Claim at v1.2.0
transfer-chicago-concordance-matrix
12659d753d9dddd2127febf032146a4a67751c8a737ac3f4d22f186dce81b69c
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": "transfer-chicago-concordance-matrix",
"prereg_branch": {
"name": "prereg/transfer-chicago-concordance-matrix",
"sha": "2805fb30a1b67b36e4b1b9fcaa1642806867ebc5"
},
"prior": 0.35,
"registered_at": "2026-11-13T01:23:02-0500",
"registrant": "Claude (round 5: Chicago concordance matrix)",
"registration": {
"alternatives": [
"Measurement: a smooth regional PM2.5 or ozone surface ranks community areas by position in the region (a north-south gradient), so its correlation with outcomes may reflect geography and not exposure.",
"Ozone: in New York ozone pairs agreed (5 of 6) because ozone is lowest in the dense core; the same chemistry may or may not hold on the EJSCREEN surface.",
"Partialling poverty removes much of what sorting adds in a city where pollution follows poverty, so agreement after partialling may still fail even if the sorting explanation is right; the raw correlation is reported beside it.",
"Racial segregation: in Chicago health outcomes follow race more closely than the poverty rate, and a poverty partial leaves that in.",
"Common valence and dependence: fifteen pairs share three exposures and a handful of outcomes, so the effective number of independent pairs is small; the bootstrap over areas carries this.",
"Different measures of the outcomes in the two cities (prevalence against emergency visits, years of life lost against death before 65)."
],
"graph_evidence": [
{
"concepts": "the New York concept mapping, reused unchanged (data/derived/graph/concordance/indicator_concepts.csv)",
"direction": "all 15 qualifying pairs of PM2.5, ozone or traffic with a Chicago health outcome have a positive literature sign",
"n_findings": 8792,
"n_studies": null,
"pubmed_ids": [],
"summary": "Literature signs are read from data/derived/graph/concordance/literature_signs.csv, computed for graph-concordance-matrix on 12 Nov 2026 (API v1.0). Of the 37 Chicago indicators, all map to a New York key; 80 of their pairs qualify in the literature (at least 3 signed studies, at least 80% one sign), 77 with a positive sign. Fifteen include PM2.5 (6), ozone (5) or traffic (4), each paired with a health outcome. No new graph request is made."
}
],
"id": "transfer-chicago-concordance-matrix",
"kind": "confirmation",
"literature": [],
"plan": "Unit: Chicago community area (n = 77). Indicators: the 37 in analyses/round5_chicago_matrix/chicago_indicators.py, each a Chicago Health Atlas topic matched to a New York indicator key, so that the New York concepts, polarities and literature signs apply unchanged. Two topics that report absence (uninsured, physically inactive) are reversed as 100 minus the value. Each topic is read at the community-area layer for the total population, at the latest period (by end year; the longer window on a tie) with a value for at least 70 of the 77 areas; a topic with no such period is dropped (fetch.py). Qualifying rule, as in New York: signed findings from at least 3 studies and at least 80% of signed findings on one side. City sign, as in New York: Spearman correlation across areas, raw, and with poverty partialled out (correlation of the residuals of each indicator's ranks on poverty's ranks); the partialled correlation decides, and the raw correlation decides for pairs that include poverty. matrix.py copies spearman_matrix, partial_matrix and decide from analyses/graph_concordance/matrix.py and was checked before registration to reproduce New York's 72 of 109, 10 of 27 and 29 of 56 from the New York files. Reported for Chicago: agreement overall and by kind of pair (the four kinds of graph_concordance/explore.py and the referee's three role kinds), each with a 95% bootstrap interval over areas; the agreement of an always-positive rule and the margin over it with an interval; the area-shuffle null (each indicator shuffled across areas, poverty fixed) and the sign-shuffle null (literature signs shuffled among qualifying pairs), 2,000 draws each; the Spearman correlation of PM2.5 with poverty with a bootstrap interval. New York is recomputed with the same code on all its qualifying pairs and on the pairs that qualify and are computable in both cities, and the difference in the air-and-traffic share on the same pairs is reported with an interval from independent bootstrap draws. Seed 20261113 (numpy default_rng([seed, 0]) for Chicago, [seed, 1] and [seed, 2] for New York), 2,000 draws. The verdict is computed by verdict() in matrix.py from the conditions above. Under PRECEDENCE rule 9 the run is repeated with 20,000 draws and with five other seeds (--draws, --seed, --tag, --no-null); if the bootstrap condition is met under some and not others, the verdict is unsettled. Scripts fixed at commit 0ba7f1a43b4ef768eee2a0516459991a9d6405b3 on local main and identified by content: {'analyses/round5_chicago_matrix/chicago_indicators.py': {'sha256': '400fd529dad0471065aacd33d66e7a7e9f7b92c710225e7e124a084228069e68', 'git_blob': '25227bd5d3cfb7b2cb158ddc07e8da20113e6452'}, 'analyses/round5_chicago_matrix/fetch.py': {'sha256': 'f6a64c975191269a8a8479ec433c14b94ebaec43f530c0e6a2380c4c4e15034f', 'git_blob': 'bd48953db77251ef99b485193153dc1cd4ae8c90'}, 'analyses/round5_chicago_matrix/matrix.py': {'sha256': '90d9f9d96345bc6368f8149073a5ab2c8fd27390ec7f8c8a8642030ccacf3b76', 'git_blob': '11a54aa41fba79bf7ad7bf98af6ea2ee03c51d5d'}}. The support condition is on pairs of an exposure with an outcome, where the literature's sign says something about a cause, and not on the overall rate, which the always-positive rule nearly reproduces (referee report critique/graph-concordance-matrix.md). Within the fifteen air and traffic pairs every literature sign is positive, so agreement there equals the always-positive rate and the comparison is between cities, not against that rule. Differences from New York that the design does not remove: PM2.5 and ozone are EPA EJSCREEN regional model surfaces, smoother than New York's street-level survey, and ozone may vary very little across the city; traffic is proximity to traffic, not vehicle miles; asthma is survey prevalence, not emergency visits; premature death is years of potential life lost; low birthweight is not restricted to term births; depression is a CDC PLACES model estimate built partly from demographic predictors. Notes per indicator are in chicago_indicators.py. New York indicators with no Chicago match: no2 (no NO2 topic); black_carbon (diesel particulate matter (DPM) exists but is a different quantity and concept); noise (no noise topic); surface_temp (no surface temperature topic); green_cover (green space proximity (EKG) is discontinued and measures distance, not cover); park_access (no park access topic); ac (no air conditioning topic); housing_problems (no maintenance topic); cockroaches (no pest topic); transit_commute (active transportation (ACT) mixes walking, cycling and public transport); walk_commute (same); asthma_child (youth asthma ED visits (CHAOHOB) are not published by community area); psych_hosp (behavioural health hospitalisations (HDBHR) are not published by community area). Disclosure: before registration the registrant read only metadata from the Chicago Health Atlas API (the topic list with names, units and descriptions; the coverage endpoint, which lists periods and population groups for 59 topics; the layer list) and, to learn the response format of the data endpoint, one response for the topic POP (total population, not an indicator in this study), of which about a dozen community-area population counts were displayed. No value of any of the 37 indicators was loaded, and no file under data/cache/chicago_atlas existed. The project's earlier Chicago claims (transfer-chicago-heat-shootings, -hot-end, -weekday-pattern) used citywide daily series, not community-area indicators. The New York side is fully seen. The registrant knows from general reading that Chicago's South and West sides are poorer, have worse health and hold the industrial corridors, that traffic is densest near the expressways and the Loop, and that EJSCREEN PM2.5 varies by about 1 microgram per cubic metre within a city.",
"prediction": {
"direction": "+",
"magnitude": "(a) Share of qualifying pairs that include PM2.5, ozone or traffic whose deciding correlation has the literature's sign: 60% in Chicago (range 40% to 80%), against 10 of 27 (37%) in New York and 6 of 15 (40%) in New York on the same fifteen pairs. (b) Spearman correlation of PM2.5 with poverty across community areas: +0.35 (range +0.10 to +0.60), against +0.11 across New York's UHF neighbourhoods (Pearson -0.06; computed from city_indicators_uhf42.csv, n = 42). (c) Descriptive: overall agreement of the deciding correlation 68% (range 58% to 78%), and a margin over the always-positive rule between -3 and +5 percentage points.",
"no_test_if": "Fewer than 12 qualifying pairs that include PM2.5, ozone or traffic have a computable deciding correlation in Chicago (for example because an air indicator has no period with 70 of 77 areas). The verdict is then no_test and the other quantities are reported as descriptive.",
"otherwise": "unsettled",
"prediction_b": "Scored hit when the Spearman correlation of PM2.5 with poverty is positive and its 95% bootstrap interval excludes zero; miss when the estimate is zero or negative; partial otherwise. It does not decide the verdict. If it is a miss, the outcome says that Chicago, as measured here, is not a city whose high-PM2.5 areas are poor, and that prediction (a) is then not a test of the sorting explanation for PM2.5.",
"prediction_c": "Reported with intervals, without a verdict.",
"refute_if": "With at least 12 such pairs, the share is at most 0.370 and the upper limit of the same bootstrap interval is below 60%.",
"support_if": "At least 12 qualifying pairs include PM2.5, ozone or traffic and have a computable deciding correlation; all of them are pairs of an exposure with a health outcome. Among them the share that agrees with the literature's sign is at least 60%, the lower limit of its 95% bootstrap interval (community areas resampled, 2,000 draws, seed 20261113) is above New York's 10 of 27 (0.370), and the share exceeds New York's share on the same pairs."
},
"prior": 0.35,
"registered_at": "2026-11-13T01:23:02-0500",
"registrant": "Claude (round 5: Chicago concordance matrix)",
"sha256": "7e47ca346696757000081b75e4c48e3e3f60b30dd99c37117791743c80c8e89f",
"statement": "Across Chicago's 77 community areas, pairs of an air pollution or traffic indicator with a health outcome agree with the sign of the System Public Graph's findings more often than the same kind of pair did across New York's 42 UHF neighbourhoods (10 of 27), as the residential-sorting explanation of the New York result predicts for a city whose polluted areas are poor.",
"transfer": {
"from_city": "New York City",
"from_claim": "graph-concordance-matrix",
"nyc_estimate": {
"air_traffic_noise_pairs_agreeing": "10 of 27",
"all_pairs_agreeing": "72 of 109",
"always_positive": "70 of 109",
"exposure_with_outcome_pairs_agreeing": "29 of 56"
},
"to_city": "Chicago"
}
},
"registration_commit": {
"commit": "499547cdfbeb5673f18fc44a3194c9db320b74a9",
"committed_at": "2026-11-13T01:23:02-05:00"
},
"registration_file_sha256": "abfa2d925d297132ad37f71f0e735be00fcc82a97ae0b8c713546f6b2225c401",
"registration_path": "claims/registered/transfer-chicago-concordance-matrix.json",
"registration_sha256": "7e47ca346696757000081b75e4c48e3e3f60b30dd99c37117791743c80c8e89f",
"registration_text": "{\n \"id\": \"transfer-chicago-concordance-matrix\",\n \"kind\": \"confirmation\",\n \"transfer\": {\n \"from_city\": \"New York City\",\n \"to_city\": \"Chicago\",\n \"from_claim\": \"graph-concordance-matrix\",\n \"nyc_estimate\": {\n \"air_traffic_noise_pairs_agreeing\": \"10 of 27\",\n \"exposure_with_outcome_pairs_agreeing\": \"29 of 56\",\n \"all_pairs_agreeing\": \"72 of 109\",\n \"always_positive\": \"70 of 109\"\n }\n },\n \"statement\": \"Across Chicago's 77 community areas, pairs of an air pollution or traffic indicator with a health outcome agree with the sign of the System Public Graph's findings more often than the same kind of pair did across New York's 42 UHF neighbourhoods (10 of 27), as the residential-sorting explanation of the New York result predicts for a city whose polluted areas are poor.\",\n \"graph_evidence\": [\n {\n \"concepts\": \"the New York concept mapping, reused unchanged (data/derived/graph/concordance/indicator_concepts.csv)\",\n \"n_findings\": 8792,\n \"n_studies\": null,\n \"direction\": \"all 15 qualifying pairs of PM2.5, ozone or traffic with a Chicago health outcome have a positive literature sign\",\n \"summary\": \"Literature signs are read from data/derived/graph/concordance/literature_signs.csv, computed for graph-concordance-matrix on 12 Nov 2026 (API v1.0). Of the 37 Chicago indicators, all map to a New York key; 80 of their pairs qualify in the literature (at least 3 signed studies, at least 80% one sign), 77 with a positive sign. Fifteen include PM2.5 (6), ozone (5) or traffic (4), each paired with a health outcome. No new graph request is made.\",\n \"pubmed_ids\": []\n }\n ],\n \"literature\": [],\n \"prediction\": {\n \"direction\": \"+\",\n \"magnitude\": \"(a) Share of qualifying pairs that include PM2.5, ozone or traffic whose deciding correlation has the literature's sign: 60% in Chicago (range 40% to 80%), against 10 of 27 (37%) in New York and 6 of 15 (40%) in New York on the same fifteen pairs. (b) Spearman correlation of PM2.5 with poverty across community areas: +0.35 (range +0.10 to +0.60), against +0.11 across New York's UHF neighbourhoods (Pearson -0.06; computed from city_indicators_uhf42.csv, n = 42). (c) Descriptive: overall agreement of the deciding correlation 68% (range 58% to 78%), and a margin over the always-positive rule between -3 and +5 percentage points.\",\n \"support_if\": \"At least 12 qualifying pairs include PM2.5, ozone or traffic and have a computable deciding correlation; all of them are pairs of an exposure with a health outcome. Among them the share that agrees with the literature's sign is at least 60%, the lower limit of its 95% bootstrap interval (community areas resampled, 2,000 draws, seed 20261113) is above New York's 10 of 27 (0.370), and the share exceeds New York's share on the same pairs.\",\n \"refute_if\": \"With at least 12 such pairs, the share is at most 0.370 and the upper limit of the same bootstrap interval is below 60%.\",\n \"no_test_if\": \"Fewer than 12 qualifying pairs that include PM2.5, ozone or traffic have a computable deciding correlation in Chicago (for example because an air indicator has no period with 70 of 77 areas). The verdict is then no_test and the other quantities are reported as descriptive.\",\n \"otherwise\": \"unsettled\",\n \"prediction_b\": \"Scored hit when the Spearman correlation of PM2.5 with poverty is positive and its 95% bootstrap interval excludes zero; miss when the estimate is zero or negative; partial otherwise. It does not decide the verdict. If it is a miss, the outcome says that Chicago, as measured here, is not a city whose high-PM2.5 areas are poor, and that prediction (a) is then not a test of the sorting explanation for PM2.5.\",\n \"prediction_c\": \"Reported with intervals, without a verdict.\"\n },\n \"plan\": \"Unit: Chicago community area (n = 77). Indicators: the 37 in analyses/round5_chicago_matrix/chicago_indicators.py, each a Chicago Health Atlas topic matched to a New York indicator key, so that the New York concepts, polarities and literature signs apply unchanged. Two topics that report absence (uninsured, physically inactive) are reversed as 100 minus the value. Each topic is read at the community-area layer for the total population, at the latest period (by end year; the longer window on a tie) with a value for at least 70 of the 77 areas; a topic with no such period is dropped (fetch.py). Qualifying rule, as in New York: signed findings from at least 3 studies and at least 80% of signed findings on one side. City sign, as in New York: Spearman correlation across areas, raw, and with poverty partialled out (correlation of the residuals of each indicator's ranks on poverty's ranks); the partialled correlation decides, and the raw correlation decides for pairs that include poverty. matrix.py copies spearman_matrix, partial_matrix and decide from analyses/graph_concordance/matrix.py and was checked before registration to reproduce New York's 72 of 109, 10 of 27 and 29 of 56 from the New York files. Reported for Chicago: agreement overall and by kind of pair (the four kinds of graph_concordance/explore.py and the referee's three role kinds), each with a 95% bootstrap interval over areas; the agreement of an always-positive rule and the margin over it with an interval; the area-shuffle null (each indicator shuffled across areas, poverty fixed) and the sign-shuffle null (literature signs shuffled among qualifying pairs), 2,000 draws each; the Spearman correlation of PM2.5 with poverty with a bootstrap interval. New York is recomputed with the same code on all its qualifying pairs and on the pairs that qualify and are computable in both cities, and the difference in the air-and-traffic share on the same pairs is reported with an interval from independent bootstrap draws. Seed 20261113 (numpy default_rng([seed, 0]) for Chicago, [seed, 1] and [seed, 2] for New York), 2,000 draws. The verdict is computed by verdict() in matrix.py from the conditions above. Under PRECEDENCE rule 9 the run is repeated with 20,000 draws and with five other seeds (--draws, --seed, --tag, --no-null); if the bootstrap condition is met under some and not others, the verdict is unsettled. Scripts fixed at commit 0ba7f1a43b4ef768eee2a0516459991a9d6405b3 on local main and identified by content: {'analyses/round5_chicago_matrix/chicago_indicators.py': {'sha256': '400fd529dad0471065aacd33d66e7a7e9f7b92c710225e7e124a084228069e68', 'git_blob': '25227bd5d3cfb7b2cb158ddc07e8da20113e6452'}, 'analyses/round5_chicago_matrix/fetch.py': {'sha256': 'f6a64c975191269a8a8479ec433c14b94ebaec43f530c0e6a2380c4c4e15034f', 'git_blob': 'bd48953db77251ef99b485193153dc1cd4ae8c90'}, 'analyses/round5_chicago_matrix/matrix.py': {'sha256': '90d9f9d96345bc6368f8149073a5ab2c8fd27390ec7f8c8a8642030ccacf3b76', 'git_blob': '11a54aa41fba79bf7ad7bf98af6ea2ee03c51d5d'}}. The support condition is on pairs of an exposure with an outcome, where the literature's sign says something about a cause, and not on the overall rate, which the always-positive rule nearly reproduces (referee report critique/graph-concordance-matrix.md). Within the fifteen air and traffic pairs every literature sign is positive, so agreement there equals the always-positive rate and the comparison is between cities, not against that rule. Differences from New York that the design does not remove: PM2.5 and ozone are EPA EJSCREEN regional model surfaces, smoother than New York's street-level survey, and ozone may vary very little across the city; traffic is proximity to traffic, not vehicle miles; asthma is survey prevalence, not emergency visits; premature death is years of potential life lost; low birthweight is not restricted to term births; depression is a CDC PLACES model estimate built partly from demographic predictors. Notes per indicator are in chicago_indicators.py. New York indicators with no Chicago match: no2 (no NO2 topic); black_carbon (diesel particulate matter (DPM) exists but is a different quantity and concept); noise (no noise topic); surface_temp (no surface temperature topic); green_cover (green space proximity (EKG) is discontinued and measures distance, not cover); park_access (no park access topic); ac (no air conditioning topic); housing_problems (no maintenance topic); cockroaches (no pest topic); transit_commute (active transportation (ACT) mixes walking, cycling and public transport); walk_commute (same); asthma_child (youth asthma ED visits (CHAOHOB) are not published by community area); psych_hosp (behavioural health hospitalisations (HDBHR) are not published by community area). Disclosure: before registration the registrant read only metadata from the Chicago Health Atlas API (the topic list with names, units and descriptions; the coverage endpoint, which lists periods and population groups for 59 topics; the layer list) and, to learn the response format of the data endpoint, one response for the topic POP (total population, not an indicator in this study), of which about a dozen community-area population counts were displayed. No value of any of the 37 indicators was loaded, and no file under data/cache/chicago_atlas existed. The project's earlier Chicago claims (transfer-chicago-heat-shootings, -hot-end, -weekday-pattern) used citywide daily series, not community-area indicators. The New York side is fully seen. The registrant knows from general reading that Chicago's South and West sides are poorer, have worse health and hold the industrial corridors, that traffic is densest near the expressways and the Loop, and that EJSCREEN PM2.5 varies by about 1 microgram per cubic metre within a city.\",\n \"alternatives\": [\n \"Measurement: a smooth regional PM2.5 or ozone surface ranks community areas by position in the region (a north-south gradient), so its correlation with outcomes may reflect geography and not exposure.\",\n \"Ozone: in New York ozone pairs agreed (5 of 6) because ozone is lowest in the dense core; the same chemistry may or may not hold on the EJSCREEN surface.\",\n \"Partialling poverty removes much of what sorting adds in a city where pollution follows poverty, so agreement after partialling may still fail even if the sorting explanation is right; the raw correlation is reported beside it.\",\n \"Racial segregation: in Chicago health outcomes follow race more closely than the poverty rate, and a poverty partial leaves that in.\",\n \"Common valence and dependence: fifteen pairs share three exposures and a handful of outcomes, so the effective number of independent pairs is small; the bootstrap over areas carries this.\",\n \"Different measures of the outcomes in the two cities (prevalence against emergency visits, years of life lost against death before 65).\"\n ],\n \"prior\": 0.35,\n \"registrant\": \"Claude (round 5: Chicago concordance matrix)\",\n \"registered_at\": \"2026-11-13T01:23:02-0500\",\n \"sha256\": \"7e47ca346696757000081b75e4c48e3e3f60b30dd99c37117791743c80c8e89f\"\n}",
"statement": "Across Chicago's 77 community areas, pairs of an air pollution or traffic indicator with a health outcome agree with the sign of the System Public Graph's findings more often than the same kind of pair did across New York's 42 UHF neighbourhoods (10 of 27), as the residential-sorting explanation of the New York result predicts for a city whose polluted areas are poor."
}History in this collection
- addedv1.0.0Nov 14, 2026this version