Two Ways of Being Misled: Composition Bias, Weak Within-Municipality Progress, and a Fixed Size Gradient in Japan's Individual Evacuation Plan Statistics, 2022–2026

Kunihisa Ohno
Additional contact information
Kunihisa Ohno: NPO Intensive Care Collaboration Network (ICON)

No yncdm_v1, SocArXiv from Center for Open Science

Abstract: Japan requires every municipality to maintain a registry of residents who cannot evacuate unaided (2013 amendment to the Disaster Countermeasures Basic Act) and, since 2021, to make best efforts to prepare an individual evacuation plan for each person listed. Progress is monitored through annual national surveys published as municipality-level tables in PDF format. We reconstruct these tables into a machine-readable panel of 1,741 municipalities across five survey waves (2022–2026; 10 source documents, plan and registry series), with documented recovery of 74 rows lost to table-rendering defects, external reconciliation of ambiguous column semantics against published national aggregates, and a complete provenance registry of corrections and known gaps. Three findings follow, and they cut in different directions. First, the pooled statistics are misleading about direction: the median creation rate across all responding municipalities falls from 19.0% to 17.8%, whereas every continuously observed subset improves, because late-entering municipalities start low and are added faster than incumbents advance. Second, and against our own initial reading, composition control does not reveal robust progress. Within the 423 municipalities observed in all five waves, the median change is only +0.77 points over four years, with a bootstrap 95% confidence interval reaching down to zero (+0.00 to +1.80; mean +3.79, CI +0.81 to +6.82; Wilcoxon p = 0.0006), and only 54.6% improved at all. The +8.3-point rise in the panel median is a shift in the position of the distribution driven by strong gains in its upper part, not the experience of a typical municipality. Third, plan creation is steeply graded by registry size in levels—the smallest quartile reaches a median of 43.3% against 9.9% for the largest—but not in growth: within-municipality change does not differ across size quartiles (Kruskal–Wallis p = 0.47; municipality-fixed-effects interactions all p > 0.55). Small municipalities are ahead, not faster, and the gap shows no tendency to close. The published series is thus misleading twice over: it reports a decline where continuously observed municipalities do not decline, and it invites the inference that composition control would reveal healthy progress, which the panel does not support. We interpret this within an institutional-debt framework: the statistical encoding of an institution (survey composition, question definitions, item discontinuation) drifts away from the institutional reality it is meant to measure, and the divergence is itself measurable—but correcting the encoding does not, here, rescue the institution's performance. The panel, extraction code, and provenance registry are archived for reuse.

Date: 2026-08-20
References: Add references at CitEc
Citations:

Downloads: (external link)
https://osf.io/download/6a86ad327020c5b09a6c574a/

Related works:
This item may be available elsewhere in EconPapers: Search for items with the same title.

Export reference: BibTeX RIS (EndNote, ProCite, RefMan) HTML/Text

Persistent link: https://EconPapers.repec.org/RePEc:osf:socarx:yncdm_v1

DOI: 10.31235/osf.io/yncdm_v1

Access Statistics for this paper

More papers in SocArXiv from Center for Open Science
Bibliographic data for series maintained by OSF ().