Claim at v1.1.0

retimed-taxi-fhv-pickups

ad2920a237b80b1b1ef5fe5e9de923ab4c2418e6260ad853cbb44e5803f92d67

Data

{
  "has_outcome": true,
  "kind": "confirmation",
  "lineage": {
    "audit": "docs/lineage_audit.md; data/derived/lineage_audit.json",
    "class": "b",
    "class_meaning": "a design, power or calibration read, pre-period data, or metadata",
    "classes": [
      "b"
    ],
    "early_sources": [
      {
        "class": "b",
        "dataset": "tlc-trip-records",
        "evidence": "Yellow-taxi pickups for February-March 2018 only, to estimate sampling error. already_seen: \"I have not tabulated 2019 or 2026 pickups by weekday.\" Referee found nothing showing 2026 counts were seen earlier.",
        "fetched": "2026-11-12T23:44:41-05:00",
        "name": "yellow trip records 2018-02"
      },
      {
        "class": "b",
        "dataset": "tlc-trip-records",
        "evidence": "Yellow-taxi pickups for February-March 2018 only, to estimate sampling error. already_seen: \"I have not tabulated 2019 or 2026 pickups by weekday.\" Referee found nothing showing 2026 counts were seen earlier.",
        "fetched": "2026-11-12T23:44:42-05:00",
        "name": "yellow trip records 2018-03"
      }
    ]
  },
  "outcome": "retimed-taxi-fhv-pickups",
  "prereg_branch": {
    "name": "prereg/retimed-taxi-fhv-pickups",
    "sha": "e6bf43b2d9b07b28713da586eb27ae541aa1dfe8"
  },
  "prior": 0.5,
  "registered_at": "2026-11-12T23:48:46-0500",
  "registrant": "Claude (round 2, re-timed consequences agent), for Jeff Vernon",
  "registration": {
    "already_seen": "The project holds TLC pickups by zone and day for January-March 2018 and 2019 (transit-fhv-price-subway, used for totals near subway complexes, not by weekday), June-October 2023 and June-September 2024 (transit-gtrain-substitution, G-train zones), and zone totals for commute-fhv-subway-distance. None was tabulated by weekday against midweek. For this registration I read yellow and for-hire pickups for January-March 2018 only, to estimate sampling error. I have not tabulated 2019 or 2026 pickups by weekday.",
    "alternatives": [
      "Yellow taxis lost share to for-hire vehicles unevenly across zones, and the two services have different weekday patterns (separate estimates by service address this).",
      "Tourists and residents, who did not change their week, make most daytime trips in Manhattan, so pickups need not follow office attendance.",
      "Weather differed between the sampled weeks of 2019 and 2026; a storm on a Monday or Friday moves the citywide ratio, though less the difference between zones.",
      "The congestion charge from January 2025 lowered trips in the CBD on all weekdays; it biases the result only if its effect differs between Monday or Friday and midweek.",
      "MTA service changes on Mondays and Fridays alter entries without altering presence, which would weaken the slope."
    ],
    "graph_evidence": [
      {
        "concepts": "taxi or ride-hail demand | commuting, working at home, weekday",
        "direction": "none",
        "n_findings": 0,
        "summary": "System Public Graph, variable searches (data/derived/graph/retimed_week/gap_checks.json and data/derived/round2/retimed_consequences/graph_gap_check.json, graph_leads.json): no finding has this outcome with commuting, working at home, day of week or daytime population as the exposure. 82 variables match taxi, ride-hailing, ridesharing or uber. The nearest are rideshare trips per taxi zone-hour as an exposure for injury crashes (odds ratio 1.046, PMID 32253258) and an elasticity of 0.004 between ride-hailing use and transit ridership with no PubMed id. Neither is a finding on demand by day of week."
      }
    ],
    "id": "retimed-taxi-fhv-pickups",
    "kind": "confirmation",
    "literature": [
      {
        "citation": "Barrero, Bloom and Davis 2023, The evolution of work from home, J Econ Perspectives",
        "doi": "10.1257/jep.37.4.23",
        "finding": "Days worked at home rose to about four times their 2019 share; no estimate for taxi demand."
      }
    ],
    "plan": "Exposure comes from the dose map only (data/derived/transit/dose_map_long.csv, built from MTA hourly entries before any outcome was queried). For a group of station complexes, E = ln(MF/TWT in Nov 2025-Oct 2026) - ln(MF/TWT in 2019), where MF is mean Monday and Friday entries and TWT mean Tuesday-to-Thursday entries, for 06:00-19:59 unless the AM peak (06:00-09:59) is named. E is negative where Mondays and Fridays thinned. Files: data/derived/round2/retimed_consequences/exposure_taxi_zone.csv, exposure_community_district.csv, exposure_groups.json; code analyses/round2/retimed_consequences/exposure.py. Estimator (analyses/round2/retimed_consequences/lib.py, written before registration): for each unit and period, r = ln(mean Monday and Friday count / mean Tuesday-to-Thursday count) over retained weeks, and Y = r(recent) - r(baseline). Retained weeks have no federal holiday and no day from 24 December to 1 January, as in the dose map; the weeks of Monday 26 January and Monday 23 February 2026 are also dropped, because subway entries on those Mondays were 38% and 20% of the week's midweek level (docs/transit.md), which was known from entries before any outcome was seen. Intervals are the 2.5 and 97.5 percentiles of 2,000 bootstrap draws that resample weeks within each period and, for a slope across units, the units as well. Outcome: pickups with pickup hour 06 to 19 by pickup zone and date, yellow taxi plus high-volume for-hire (Uber, Lyft), from TLC trip records. This is a sample of months, not full years: February and March 2019 (7 retained weeks; high-volume for-hire records begin in February 2019; local parquet copies) against February and March 2026 (6 retained weeks; yellow from local parquet, high-volume for-hire by remote column scan of the TLC parquet files, which are not saved). Green taxis and other for-hire bases are left out. Units: the taxi zones that contain at least one reliable station complex (150) and have at least 200 pickups per weekday in the band in both periods. b is the unweighted OLS slope of Y on E. Sensitivity, reported and not deciding: slope weighted by 2019 pickups; yellow and for-hire separately; zones outside the CBD only; hours 06-09 against E for the AM peak. Placebo: yellow taxi February-March 2018 against 2019. Power, from data outside the test: random half-splits of the 10 retained weeks of January-March 2018 (yellow plus all for-hire, 147 zones) give a null standard deviation of 0.166 for b and 0.016 for the contrast; scaled to 7 and 6 weeks the expected standard errors are about 0.15 and 0.014 (design_taxi_placebo_2018.json). The smallest slope detectable with 80% power is about 0.41. At b = 0.5 the predicted contrast is -0.038, 2.7 standard errors from zero (D = 7.6, above the kit's threshold of 4). A true slope of zero would meet refute_if about half the time. The split-half error leaves out differences between zones that built up over seven years, so the real interval may be wider. Verdict by concord.claims.PRECEDENCE rules 1 to 4: support_if met gives confirmed, refute_if met gives contested, otherwise unsettled; no_test if the registered data cannot be read. Secondary and placebo results decide nothing.",
    "prediction": {
      "direction": "+",
      "magnitude": "Primary: the slope b of Y (pickups) on E (entries) across taxi zones is between 0.3 and 1.0. Secondary measure that must agree in sign: Y pooled over the third of zones with the most thinning (E <= -0.0639; pooled E = -0.109) minus Y pooled over the third with the least (E >= -0.0421; pooled E = -0.032) is between -0.02 and -0.08.",
      "refute_if": "The 95% interval of b lies below zero, or its upper bound is below 0.3 (any effect is smaller than registered).",
      "support_if": "The 95% interval of b lies above zero and the 95% interval of the top-third minus bottom-third contrast lies below zero."
    },
    "prior": 0.5,
    "registered_at": "2026-11-12T23:48:46-0500",
    "registrant": "Claude (round 2, re-timed consequences agent), for Jeff Vernon",
    "sha256": "a4208ae183234df3c7030abc703801785f69d18c46bc96534778383bc670e6b7",
    "statement": "Between February-March 2019 and February-March 2026, daytime taxi and high-volume for-hire pickups on Mondays and Fridays fell relative to Tuesday-to-Thursday pickups in the taxi zones whose subway entries thinned most on Mondays and Fridays, in proportion to between 0.3 and 1.0 of the thinning of entries."
  },
  "registration_commit": {
    "commit": "e0b4eecd70952329bfd68a44e5e44f96018c0d5e",
    "committed_at": "2026-11-12T23:48:54-05:00"
  },
  "registration_file_sha256": "904ebe298c9f11d9a11dacc86d7ee6430cbd09007e899bbe1ee6fc44786e65aa",
  "registration_path": "claims/registered/retimed-taxi-fhv-pickups.json",
  "registration_sha256": "a4208ae183234df3c7030abc703801785f69d18c46bc96534778383bc670e6b7",
  "registration_text": "{\n \"id\": \"retimed-taxi-fhv-pickups\",\n \"kind\": \"confirmation\",\n \"statement\": \"Between February-March 2019 and February-March 2026, daytime taxi and high-volume for-hire pickups on Mondays and Fridays fell relative to Tuesday-to-Thursday pickups in the taxi zones whose subway entries thinned most on Mondays and Fridays, in proportion to between 0.3 and 1.0 of the thinning of entries.\",\n \"graph_evidence\": [\n  {\n   \"concepts\": \"taxi or ride-hail demand | commuting, working at home, weekday\",\n   \"n_findings\": 0,\n   \"direction\": \"none\",\n   \"summary\": \"System Public Graph, variable searches (data/derived/graph/retimed_week/gap_checks.json and data/derived/round2/retimed_consequences/graph_gap_check.json, graph_leads.json): no finding has this outcome with commuting, working at home, day of week or daytime population as the exposure. 82 variables match taxi, ride-hailing, ridesharing or uber. The nearest are rideshare trips per taxi zone-hour as an exposure for injury crashes (odds ratio 1.046, PMID 32253258) and an elasticity of 0.004 between ride-hailing use and transit ridership with no PubMed id. Neither is a finding on demand by day of week.\"\n  }\n ],\n \"literature\": [\n  {\n   \"citation\": \"Barrero, Bloom and Davis 2023, The evolution of work from home, J Econ Perspectives\",\n   \"doi\": \"10.1257/jep.37.4.23\",\n   \"finding\": \"Days worked at home rose to about four times their 2019 share; no estimate for taxi demand.\"\n  }\n ],\n \"prediction\": {\n  \"direction\": \"+\",\n  \"magnitude\": \"Primary: the slope b of Y (pickups) on E (entries) across taxi zones is between 0.3 and 1.0. Secondary measure that must agree in sign: Y pooled over the third of zones with the most thinning (E <= -0.0639; pooled E = -0.109) minus Y pooled over the third with the least (E >= -0.0421; pooled E = -0.032) is between -0.02 and -0.08.\",\n  \"support_if\": \"The 95% interval of b lies above zero and the 95% interval of the top-third minus bottom-third contrast lies below zero.\",\n  \"refute_if\": \"The 95% interval of b lies below zero, or its upper bound is below 0.3 (any effect is smaller than registered).\"\n },\n \"plan\": \"Exposure comes from the dose map only (data/derived/transit/dose_map_long.csv, built from MTA hourly entries before any outcome was queried). For a group of station complexes, E = ln(MF/TWT in Nov 2025-Oct 2026) - ln(MF/TWT in 2019), where MF is mean Monday and Friday entries and TWT mean Tuesday-to-Thursday entries, for 06:00-19:59 unless the AM peak (06:00-09:59) is named. E is negative where Mondays and Fridays thinned. Files: data/derived/round2/retimed_consequences/exposure_taxi_zone.csv, exposure_community_district.csv, exposure_groups.json; code analyses/round2/retimed_consequences/exposure.py. Estimator (analyses/round2/retimed_consequences/lib.py, written before registration): for each unit and period, r = ln(mean Monday and Friday count / mean Tuesday-to-Thursday count) over retained weeks, and Y = r(recent) - r(baseline). Retained weeks have no federal holiday and no day from 24 December to 1 January, as in the dose map; the weeks of Monday 26 January and Monday 23 February 2026 are also dropped, because subway entries on those Mondays were 38% and 20% of the week's midweek level (docs/transit.md), which was known from entries before any outcome was seen. Intervals are the 2.5 and 97.5 percentiles of 2,000 bootstrap draws that resample weeks within each period and, for a slope across units, the units as well. Outcome: pickups with pickup hour 06 to 19 by pickup zone and date, yellow taxi plus high-volume for-hire (Uber, Lyft), from TLC trip records. This is a sample of months, not full years: February and March 2019 (7 retained weeks; high-volume for-hire records begin in February 2019; local parquet copies) against February and March 2026 (6 retained weeks; yellow from local parquet, high-volume for-hire by remote column scan of the TLC parquet files, which are not saved). Green taxis and other for-hire bases are left out. Units: the taxi zones that contain at least one reliable station complex (150) and have at least 200 pickups per weekday in the band in both periods. b is the unweighted OLS slope of Y on E. Sensitivity, reported and not deciding: slope weighted by 2019 pickups; yellow and for-hire separately; zones outside the CBD only; hours 06-09 against E for the AM peak. Placebo: yellow taxi February-March 2018 against 2019. Power, from data outside the test: random half-splits of the 10 retained weeks of January-March 2018 (yellow plus all for-hire, 147 zones) give a null standard deviation of 0.166 for b and 0.016 for the contrast; scaled to 7 and 6 weeks the expected standard errors are about 0.15 and 0.014 (design_taxi_placebo_2018.json). The smallest slope detectable with 80% power is about 0.41. At b = 0.5 the predicted contrast is -0.038, 2.7 standard errors from zero (D = 7.6, above the kit's threshold of 4). A true slope of zero would meet refute_if about half the time. The split-half error leaves out differences between zones that built up over seven years, so the real interval may be wider. Verdict by concord.claims.PRECEDENCE rules 1 to 4: support_if met gives confirmed, refute_if met gives contested, otherwise unsettled; no_test if the registered data cannot be read. Secondary and placebo results decide nothing.\",\n \"already_seen\": \"The project holds TLC pickups by zone and day for January-March 2018 and 2019 (transit-fhv-price-subway, used for totals near subway complexes, not by weekday), June-October 2023 and June-September 2024 (transit-gtrain-substitution, G-train zones), and zone totals for commute-fhv-subway-distance. None was tabulated by weekday against midweek. For this registration I read yellow and for-hire pickups for January-March 2018 only, to estimate sampling error. I have not tabulated 2019 or 2026 pickups by weekday.\",\n \"alternatives\": [\n  \"Yellow taxis lost share to for-hire vehicles unevenly across zones, and the two services have different weekday patterns (separate estimates by service address this).\",\n  \"Tourists and residents, who did not change their week, make most daytime trips in Manhattan, so pickups need not follow office attendance.\",\n  \"Weather differed between the sampled weeks of 2019 and 2026; a storm on a Monday or Friday moves the citywide ratio, though less the difference between zones.\",\n  \"The congestion charge from January 2025 lowered trips in the CBD on all weekdays; it biases the result only if its effect differs between Monday or Friday and midweek.\",\n  \"MTA service changes on Mondays and Fridays alter entries without altering presence, which would weaken the slope.\"\n ],\n \"prior\": 0.5,\n \"registrant\": \"Claude (round 2, re-timed consequences agent), for Jeff Vernon\",\n \"registered_at\": \"2026-11-12T23:48:46-0500\",\n \"sha256\": \"a4208ae183234df3c7030abc703801785f69d18c46bc96534778383bc670e6b7\"\n}",
  "statement": "Between February-March 2019 and February-March 2026, daytime taxi and high-volume for-hire pickups on Mondays and Fridays fell relative to Tuesday-to-Thursday pickups in the taxi zones whose subway entries thinned most on Mondays and Fridays, in proportion to between 0.3 and 1.0 of the thinning of entries."
}

History in this collection

  1. addedv1.0.0Nov 14, 2026this version