Claim at v1.1.0

graph-concordance-matrix

101b5d64804c3f65b69076191950c19df371f5d6fec361f0814654e4b3554249

Data

{
  "has_outcome": true,
  "kind": "confirmation",
  "lineage": {
    "audit": "docs/lineage_audit.md; data/derived/lineage_audit.json",
    "class": "none",
    "class_meaning": "the audit found no city-data source fetched before registration",
    "classes": [],
    "early_sources": []
  },
  "outcome": "graph-concordance-matrix",
  "prereg_branch": {
    "name": "prereg/graph-concordance-matrix",
    "sha": "1a049aec1b4ba4c7df0f3c5bbe9dd9bd45f0a489"
  },
  "prior": 0.6,
  "registered_at": "2026-11-12T22:49:55-0500",
  "registrant": "Claude (graph discovery study: concordance matrix)",
  "registration": {
    "alternatives": [
      "Ecological confounding: neighbourhood correlations reflect who lives where (poverty, age, race, migration), not the individual-level relation in the literature.",
      "Common valence: nearly all qualifying literature signs are positive between two adverse conditions, so agreement may show only that adverse conditions cluster in the same neighbourhoods; the 'always positive' benchmark and the sign-shuffle null measure this.",
      "Extraction and grounding error in the graph (about a fifth of API findings; negations in variable names) biases the literature sign.",
      "Measurement: survey estimates for 42 areas are noisy, and self-reported diagnosis depends on access to care.",
      "Dependence: pairs share indicators, so the effective number of independent pairs is much smaller than the count."
    ],
    "graph_evidence": [
      {
        "concepts": "80 concepts mapped to 48 of 56 city indicators (data/derived/graph/concordance/indicator_concepts.csv)",
        "direction": "105 of 109 qualifying pairs have a positive literature sign after polarity",
        "n_findings": 8792,
        "n_studies": null,
        "pubmed_ids": [],
        "summary": "Of 1,540 indicator pairs, 1,225 have a concept at both ends, 277 have at least one finding, 166 have signed findings from at least 3 studies and 109 of those have at least 80% of signed findings on one side (data/derived/graph/concordance/literature_signs.csv, written before this registration; API v1.0, 12 Nov 2026, associations not filtered against the Dataverse CSV because no local copy exists on this machine)."
      }
    ],
    "id": "graph-concordance-matrix",
    "kind": "confirmation",
    "literature": [],
    "plan": "Unit: UHF42 neighbourhood (n = 42). Indicators: the 56 in analyses/graph_concordance/indicators.py, fixed before this registration; EHDP measures use the latest period with at least 40 areas reported, and an indicator with no such period is dropped. Project tables (shootings, vacant land, modelled noise, population density) are aggregated from tracts to UHF42 by the tract's representative point. Literature sign: for each pair, all API associations linking any concept of one indicator to any concept of the other, in either order; a finding is signed when tools/graph_evidence.normalise gives a direction, the statistic is not a group difference and it has no data-check flag; the sign is multiplied by the two polarities and flipped for each variable name carrying a negation word (analyses/graph_concordance/graph_signs.py). A pair qualifies with signed findings from at least 3 studies and at least 80% of signed findings on one side. City sign: Spearman correlation across neighbourhoods, raw, and partialled for poverty (correlation of the residuals of each indicator's ranks on poverty's ranks). The poverty-partialled correlation decides; for pairs that include poverty the raw correlation decides. Reported: agreement rate raw and deciding; the area-shuffle null; a 95% bootstrap interval from resampling neighbourhoods (2,000 draws); a second null that shuffles literature signs among qualifying pairs; the agreement a rule of 'always positive' would reach, since almost all qualifying literature signs are positive; agreement among pairs with |rho| of at least 0.2 and 0.3; and the rate with the five previously analysed pairs removed. Individual pairs, including the strongest disagreements and agreements (ranked by |deciding rho| among qualifying pairs), are exploratory leads and not tests. Pairs with a concept at both ends and no findings, and indicators with no concept, are listed as gaps. Disclosure: no city correlation for this matrix was computed before registration and no EHDP value was loaded in this study; only EHDP metadata (measure names, periods, geographies) was read to choose indicators. The project has already analysed, at UHF42, NO2 and PM2.5 against asthma ED visits (mob-asthma-pollution: positive after poverty adjustment, interval includes no effect), walking and cycling to work against obesity (mob-active-commute-obesity, graph-confounders-commute-obesity: negative), commute time against depression with exercise, obesity and subway access as covariates (commute-time-depression), and rodents, cockroaches and housing problems against child asthma ED visits (nbhd-rodents-asthma: positive); at tract level vacant land against shootings (nbhd-vacant-shootings: positive) and modelled noise against 311 complaints; at NTA and community district level green space, surface temperature, air conditioning and heat illness (heat claims). Of the 109 qualifying pairs, five overlap pairs whose direction the project has seen: [('asthma_adult', 'no2'), ('asthma_adult', 'pm25'), ('asthma_adult', 'cockroaches'), ('asthma_adult', 'poverty'), ('asthma_adult', 'asthma_child')]. The registrant also knows from general knowledge that NO2 and traffic are highest in Manhattan's wealthy core and that most adverse health outcomes in New York follow poverty.",
    "prediction": {
      "direction": "+",
      "magnitude": "Share of qualifying pairs whose city correlation has the literature's sign: 65% (range 55% to 75%) for the deciding correlation, 78% (range 68% to 88%) for the raw correlation. The null is 50%.",
      "refute_if": "The agreement rate of the deciding correlation is at most 40% and below the 2.5th percentile of the same null.",
      "support_if": "The agreement rate of the deciding correlation is at least 60% and above the 97.5th percentile of the null in which each indicator's values are shuffled across neighbourhoods independently (2,000 shuffles, poverty held fixed)."
    },
    "prior": 0.6,
    "registered_at": "2026-11-12T22:49:55-0500",
    "registrant": "Claude (graph discovery study: concordance matrix)",
    "sha256": "2421c9a2e18c1b5a67d6b0a919af4609389d6d79e2e0aa3472dda1321f8d5ff4",
    "statement": "Across New York City's 42 UHF neighbourhoods, the sign of the correlation between two published indicators matches the majority sign of the System Public Graph's findings for the same pair of concepts more often than chance, for the pairs on which the graph holds at least 3 studies and at least 80% sign consistency."
  },
  "registration_commit": {
    "commit": "14974809d38dfb08f5d04876755b33ef49c76dc1",
    "committed_at": "2026-11-12T22:49:56-05:00"
  },
  "registration_file_sha256": "b6c21cac878f98915fba13e8e5a153d8ec1634ab072bfb29bb3e3f3c9a980a84",
  "registration_path": "claims/registered/graph-concordance-matrix.json",
  "registration_sha256": "2421c9a2e18c1b5a67d6b0a919af4609389d6d79e2e0aa3472dda1321f8d5ff4",
  "registration_text": "{\n \"id\": \"graph-concordance-matrix\",\n \"kind\": \"confirmation\",\n \"statement\": \"Across New York City's 42 UHF neighbourhoods, the sign of the correlation between two published indicators matches the majority sign of the System Public Graph's findings for the same pair of concepts more often than chance, for the pairs on which the graph holds at least 3 studies and at least 80% sign consistency.\",\n \"graph_evidence\": [\n  {\n   \"concepts\": \"80 concepts mapped to 48 of 56 city indicators (data/derived/graph/concordance/indicator_concepts.csv)\",\n   \"n_findings\": 8792,\n   \"n_studies\": null,\n   \"direction\": \"105 of 109 qualifying pairs have a positive literature sign after polarity\",\n   \"summary\": \"Of 1,540 indicator pairs, 1,225 have a concept at both ends, 277 have at least one finding, 166 have signed findings from at least 3 studies and 109 of those have at least 80% of signed findings on one side (data/derived/graph/concordance/literature_signs.csv, written before this registration; API v1.0, 12 Nov 2026, associations not filtered against the Dataverse CSV because no local copy exists on this machine).\",\n   \"pubmed_ids\": []\n  }\n ],\n \"literature\": [],\n \"prediction\": {\n  \"direction\": \"+\",\n  \"magnitude\": \"Share of qualifying pairs whose city correlation has the literature's sign: 65% (range 55% to 75%) for the deciding correlation, 78% (range 68% to 88%) for the raw correlation. The null is 50%.\",\n  \"support_if\": \"The agreement rate of the deciding correlation is at least 60% and above the 97.5th percentile of the null in which each indicator's values are shuffled across neighbourhoods independently (2,000 shuffles, poverty held fixed).\",\n  \"refute_if\": \"The agreement rate of the deciding correlation is at most 40% and below the 2.5th percentile of the same null.\"\n },\n \"plan\": \"Unit: UHF42 neighbourhood (n = 42). Indicators: the 56 in analyses/graph_concordance/indicators.py, fixed before this registration; EHDP measures use the latest period with at least 40 areas reported, and an indicator with no such period is dropped. Project tables (shootings, vacant land, modelled noise, population density) are aggregated from tracts to UHF42 by the tract's representative point. Literature sign: for each pair, all API associations linking any concept of one indicator to any concept of the other, in either order; a finding is signed when tools/graph_evidence.normalise gives a direction, the statistic is not a group difference and it has no data-check flag; the sign is multiplied by the two polarities and flipped for each variable name carrying a negation word (analyses/graph_concordance/graph_signs.py). A pair qualifies with signed findings from at least 3 studies and at least 80% of signed findings on one side. City sign: Spearman correlation across neighbourhoods, raw, and partialled for poverty (correlation of the residuals of each indicator's ranks on poverty's ranks). The poverty-partialled correlation decides; for pairs that include poverty the raw correlation decides. Reported: agreement rate raw and deciding; the area-shuffle null; a 95% bootstrap interval from resampling neighbourhoods (2,000 draws); a second null that shuffles literature signs among qualifying pairs; the agreement a rule of 'always positive' would reach, since almost all qualifying literature signs are positive; agreement among pairs with |rho| of at least 0.2 and 0.3; and the rate with the five previously analysed pairs removed. Individual pairs, including the strongest disagreements and agreements (ranked by |deciding rho| among qualifying pairs), are exploratory leads and not tests. Pairs with a concept at both ends and no findings, and indicators with no concept, are listed as gaps. Disclosure: no city correlation for this matrix was computed before registration and no EHDP value was loaded in this study; only EHDP metadata (measure names, periods, geographies) was read to choose indicators. The project has already analysed, at UHF42, NO2 and PM2.5 against asthma ED visits (mob-asthma-pollution: positive after poverty adjustment, interval includes no effect), walking and cycling to work against obesity (mob-active-commute-obesity, graph-confounders-commute-obesity: negative), commute time against depression with exercise, obesity and subway access as covariates (commute-time-depression), and rodents, cockroaches and housing problems against child asthma ED visits (nbhd-rodents-asthma: positive); at tract level vacant land against shootings (nbhd-vacant-shootings: positive) and modelled noise against 311 complaints; at NTA and community district level green space, surface temperature, air conditioning and heat illness (heat claims). Of the 109 qualifying pairs, five overlap pairs whose direction the project has seen: [('asthma_adult', 'no2'), ('asthma_adult', 'pm25'), ('asthma_adult', 'cockroaches'), ('asthma_adult', 'poverty'), ('asthma_adult', 'asthma_child')]. The registrant also knows from general knowledge that NO2 and traffic are highest in Manhattan's wealthy core and that most adverse health outcomes in New York follow poverty.\",\n \"alternatives\": [\n  \"Ecological confounding: neighbourhood correlations reflect who lives where (poverty, age, race, migration), not the individual-level relation in the literature.\",\n  \"Common valence: nearly all qualifying literature signs are positive between two adverse conditions, so agreement may show only that adverse conditions cluster in the same neighbourhoods; the 'always positive' benchmark and the sign-shuffle null measure this.\",\n  \"Extraction and grounding error in the graph (about a fifth of API findings; negations in variable names) biases the literature sign.\",\n  \"Measurement: survey estimates for 42 areas are noisy, and self-reported diagnosis depends on access to care.\",\n  \"Dependence: pairs share indicators, so the effective number of independent pairs is much smaller than the count.\"\n ],\n \"prior\": 0.6,\n \"registrant\": \"Claude (graph discovery study: concordance matrix)\",\n \"registered_at\": \"2026-11-12T22:49:55-0500\",\n \"sha256\": \"2421c9a2e18c1b5a67d6b0a919af4609389d6d79e2e0aa3472dda1321f8d5ff4\"\n}",
  "statement": "Across New York City's 42 UHF neighbourhoods, the sign of the correlation between two published indicators matches the majority sign of the System Public Graph's findings for the same pair of concepts more often than chance, for the pairs on which the graph holds at least 3 studies and at least 80% sign consistency."
}

History in this collection

  1. addedv1.0.0Nov 14, 2026this version