Is correction triggered by the step or by the gap? Results of a pre-registered test of Route 1 of the Correction Freeze Problem
Description
Route 1 of the Correction Freeze Problem (CFP; postulated by Professor Vuong Quan Hoang as a mathematical apparatus), the shifting-baseline ratchet, holds that a belief is corrected only when a single step in the true state clears a perceptibility threshold, so that a run of small steps can carry any amount of drift without triggering correction. We tested this claim against a rival rule, under which correction is triggered by the gap accumulated since the last revision. The test used 189,827 monthly answers from 24,977 respondents in the New York Fed Survey of Consumer Expectations (June 2013 to January 2026), with monthly changes in retail gasoline prices as the step. Every prediction, falsifier, and model was registered on OSF before the data were downloaded. A simulation run before data access showed that the registered measure of the gap was biased in favour of the rival, and it was replaced in a timestamped addendum. Respondents were split into an exploration sample, used only to estimate the threshold, and a confirmatory sample, opened after that estimate was archived. The threshold was estimated at a 13% monthly gasoline change, with a wide bootstrap interval of 1.5% to 13%. In the confirmatory sample, two of the three pre-registered falsifiers fired. The effect of the latest step was small, with a 95% interval that included zero, and the accumulated gap predicted revisions about three times as strongly as the step. Route 1 therefore fails this test. One registered robustness check qualifies the failure: when regional prices are used with survey-month fixed effects, which remove nationwide news, both the step effect and the gap effect disappear. The evidence against Route 1 may therefore reflect common national shocks rather than each respondent’s own accumulated gap. A forward holdout on months not yet published remains pending.