you put them there and randomly assigned to the target classes. The number of images per prime class should be the same.

As explained before, the images for the targets are Chinese characters obtained from http://bkpayne.web.unc.edu/research-materials/ and divided into positive (pos folder) and negative (neg folder) classes. You may replace those images – you should only make sure that the number of images per target class is the same. The number of images determines the number of trials that are run per block.

Further configuration via Constants

The Constants class in amp/models.py contains further configuration settings such as the number of blocks (num_rounds) and the display timings that determine for how long the prime and target image are shown.

Code structure and page sequence

Models

A custom Trial class is defined in amp/models.py that stores information for each trial per player such as displayed prime / target and participant's response. The trials are set up in creating_session() in class Subsession where the prime and target images are detected in the respective folders and random assignment of primes and targets takes place (see sample_targets_and_primes()).

Pages and templates

The page sequence consists of four classes in amp/pages.py:

1. IntroPage
2. AMPPracticePage
3. AMPPage
4. AMPFinished

AMPPracticePage is derived from AMPPage and uses the same template (AMPPage.html) but sets is_practice to True. On the practice page, the participant can try out the experiment process without recording any measurments. Primes and targets for the practice page are loaded from the respective folders with _practice suffix in _static/amp/. The real test is then implemented in AMPPage, especially in the JavaScript functions of the HTML template. The randomized Trial objects are loaded for the participant for the given round and passed to the template where they are displayed. During the test, the response times and keys are recorded and submitted when the next block is loaded or all blocks are finished. The submitted trial responses are handled and stored to the database in the method AMPPage.before_next_page().

Data export

Since the measurements are stored using the custom data model Trials (see this blog post or Konrad 2018 for more on custom data models with oTree), the data is not exported automatically using oTree's data export page. However, two methods are provided to obtain the data in hierarchically structured JSON format:

1. You can access the page https://<SERVER>/custom_export/ (e.g. http://localhost:8000/custom_export/ on a local development machine) which, after logging in, lets you download the data.
2. You can use the data_exporter.py script, e.g. by executing python data_exporter.py my_data.json in the terminal, which will store the JSON data to my_data.json.

For later processing of the JSON data, you may use the jsonlite package for R or the built-in json module in Python.

Tests

Automated tests are implemented in amp/tests.py and can be run via otree test amp.

License

Apache License 2.0. See LICENSE file.