{
 "name": "a dispatch to a population: does a population split on a question the way a measured population did?",
 "registered": "2026-10-05",
 "honesty": "While checking that the data exists, the 2024 overall share answering too much was printed for both wordings (welfare 0.308, N 1635; assistance to the poor 0.091, N 1595). The overall 2024 gap is therefore NOT a blind test and is reported only as seen. No 2024 answer has been read by party, ideology, age or any other subgroup, and no individual 2024 answer has been read. Those are the held-out outcomes.",
 "data": {
  "source": "General Social Survey cumulative file 1972-2024, Release 3a (NORC, July 2026), public",
  "question": "Are we spending too much, too little, or about the right amount on [welfare | assistance to the poor]? GSS split ballot: NATFARE (welfare) and NATFAREY (assistance to the poor); each respondent is asked one wording.",
  "answers": "1 too little, 2 about right, 3 too much; other codes excluded",
  "weights": "WTSSPS for measured shares; unweighted for individual scores",
  "train": "years 2016, 2018, 2021, 2022",
  "test": "year 2024 only"
 },
 "population": "every 2024 respondent asked either wording with a valid answer; N is reported. Each life is described only by age, sex, race, highest degree, party identification (7 points), political views (7 points), region and the survey year.",
 "dispatch": "the wording the respondent was actually asked",
 "methods": {
  "carry_forward": "the 2022 weighted share in the same cell",
  "regression": "multinomial logistic regression of the answer on the facts, the wording, wording x party and wording x political views, fitted on train",
  "voice": "a language model answers as the person. System: You are a person living in the United States. The facts of your life are below; answer as yourself. User: the facts in plain words, then the question with the assigned wording, then: Reply with only one of: too little, about right, too much. Temperature 0. Unparseable answers are did-not-run and reported.",
  "anchored": "the regression's log-odds of each answer plus b times the voice's answer indicator, b fitted on a voiced sample of 2022 respondents (train)"
 },
 "outcomes": {
  "primary": "the weighted share answering too much in 2024 in six cells: party group (Democrat: partyid 0-2; Independent: 3; Republican: 4-6) by wording. Mean absolute error in percentage points over the six cells, per method, with 95% intervals from 2,000 bootstrap resamples of 2024 respondents.",
  "also": "the wording gap within each party group; individual log loss and accuracy; the same by political views and by age group; the overall 2024 gap (already seen, labelled as such)",
  "decision": "A method matches the measured split if its six-cell MAE is at most 3.0 points AND each party group's predicted wording gap falls inside that gap's 95% sampling interval. The voice (or anchored voice) is credited with adding something only if its six-cell MAE is lower than the regression's, with a 95% interval for the difference below zero. Every method is reported whatever happens."
 },
 "seed": 20261005
}