Claim at v1.0.0
afford-poor-count-shift
b7c150c33b80ed59928bcdee2477f5294067e7568f4e840e04abe14b0cbcd40b
Data
{
"has_outcome": true,
"kind": "confirmation",
"lineage": {
"audit": "docs/lineage_audit.md; data/derived/lineage_audit.json",
"class": "e",
"class_meaning": "cannot tell",
"classes": [
"e"
],
"early_sources": [
{
"class": "e",
"dataset": "file",
"evidence": "Same workbook, which holds the outcome (post-2000 poverty counts). already_seen: \"no poverty rate or population after 2000 was read\", because the simulated output was cut at 70 lines. Referee: the claim's block starts at line 70, so the statement is \"consistent with the output to the line\" but \"The repository cannot confirm that the output was cut\"; the full simulated block was written to a file.",
"fetched": "2026-11-13T07:37:28-05:00"
}
]
},
"outcome": "afford-poor-count-shift",
"prereg_branch": {
"name": "prereg/afford-poor-count-shift",
"sha": "e2aff5c758ca8ec86504c03c5d234da787224f48"
},
"prior": 0.5,
"registered_at": "2026-11-13T07:47:06-0500",
"registrant": "Claude (affordability agent)",
"registration": {
"already_seen": "Read before registration: the 2000 values of every indicator for the 55 sub-borough areas and the city (rent, income, college share, poverty, population), which fix the low-start group; the workbook's column names, data dictionary and the count of non-missing cells by year. Not read directly: any value after 2000 for rent, college share, poverty or population. One leak occurred. The code test (`gentrification_test.py --simulate`) was meant to replace later-year values with noise; it multiplied the real values by lognormal noise instead (sigma 0.3 on each annual value, seed 0), and I read its output: k = 6, k2 = 7, Spearman 0.38, and the list of areas it classed as gentrified (Brooklyn Heights/Fort Greene, Bushwick, Coney Island, Lower East Side/Chinatown, North Crown Heights/Prospect Heights, Williamsburg/Greenpoint). That output is a noisy view of the outcome: the noise is about as large as the real variation between areas. The thresholds (k of 12 or more, 4 to 11, 1 or fewer) and the script were written before that run and were not changed after it. Prior knowledge: the Furman Center's 2016 report classed 15 of 22 low-income sub-borough areas as gentrifying on rent growth above the median from 1990 to 2014, with one condition (rent) where this test has two; Alroy 2023 classed 14 of 42 neighbourhoods as gentrifying or hypergentrifying on rent growth from 2000 to 2017. Also seen before registration, none of them a deciding variable: executed evictions by community district and year (6z8x-wfk4), IRS county outflows, and the graph findings listed. For this claim: no poverty rate or population after 2000 was read, in any form; the simulated run printed only the first claim's block before the output was cut at 70 lines.",
"alternatives": [
"A fall in the poor count is not displacement. Poor residents can leave, die, or stop being poor; the count cannot tell these apart.",
"The poverty line is national and takes no account of rent, so a household above it may still be unable to pay a higher rent.",
"Citywide poverty fell over the period; the comparison with other low-rent areas removes the common part.",
"New housing adds residents who are not poor and lowers the rate without changing the count; the test uses the count for that reason.",
"Public housing and regulated units keep poor residents in place in some gentrifying areas (Lower East Side, Harlem) and weaken the contrast.",
"The 2000 poverty count is from the census long form and the later ones from the ACS, which measure income over different reference periods; this affects both groups alike."
],
"can_it_fail": {
"assumption": "The spread of the log change in the poor count across SBAs is not known before the data are read; I assume a standard deviation between 0.20 and 0.35.",
"power": "For a true difference of -0.30 the support condition is met with probability 0.90 (sd 0.20, 8 areas), 0.73 (sd 0.25, 8 areas), 0.60 (sd 0.25, 5 areas) or 0.34 to 0.43 (sd 0.35). With no true difference it is met with probability 0.025. The smallest difference detectable with 80% power is 0.26 to 0.52.",
"reading": "The test can fail. An unsettled verdict is likely if the true difference is under 0.25.",
"standard_error": "With 8 gentrified and 14 other areas the standard error of the difference is 0.09 (sd 0.20), 0.11 (sd 0.25) or 0.16 (sd 0.35); with 5 and 17 it is 0.10, 0.13 or 0.18."
},
"graph_evidence": [
{
"association_id": "variable-name search 'gentrification', 'gentrifying', 'gentrified'",
"concepts": [
"gentrification (variable names)",
"health and injury outcomes"
],
"direction": "0",
"n_findings": 33,
"n_studies": 15,
"pubmed_ids": [
37018221,
36367672,
37728889,
37678111,
24022181,
29272306,
29709759
],
"summary": "The graph holds 33 findings in 15 studies with gentrification as a variable name; all concern health or injury outcomes and none concerns how often low-rent neighbourhoods gentrify. The studies' own counts, read in the abstracts, give a prevalence: Alroy 2023 classed 14 of 42 New York neighbourhoods as gentrifying or hypergentrifying (rent growth 2000 to 2017); Bhavsar 2022 found that 5 of 28 eligible block groups in Durham gentrified; Spitzer 2023 classed 23.8% of 59,379 urban tracts as gentrifying. Gao 2023 found opposite signs for severe maternal morbidity under two definitions (OR 0.89 and 1.18), and Bhavsar reports that results depend on the definition."
},
{
"association_id": "variable-name search 'displacement', 'eviction', 'rent burden', 'housing cost burden'",
"concepts": [
"displacement, eviction, rent burden (variable names)",
"health outcomes"
],
"direction": "+",
"n_findings": 150,
"n_studies": 60,
"pubmed_ids": [
29272306,
35000688,
38007965,
36120757
],
"summary": "About 150 findings in about 60 studies (collection still running at registration; counts approximate). Lim 2017 followed 12,882 residents of gentrifying New York neighbourhoods who had hospital records: 23% were displaced between 2006 and 2014, and the displaced had 1.8 times (1.5 to 2.2) the rate of emergency visits and hospitalisations, mainly for mental health. Schwartz 2022: New York evictions in 2017 and loss of Medicaid coverage, OR 1.63 (1.38 to 1.92). Graetz 2024: an eviction judgment and 40% (36 to 43) higher mortality; the graph stores the percentages 40, 19, 16 and 12 as hazard ratios. The graph holds no finding that compares the rate at which low-income residents leave gentrifying and non-gentrifying neighbourhoods, which is the disputed question (Freeman and Braconi 2004; Ding, Hwang and Divringi 2016; Dragan, Ellen and Glied 2020)."
}
],
"graph_evidence_note": "As for afford-gentrification-prevalence. The graph has no finding on counts of poor residents and none that compares exit rates of low-income residents between gentrifying and other poor neighbourhoods. Lim 2017 gives the only number: 23% of a hospital-using cohort in gentrifying New York neighbourhoods moved to a lower-income area or out in eight years, with no comparison group in the graph record.",
"id": "afford-poor-count-shift",
"kind": "confirmation",
"literature": [
{
"citation": "NYU Furman Center. State of New York City's Housing and Neighborhoods in 2015, Focus on Gentrification. 2016.",
"doi": null,
"finding": "15 of 55 sub-borough areas classed as gentrifying (low income in 1990, rent growth above the median 1990 to 2010-2014), 7 as non-gentrifying, 33 as higher-income."
},
{
"citation": "Freeman L, Braconi F. Gentrification and displacement: New York City in the 1990s. J Am Plann Assoc 2004;70:39-52.",
"doi": "10.1080/01944360408976337",
"finding": "Poor households in gentrifying New York neighbourhoods were less likely to move than poor households elsewhere."
},
{
"citation": "Dragan K, Ellen IG, Glied S. Does gentrification displace poor children and their families? Reg Sci Urban Econ 2020;83:103481.",
"doi": "10.1016/j.regsciurbeco.2019.103481",
"finding": "Low-income children in New York born 2006 to 2008 in gentrifying areas were no more likely to move than those in areas that stayed poor."
}
],
"plan": "Same script and data as afford-gentrification-prevalence (analyses/affordability/gentrification_test.py, claim2 block), run unchanged. This claim is registered before that script has been run on the real panel, so the group membership is not known at registration except through the leak described under already_seen.",
"prediction": {
"definitions": "Poor count: pop_pov_pct / 100 x pop_num. Change: log of the mean-of-2022-2024 count over the 2000 count. Groups: the low-start SBAs classed gentrified and not gentrified by the primary definition of afford-gentrification-prevalence (rent growth and college change from 2000, both above the city's). Interval: Welch t, 95%.",
"direction": "-",
"magnitude": "Difference in mean log change of the poor count, gentrified minus not gentrified, between -0.45 and -0.10; best guess -0.25 (about 22% fewer poor residents relative to the comparison areas).",
"otherwise": "Unsettled. If either group has fewer than 3 SBAs the verdict is no_test.",
"refute_if": "The lower limit of the difference is above 0.",
"secondary_registered": "Reported, not deciding: (b) the same contrast for the change, in points, of the share of low-income renter households (at or below 80% of area median income) paying 30% or more of income in rent, 2005-2009 to 2020-2024 (rburden_mod_ami_le80); (c) totals of the poor count in 2000 and 2022-2024 for gentrified low-start, other low-start, the other 33 SBAs and all 55.",
"support_if": "The upper limit of the difference (gentrified minus not) is below 0."
},
"prior": 0.5,
"prior_note": "Probability that the upper limit is below 0. About 0.03 for contested, the rest unsettled or no_test.",
"registered_at": "2026-11-13T07:47:06-0500",
"registrant": "Claude (affordability agent)",
"sha256": "48a9801226fd5e68c65363e8728563631b39983abda37e120af9b668f3515aa3",
"statement": "Among the 22 low-rent sub-borough areas of 2000, the number of residents below the poverty line fell more between 2000 and 2022-2024 in the areas that gentrified (as defined in afford-gentrification-prevalence) than in those that did not."
},
"registration_commit": {
"commit": "584e57e3a7c546b2c87aa670be25ecc539ee0e99",
"committed_at": "2026-11-13T08:05:56-05:00"
},
"registration_file_sha256": "24e0a14398397e7d66fef039100c476cf12d7fda23fea88f670c0cd146688371",
"registration_path": "claims/registered/afford-poor-count-shift.json",
"registration_sha256": "48a9801226fd5e68c65363e8728563631b39983abda37e120af9b668f3515aa3",
"registration_text": "{\n \"id\": \"afford-poor-count-shift\",\n \"kind\": \"confirmation\",\n \"statement\": \"Among the 22 low-rent sub-borough areas of 2000, the number of residents below the poverty line fell more between 2000 and 2022-2024 in the areas that gentrified (as defined in afford-gentrification-prevalence) than in those that did not.\",\n \"graph_evidence\": [\n {\n \"association_id\": \"variable-name search 'gentrification', 'gentrifying', 'gentrified'\",\n \"concepts\": [\n \"gentrification (variable names)\",\n \"health and injury outcomes\"\n ],\n \"n_findings\": 33,\n \"n_studies\": 15,\n \"direction\": \"0\",\n \"summary\": \"The graph holds 33 findings in 15 studies with gentrification as a variable name; all concern health or injury outcomes and none concerns how often low-rent neighbourhoods gentrify. The studies' own counts, read in the abstracts, give a prevalence: Alroy 2023 classed 14 of 42 New York neighbourhoods as gentrifying or hypergentrifying (rent growth 2000 to 2017); Bhavsar 2022 found that 5 of 28 eligible block groups in Durham gentrified; Spitzer 2023 classed 23.8% of 59,379 urban tracts as gentrifying. Gao 2023 found opposite signs for severe maternal morbidity under two definitions (OR 0.89 and 1.18), and Bhavsar reports that results depend on the definition.\",\n \"pubmed_ids\": [\n 37018221,\n 36367672,\n 37728889,\n 37678111,\n 24022181,\n 29272306,\n 29709759\n ]\n },\n {\n \"association_id\": \"variable-name search 'displacement', 'eviction', 'rent burden', 'housing cost burden'\",\n \"concepts\": [\n \"displacement, eviction, rent burden (variable names)\",\n \"health outcomes\"\n ],\n \"n_findings\": 150,\n \"n_studies\": 60,\n \"direction\": \"+\",\n \"summary\": \"About 150 findings in about 60 studies (collection still running at registration; counts approximate). Lim 2017 followed 12,882 residents of gentrifying New York neighbourhoods who had hospital records: 23% were displaced between 2006 and 2014, and the displaced had 1.8 times (1.5 to 2.2) the rate of emergency visits and hospitalisations, mainly for mental health. Schwartz 2022: New York evictions in 2017 and loss of Medicaid coverage, OR 1.63 (1.38 to 1.92). Graetz 2024: an eviction judgment and 40% (36 to 43) higher mortality; the graph stores the percentages 40, 19, 16 and 12 as hazard ratios. The graph holds no finding that compares the rate at which low-income residents leave gentrifying and non-gentrifying neighbourhoods, which is the disputed question (Freeman and Braconi 2004; Ding, Hwang and Divringi 2016; Dragan, Ellen and Glied 2020).\",\n \"pubmed_ids\": [\n 29272306,\n 35000688,\n 38007965,\n 36120757\n ]\n }\n ],\n \"graph_evidence_note\": \"As for afford-gentrification-prevalence. The graph has no finding on counts of poor residents and none that compares exit rates of low-income residents between gentrifying and other poor neighbourhoods. Lim 2017 gives the only number: 23% of a hospital-using cohort in gentrifying New York neighbourhoods moved to a lower-income area or out in eight years, with no comparison group in the graph record.\",\n \"literature\": [\n {\n \"citation\": \"NYU Furman Center. State of New York City's Housing and Neighborhoods in 2015, Focus on Gentrification. 2016.\",\n \"doi\": null,\n \"finding\": \"15 of 55 sub-borough areas classed as gentrifying (low income in 1990, rent growth above the median 1990 to 2010-2014), 7 as non-gentrifying, 33 as higher-income.\"\n },\n {\n \"citation\": \"Freeman L, Braconi F. Gentrification and displacement: New York City in the 1990s. J Am Plann Assoc 2004;70:39-52.\",\n \"doi\": \"10.1080/01944360408976337\",\n \"finding\": \"Poor households in gentrifying New York neighbourhoods were less likely to move than poor households elsewhere.\"\n },\n {\n \"citation\": \"Dragan K, Ellen IG, Glied S. Does gentrification displace poor children and their families? Reg Sci Urban Econ 2020;83:103481.\",\n \"doi\": \"10.1016/j.regsciurbeco.2019.103481\",\n \"finding\": \"Low-income children in New York born 2006 to 2008 in gentrifying areas were no more likely to move than those in areas that stayed poor.\"\n }\n ],\n \"prediction\": {\n \"direction\": \"-\",\n \"magnitude\": \"Difference in mean log change of the poor count, gentrified minus not gentrified, between -0.45 and -0.10; best guess -0.25 (about 22% fewer poor residents relative to the comparison areas).\",\n \"definitions\": \"Poor count: pop_pov_pct / 100 x pop_num. Change: log of the mean-of-2022-2024 count over the 2000 count. Groups: the low-start SBAs classed gentrified and not gentrified by the primary definition of afford-gentrification-prevalence (rent growth and college change from 2000, both above the city's). Interval: Welch t, 95%.\",\n \"support_if\": \"The upper limit of the difference (gentrified minus not) is below 0.\",\n \"refute_if\": \"The lower limit of the difference is above 0.\",\n \"otherwise\": \"Unsettled. If either group has fewer than 3 SBAs the verdict is no_test.\",\n \"secondary_registered\": \"Reported, not deciding: (b) the same contrast for the change, in points, of the share of low-income renter households (at or below 80% of area median income) paying 30% or more of income in rent, 2005-2009 to 2020-2024 (rburden_mod_ami_le80); (c) totals of the poor count in 2000 and 2022-2024 for gentrified low-start, other low-start, the other 33 SBAs and all 55.\"\n },\n \"plan\": \"Same script and data as afford-gentrification-prevalence (analyses/affordability/gentrification_test.py, claim2 block), run unchanged. This claim is registered before that script has been run on the real panel, so the group membership is not known at registration except through the leak described under already_seen.\",\n \"can_it_fail\": {\n \"assumption\": \"The spread of the log change in the poor count across SBAs is not known before the data are read; I assume a standard deviation between 0.20 and 0.35.\",\n \"standard_error\": \"With 8 gentrified and 14 other areas the standard error of the difference is 0.09 (sd 0.20), 0.11 (sd 0.25) or 0.16 (sd 0.35); with 5 and 17 it is 0.10, 0.13 or 0.18.\",\n \"power\": \"For a true difference of -0.30 the support condition is met with probability 0.90 (sd 0.20, 8 areas), 0.73 (sd 0.25, 8 areas), 0.60 (sd 0.25, 5 areas) or 0.34 to 0.43 (sd 0.35). With no true difference it is met with probability 0.025. The smallest difference detectable with 80% power is 0.26 to 0.52.\",\n \"reading\": \"The test can fail. An unsettled verdict is likely if the true difference is under 0.25.\"\n },\n \"already_seen\": \"Read before registration: the 2000 values of every indicator for the 55 sub-borough areas and the city (rent, income, college share, poverty, population), which fix the low-start group; the workbook's column names, data dictionary and the count of non-missing cells by year. Not read directly: any value after 2000 for rent, college share, poverty or population. One leak occurred. The code test (`gentrification_test.py --simulate`) was meant to replace later-year values with noise; it multiplied the real values by lognormal noise instead (sigma 0.3 on each annual value, seed 0), and I read its output: k = 6, k2 = 7, Spearman 0.38, and the list of areas it classed as gentrified (Brooklyn Heights/Fort Greene, Bushwick, Coney Island, Lower East Side/Chinatown, North Crown Heights/Prospect Heights, Williamsburg/Greenpoint). That output is a noisy view of the outcome: the noise is about as large as the real variation between areas. The thresholds (k of 12 or more, 4 to 11, 1 or fewer) and the script were written before that run and were not changed after it. Prior knowledge: the Furman Center's 2016 report classed 15 of 22 low-income sub-borough areas as gentrifying on rent growth above the median from 1990 to 2014, with one condition (rent) where this test has two; Alroy 2023 classed 14 of 42 neighbourhoods as gentrifying or hypergentrifying on rent growth from 2000 to 2017. Also seen before registration, none of them a deciding variable: executed evictions by community district and year (6z8x-wfk4), IRS county outflows, and the graph findings listed. For this claim: no poverty rate or population after 2000 was read, in any form; the simulated run printed only the first claim's block before the output was cut at 70 lines.\",\n \"alternatives\": [\n \"A fall in the poor count is not displacement. Poor residents can leave, die, or stop being poor; the count cannot tell these apart.\",\n \"The poverty line is national and takes no account of rent, so a household above it may still be unable to pay a higher rent.\",\n \"Citywide poverty fell over the period; the comparison with other low-rent areas removes the common part.\",\n \"New housing adds residents who are not poor and lowers the rate without changing the count; the test uses the count for that reason.\",\n \"Public housing and regulated units keep poor residents in place in some gentrifying areas (Lower East Side, Harlem) and weaken the contrast.\",\n \"The 2000 poverty count is from the census long form and the later ones from the ACS, which measure income over different reference periods; this affects both groups alike.\"\n ],\n \"prior\": 0.5,\n \"prior_note\": \"Probability that the upper limit is below 0. About 0.03 for contested, the rest unsettled or no_test.\",\n \"registrant\": \"Claude (affordability agent)\",\n \"registered_at\": \"2026-11-13T07:47:06-0500\",\n \"sha256\": \"48a9801226fd5e68c65363e8728563631b39983abda37e120af9b668f3515aa3\"\n}",
"statement": "Among the 22 low-rent sub-borough areas of 2000, the number of residents below the poverty line fell more between 2000 and 2022-2024 in the areas that gentrified (as defined in afford-gentrification-prevalence) than in those that did not."
}History in this collection
- addedv1.0.0Nov 14, 2026this version