Vote-Buying List Experiment for Sensitive Questions

# List Experiment on Vote Buying (item count technique), Kenya 2007 instrument
# Instrument source (wording quoted verbatim):
#   Graeme Blair, Alexander Coppock, and Margaret Moor (2020), "When to Worry
#   about Sensitivity Bias: A Social Reference Theory and Evidence from 30
#   Years of List Experiments." American Political Science Review 114(4):1297-1315.
#   DOI: 10.1017/S0003055420000374   (open access, CC BY 4.0)
#   The wording is the paper's worked example, p. 1302 ("List Experiments to
#   Reduce Sensitivity Bias"), which reproduces the list experiment in
#   Eric Kramon (2016), "Where is vote buying effective? Evidence from a list
#   experiment in Kenya." Electoral Studies 44:397-408.
#   DOI: 10.1016/j.electstud.2016.09.006
# Method and estimator:
#   Graeme Blair and Kosuke Imai (2012), "Statistical Analysis of List
#   Experiments." Political Analysis 20(1):47-77. DOI: 10.1093/pan/mpr048
# Replication data for Blair et al. (2020): Harvard Dataverse,
#   DOI 10.7910/DVN/YUXHZT, CC0 1.0. It is not used here. No respondent-level
#   data is redistributed, and the sample items below are illustrative.
#
# Instrument (control group, quoted from p. 1302):
#   "Election campaigns are a busy time in our country. I am going to read you
#   a list of some of things that people have told us happened to them during
#   the 2007 campaign. I am going to read you the whole list, and then I want
#   you to tell me how many of the different things happened to you. Please do
#   not tell me which of the things happened to you, just how many. If you
#   would like me to repeat the list, I will do so."
#   1. Politicians put up posters or signs in the area where you live.
#   2. You read the newspaper almost every day to learn about the campaign.
#   3. You met a politician personally to discuss his or her candidacy.
#   4. You discussed the campaign with a chief or another traditional leader.
# Treatment group ("Received" arm): the same script with a fifth item,
#   5. You received money from a party or politician.
# The paper notes that item order is typically randomized and the sensitive
# item is not necessarily the last item. Here it is last, as printed.
#
# How Potato runs it: conditions.factors.list_group assigns each participant to
# control or treatment (assignment: balanced). The sensitive statement is
# removed from the control group's page on the server. The participant types a
# count. Run
#   python -m potato.export --config config.yaml --format survey_experiments --output out/
# for the difference-in-means estimate with a Welch standard error.
#
# Adaptations:
#   - Kramon's survey had a second treatment arm ("Influenced"), whose fifth
#     item read "You voted for a party or politician because they gave you
#     money during the campaign." One scheme carries one sensitive statement,
#     so only the "Received" arm is built.
#   - The original was read aloud by an interviewer in a face-to-face survey.
#     Here it is a self-administered page, and the repeat-the-list offer has no
#     equivalent because the list stays on screen.
#   - Each sample item is one illustrative interview slot. With
#     max_annotations_per_user: 1 a participant answers once, as a respondent
#     would. Nobody's answers from the 2007 survey are included.
#   - randomized_response is supported by Potato but not added, because no
#     verified published wording was collected for it.
#
# Trainer: none. No trainer in potato/training/registry.py consumes
# list_experiment, so this design has no model_training block.

annotation_task_name: "List Experiment: Vote Buying (Kenya 2007 Instrument)"
task_dir: "."

data_files:
  - sample-data.json

item_properties:
  id_key: "id"
  text_key: "text"

output_annotation_dir: "annotation_output/"
export_annotation_format: "jsonl"

port: 8000

conditions:
  factors:
    list_group: [control, treatment]
  assignment: balanced

annotation_schemes:
  - annotation_type: list_experiment
    name: campaign_list
    description: "Election campaigns are a busy time in our country. I am going to read you a list of some of things that people have told us happened to them during the 2007 campaign. I am going to read you the whole list, and then I want you to tell me how many of the different things happened to you. Please do not tell me which of the things happened to you, just how many. If you would like me to repeat the list, I will do so."
    items:
      - "Politicians put up posters or signs in the area where you live."
      - "You read the newspaper almost every day to learn about the campaign."
      - "You met a politician personally to discuss his or her candidacy."
      - "You discussed the campaign with a chief or another traditional leader."
    sensitive_item: "You received money from a party or politician."
    treatment:
      factor: list_group
      level: treatment
    count_prompt: "How many of these things happened to you?"
    label_requirement:
      required: true

annotation_instructions: |
  This page reproduces the wording of a published list experiment on vote
  buying. Read the list, then enter how many of the statements apply to you.
  Do not say which ones. Half of participants see a list with one more
  statement than the others, and the difference between the two groups'
  average counts estimates how many people received money, without any
  individual answer revealing it. Your answer here is a demonstration and is
  not part of any study.

user_config:
  allow_all_users: true
max_annotations_per_user: 1
num_annotators_per_item: 2