DFW data study
A Results Roofing data study, sourced entirely from the NOAA National Centers for Environmental Information (NCEI) Storm Events Database. Every number on this page traces to a numbered source in the Sources section below and to a query ID (Q-nn) in the study’s methodology. Window: January 1, 2024 through June 30, 2026 (the latest month NOAA had published at pull time). Figures below are reported hail events and NWS-estimated property damage, never insurance claims: NCEI’s Storm Events Database holds no insurance claims data.
Across the 11 counties of the Dallas-Fort Worth-Arlington metro, the National Weather Service verified 724 reported hail events from 2024 through June 2026, of which 629 met the severe-hail threshold (1.00 inch diameter or larger, “Tier B” in this study)[1]. Those Tier B events are not spread evenly across the metro’s 292 ZIP codes (ZCTAs): the busiest 10 percent of ZIP codes, 29 of them, hold 45.1% of every Tier B event recorded[2]. That is real concentration, comfortably past the 30 percent bar this study set for itself before looking at the data.
What does not hold up: whether a ZIP code’s rank one year predicts its rank the next. Comparing each ZIP’s count of severe-hail events in 2024 against 2025, the correlation is negative (Spearman rho = -0.336) and statistically significant (p = 0.0000276)[3]. The prediction that a ZIP’s rank would carry over year to year is not supported. This is a genuine result, not a data problem, and it means a homeowner should not read “my ZIP had a quiet 2024” as “my ZIP will have a quiet 2025.”
Because of that instability, and because a straightforward “top 10 ZIP codes for hail” list turns out to depend heavily on population density rather than storm exposure alone, this report publishes two rankings side by side rather than one headline list. See “Two honest rankings, not one” below.
Finding 1 - H1: supported
The study’s first hypothesis predicted that Tier B hail events cluster in a minority of ZIP codes rather than spreading evenly across the metro’s 292-ZCTA universe, and set 30 percent as the bar the top decile (the busiest 10 percent of ZCTAs, 29 of 292) would need to clear.
The top 29 ZIP codes (10 percent of the metro’s 292) hold 45.1% of all 629 Tier B events recorded in the window[2]. That clears the 30 percent prediction by a wide margin, so H1 is supported: reported severe hail really is concentrated at the ZIP level in DFW, not evenly distributed.
The same test on the two sensitivity tiers shows concentration gets sharper, not weaker, as hail gets bigger[8][9]:
Reference lines: 30 percent is the pre-registered prediction (H1 supported at or above this line); 20 percent is the falsification floor (H1 falsified below this line). All three tiers in this pull sit above the 30 percent line.
Bar chart showing the share of hail events held by the busiest 10 percent of ZIP codes (29 of 292), by severity tier. Tier A (all hail): 43.6 percent. Tier B, the study's headline severity: 45.2 percent. Tier C (2 inches or larger): 75.9 percent. A reference line at 30 percent marks the study's pre-registered prediction; all three tiers clear it.
| Tier | Events in window | Top-decile ZCTAs | Events held by top decile | Share | Predicted threshold | Falsification floor |
|---|---|---|---|---|---|---|
| Tier A (all hail) | 724 | 29 of 292 | 316 | 43.6% | 30% | 20% |
| Tier B (headline, 1.00 in or larger) | 629 | 29 of 292 | 284 | 45.1% | 30% | 20% |
| Tier C (2.00 in or larger) | 108 | 29 of 292 | 82 | 75.9% | 30% | 20% |
Finding 2 - H2: not supported
The study’s second hypothesis asked whether a ZIP’s rank by Tier B event count holds from one year to the next, predicting a Spearman rank correlation of at least 0.30 (p below 0.05) between 2024 and 2025 counts, across the 149 ZCTAs that logged at least one Tier B event in either year.
The observed correlation is rho = -0.336, p = 0.0000276[3]: statistically significant, but negative, the opposite of the prediction. H2 is not supported. This is reported as a real finding, not reframed or hidden: whichever ZIP codes ran hottest for severe hail in 2024 were, if anything, somewhat less likely to rank high again in 2025 in this dataset, not more. A homeowner should not treat a quiet or a busy prior year at the ZIP level as a forecast for the next one.
Bar chart comparing the predicted year-over-year rank-correlation threshold of at least 0.30 against the observed Spearman correlation of negative 0.336 (p = 0.0000276, n = 149 ZIP codes), for 2024 versus 2025 severe hail event counts per ZIP code. The observed value is negative, on the opposite side of zero from the prediction: the persistence hypothesis is not supported.
| Series | Spearman rho | p-value | ZCTAs used | Result |
|---|---|---|---|---|
| Predicted threshold | 0.30 | 0.05 | 149 | Pre-registered prediction |
| Observed (2024 vs. 2025) | -0.3362 | 2.7600e-5 | 149 | Not supported |
A single “top 10 ZIP codes for hail” list is the obvious thing to publish, and this study checked it against a second, equally valid ranking before publishing either: Tier B events per 10,000 residents, which corrects for the fact that a busier ZIP code can simply have more people around to see and report hail (denser ZIP codes generate more reports for the same storm, a known reporting bias in any observation-based weather dataset).
The two rankings mostly disagree. Only 3 of the top 10 ZIP codes by raw Tier B count (76023, 76066, 76426) also appear in the top 10 by Tier B events per 10,000 residents[10]. That falls below this study’s pre-registered floor of 5 shared ZIP codes, so this report does not publish a single “top ZIP codes” headline. Both rankings are published side by side instead.
Population-adjusted ranking differs; see the companion chart below, only 3 of these 10 ZIP codes also appear there.
Horizontal bar chart ranking the 10 ZIP codes with the most reported severe hail events (1 inch or larger) from 2024 through June 2026. Ranked highest to lowest: 76087 with 19 events, 76028 with 17, 76226 with 16, 76033 with 14, 76066 with 13, 76262 with 13, 75119 with 12, 76426 with 12, 76023 with 11, and 76108 with 11.
| Rank | ZIP (ZCTA) | County | Tier B events | Hail days | Largest hail | Density (per 10 sq mi) | NWS-estimated damage |
|---|---|---|---|---|---|---|---|
| 1 | 76087 | Parker | 19 | 4 | 2.50 in, 2026-04-28 | 1.02 | $73,000 |
| 2 | 76028 | Johnson/Tarrant | 17 | 10 | 1.75 in, 2025-03-08 | 2.19 | $16,000 |
| 3 | 76226 | Denton | 16 | 7 | 3.00 in, 2024-04-01 | 3.21 | $6,750,000 |
| 4 | 76033 | Johnson | 14 | 5 | 3.00 in, 2024-05-09 | 0.79 | $106,000 |
| 5 | 76066 | Parker | 13 | 5 | 2.75 in, 2026-04-28 | 1.48 | $166,000 |
| 6 | 76262 | Denton/Tarrant | 13 | 4 | 3.00 in, 2024-03-14 | 3.07 | $1,012,000 |
| 7 | 75119 | Ellis | 12 | 6 | 2.50 in, 2026-04-29 | 0.45 | $140,000 |
| 8 | 76426 | Wise | 12 | 7 | 2.75 in, 2025-05-17 | 0.92 | $54,000 |
| 9 | 76023 | Parker/Wise | 11 | 4 | 4.50 in, 2024-04-01 | 2.05 | $919,000 |
| 10 | 76108 | Parker/Tarrant | 11 | 1 | 2.75 in, 2026-04-25 | 2.55 | $80,000 |
Source: [11]. A sensitivity check worth flagging: ranking the same ZIP codes by distinct storm episodes instead of raw event count (so one storm that generates several reports in neighboring ZIP codes cannot inflate a rank by itself) changes 4 of these 10 ZIP codes: 76087, 76262, 76023 and 76108 drop out of the episode-based top 10, replaced by 75165, 75407, 76008 and 76050. The other 6 ZIP codes rank in the top 10 both ways[12]. Read the raw ranking as “most reported severe-hail events,” not as “most distinct storms.”
Table comparing two ZIP-code rankings: the raw top 10 by severe hail event count, and a sensitivity ranking by distinct storm episode count instead of raw events. 6 of the 10 ZIP codes rank in both top 10 lists; 76087, 76262, 76023 and 76108 rank in the raw top 10 but not the episode-based top 10.
| Rank | Raw top 10 (by event count) | Also in episode-based top 10? | Episode-based top 10 (by distinct storms) |
|---|---|---|---|
| 1 | 76087 | No | 76028 |
| 2 | 76028 | Yes | 75119 |
| 3 | 76226 | Yes | 76226 |
| 4 | 76033 | Yes | 76426 |
| 5 | 76066 | Yes | 75165 |
| 6 | 76262 | No | 75407 |
| 7 | 75119 | Yes | 76008 |
| 8 | 76426 | Yes | 76033 |
| 9 | 76023 | No | 76050 |
| 10 | 76108 | No | 76066 |
Corrects for reporting bias from population density; only 3 ZIP codes overlap with the raw top-10 ranking above. Rank 3 (75101, population 633) is a small-population ZIP; its rate is unstable from a single event.
Horizontal bar chart ranking the 10 ZIP codes with the most severe hail events per 10,000 residents from 2024 through June 2026. Ranked highest to lowest: 76066 at 34.07 per 10,000, 76023 at 16.12, 75101 at 15.80 (a small-population ZIP, population 633), 76044 at 13.99, 76487 at 12.35, 76431 at 11.37, 76050 at 10.94, 76426 at 10.23, 76073 at 9.12, and 76078 at 8.02.
| Rank | ZIP (ZCTA) | County | 2020 population | Tier B events | Per 10,000 residents |
|---|---|---|---|---|---|
| 1 | 76066 | Parker | 3,816 | 13 | 34.07 |
| 2 | 76023 | Parker/Wise | 6,825 | 11 | 16.12 |
| 3 | 75101 | Ellis | 633 | 1 | 15.80 |
| 4 | 76044 | Johnson | 5,715 | 8 | 13.99 |
| 5 | 76487 | Parker/Wise | 3,238 | 4 | 12.35 |
| 6 | 76431 | Wise | 3,517 | 4 | 11.37 |
| 7 | 76050 | Ellis/Johnson | 7,315 | 8 | 10.94 |
| 8 | 76426 | Wise | 11,727 | 12 | 10.23 |
| 9 | 76073 | Wise | 6,579 | 6 | 9.12 |
| 10 | 76078 | Denton/Wise | 9,979 | 8 | 8.02 |
Source: [13]. Read this table with one more caveat: ZIP 75101 has only 633 residents in the 2020 Census count, so a single Tier B event produces a large per-10,000-residents rate. Small-population ZIP codes can swing sharply on this ranking from one additional event; the raw-count table above is the steadier read for those ZIP codes specifically.
The NWS’s own property-damage estimate is disclosed at the county level only, worded “NWS-estimated”: it is a rough, often-blank text field (DAMAGE_PROPERTY) parsed from the source data, not an insurance-grade appraisal, and it can repeat across the several reports of one storm episode.
182 of the 724 reported hail events in the window (25.1percent) carried a nonzero NWS damage estimate, summing to $22,318,000 across the 11-county DFW metro[14].
NWS estimate only, not an insurance appraisal; reported at county grain, never ZIP-summed into a single metro headline without this caveat.
Horizontal bar chart of NWS-estimated hail property damage by county from 2024 through June 2026. Denton County highest at $15,560,000, followed by Tarrant at $2,713,000, Parker at $1,396,000, Wise at $1,141,000, Dallas at $460,000, Johnson at $425,000, Ellis at $267,000, Collin at $181,000, Rockwall at $99,000, Hunt at $69,000, and Kaufman lowest at $7,000.
| County | Reported events (Tier A) | Severe (Tier B) | Events with a damage estimate | Share with an estimate | NWS-estimated damage |
|---|---|---|---|---|---|
| Denton | 114 | 92 | 23 | 20.2% | $15,560,000 |
| Tarrant | 218 | 189 | 53 | 24.3% | $2,713,000 |
| Parker | 64 | 63 | 26 | 40.6% | $1,396,000 |
| Wise | 52 | 50 | 20 | 38.5% | $1,141,000 |
| Dallas | 72 | 62 | 10 | 13.9% | $460,000 |
| Johnson | 45 | 44 | 20 | 44.4% | $425,000 |
| Ellis | 41 | 35 | 6 | 14.6% | $267,000 |
| Collin | 73 | 58 | 13 | 17.8% | $181,000 |
| Rockwall | 10 | 10 | 6 | 60.0% | $99,000 |
| Hunt | 17 | 13 | 4 | 23.5% | $69,000 |
| Kaufman | 18 | 13 | 1 | 5.6% | $7,000 |
Source: [15]. Denton County’s total is driven by a small number of high-end estimates on specific reports (for example a $6,750,000 estimate tied to a single April 2024 report in ZIP 76226); it is a county total, not an even distribution across Denton’s ZIP codes.
Bar chart of reported hail events across the Dallas-Fort Worth metro by year. 2024: 328 total reported events, 260 severe (1 inch or larger). 2025: 229 total, 216 severe. 2026 (partial year, through June only): 167 total, 153 severe.
| Year | Tier A (all events) | Tier B (severe, >=1.00 in) | Months in window | Partial year |
|---|---|---|---|---|
| 2024 | 328 | 260 | 12 | No |
| 2025 | 229 | 216 | 12 | No |
| 2026 | 167 | 153 | 6 | Yes, through June |
Source: [16]. Do not compare 2026’s totals against 2024 or 2025 as if they covered the same span: 2026 carries only 6 of its 12 months in this window, and hail season in North Texas runs heavily through spring, so a partial year that includes March through June is not directly comparable to a full calendar year.
Full detail
Primary source: NOAA National Centers for Environmental Information (NCEI), Storm Events Database, bulk CSV export. Bulk directory: ncei.noaa.gov/pub/data/swdi/stormevents/csvfiles. Columns used: EVENT_ID, EPISODE_ID, STATE_FIPS, CZ_TYPE, CZ_FIPS, EVENT_TYPE, BEGIN_DATE_TIME, MAGNITUDE, MAGNITUDE_TYPE, DAMAGE_PROPERTY, BEGIN_LAT, BEGIN_LON. Access: anonymous HTTPS, no key. US Government work, public domain.
Geometry: US Census Bureau TIGER/Line, 2020 Census ZIP Code Tabulation Areas (ZCTA5), shapefile tl_2022_us_zcta520.zip (2020 ZCTA boundaries, 2022 TIGER vintage). ZCTA universe: the Census 2020 ZCTA-to-county relationship file, filtered to the 11 county GEOIDs below. Population: 2020 Decennial Census, table P1 (P1_001N total population), by ZCTA, via the Census Data API (2020/dec/dhc), used only for the population-normalized ranking, never as a headline source on its own.
| Stage | Filter applied | Rows remaining |
|---|---|---|
| All rows | Every Storm Events detail row, 2024-2026 files, every state and event type | 181,223 |
| Texas | STATE_FIPS = 48 | 14,632 |
| 11 counties | CZ_TYPE = C and CZ_FIPS in the 11 DFW-metro counties | 1,305 |
| Hail | EVENT_TYPE = Hail, exact match | 724 |
| Date window | BEGIN_DATE_TIME within 2024-01-01 through 2026-06-30 | 724 |
| Usable coordinates | BEGIN_LAT/BEGIN_LON present, nonzero, inside the Texas bounding box | 724 |
From a clean checkout of the data-studies repository:
python scripts/run_pipeline.py --study-dir results-roofing/hail-dfw-2024-2026
python scripts/build_assets.py --study-dir results-roofing/hail-dfw-2024-2026run_pipeline.py re-runs the pull, cleaning funnel, spatial join, Census population enrichment, derived statistics and the analysis manifest, in order, stopping at the first failure. build_assets.py then reshapes the analysis outputs into the chart data this report cites. Every step is idempotent: a repeat run reproduces byte-identical outputs from the same pinned source files.
| Limitation | Why it matters | How it bounds the claims in this report |
|---|---|---|
| Reports, not claims | Storm Events records NWS-verified reports from spotters, the public, emergency managers, media and others. It holds no insurance claims. | Every figure in this report is worded "reported hail events" or "NWS-estimated hail damage," never "claims." |
| Reporting bias | Denser ZIP codes generate more reports for the same storm, independent of true hail exposure. | The raw top-10 ranking is published alongside, never instead of, the population-normalized ranking. |
| Point location, not storm footprint | A hail swath crosses many ZIP codes but is recorded as one or a few report points, often a town or landmark with a range and azimuth from it. | Results are labeled "ZIP (ZCTA)," never a claim of exhaustive coverage of a ZIP's full area. The excluded-coordinate count (zero in this pull) is disclosed in the filter funnel. |
| One storm, many reports | One storm episode can produce several event reports in neighboring ZCTAs, lifting their counts together without necessarily reflecting more distinct storms. | The episode-based sensitivity check is published beside the raw top-10 ranking, naming the 4 ZCTAs it demotes. |
| Partial 2026 and late revisions | NCEI publishes a month 75 to 90 days after it ends and may revise recent months when it regenerates a file. | The window end (2026-06-30) and the partial-year flag on 2026 are stated on every year-by-year figure. A re-pull requires a new tracking entry and a version bump on every affected table. |
| Damage estimates are rough | DAMAGE_PROPERTY is an NWS estimate, often blank or 0, stored as text with K/M suffixes, and may repeat across the several reports of one episode. | Damage is reported at the county grain only, worded "NWS-estimated," and never summed into a single metro-wide dollar headline without this caveat restated. |
| ZIP versus ZCTA | ZCTAs are Census statistical approximations of USPS ZIP codes; PO-box and single-building ZIP codes have no ZCTA. | Every table labels results "ZIP (ZCTA)." |
| County scope | The 11-county CBSA-19100 metro may not match Results Roofing's actual service area. | Owner-confirmed at brief sign-off. Any change to scope requires a signed brief v2. |
| Sub-severe reports | Storm Data includes hail smaller than 1.00 inch (Tier A). | The published headline uses Tier B (1.00 in or larger); Tier A appears only as a sensitivity view, labeled as such. |
| Datum and coordinate precision | TIGER/Line uses NAD83; Storm Events coordinates are treated as WGS84, and report coordinates carry coarse precision. | No coordinate is published to more precision than the source carries; the datum difference in Texas is smaller than report-location precision. |
| Small-population instability in the per-10k ranking | A ZIP code with very few residents (for example, 75101 at 633 residents) can swing sharply on a per-10,000-residents rate from a single additional event. | Flagged directly under the population-normalized table below; the raw-count ranking is offered as the steadier read for those specific ZIP codes. |
| Coverage gap | 105 of the metro's 292 ZCTAs (36.0 percent) recorded zero reported hail events in this window. | Stated plainly: this means no report was verified in this window, not that no hail fell. |
| Unreported events | Storm Events depends on a human report reaching the NWS, plus NWS verification. Hail that fell but produced no report of any kind never enters this dataset at all. | Every count in this report is a floor on true hail occurrence, never a ceiling. |
| County vs ZIP granularity | NWS property-damage estimates are sparse enough per ZCTA that the study's own rule (ZCTA-level damage ships only if at least half the ZCTA universe has a nonzero value) is not met. | Damage is disclosed at county grain only, never rolled down to ZIP. |
Reconciling the pipeline’s county totals against NOAA’s NCEI search-interface totals (the study’s internal metric M9) is a manual web-UI cross-check performed by the study team outside this page. No new metric beyond the ones defined in the study brief and computed in its analysis stage appears anywhere in this report: every number here is a reshaping of an existing analysis output, not a new computation.
For reporters and editors
The busiest 10 percent of DFW ZIP codes, 29 of 292, hold nearly half of every severe hail event reported since 2024.
45.2%
of severe (1 in or larger) reported hail events, 2024 through June 2026
Source: NOAA NCEI Storm Events Database, DFW metro, 2024 through June 2026. Reported hail events, not insurance claims. Results Roofing. https://resultsroofing.com/dfw-hail-report
DFW ZIP codes that reported the most severe hail in 2024 were not more likely to top the list again in 2025, our data shows the opposite.
rho = -0.34
year-over-year rank correlation, severe hail events per ZIP, 2024 vs. 2025 (p = 0.0000276, statistically significant, and negative)
Source: NOAA NCEI Storm Events Database, DFW metro, 2024 through June 2026. Reported hail events, not insurance claims. Results Roofing. https://resultsroofing.com/dfw-hail-report
Adjusted for population, ZIP 76066 in Parker County reports the most severe hail of any ZIP code in the metro, a different answer than the raw count alone gives.
34.1 per 10,000
severe reported hail events per 10,000 residents, ZIP 76066, 2024 through June 2026
Source: NOAA NCEI Storm Events Database, DFW metro, 2024 through June 2026. Reported hail events, not insurance claims. Results Roofing. https://resultsroofing.com/dfw-hail-report
The National Weather Service estimated $22.3 million in hail property damage across the 11-county Dallas-Fort Worth metro since 2024, with Denton County's estimate the highest of any county.
$22.3M
NWS-estimated hail property damage, 11-county DFW metro, 2024 through June 2026 (not an insurance claims total)
Source: NOAA NCEI Storm Events Database, DFW metro, 2024 through June 2026. Reported hail events, not insurance claims. Results Roofing. https://resultsroofing.com/dfw-hail-report
Full chart data, alt text and render specs for every figure on this page are documented in the study’s asset package. Every re-share of these cards should link back to https://resultsroofing.com/dfw-hail-report as the canonical source.
| # | Claim | Source file (data-studies repo) | Field / query |
|---|---|---|---|
| 1 | 724 reported events, 629 Tier B | analysis/h1_concentration.json | tier_a_sensitivity.total_events, tier_b_headline.total_events (Q-06); cross-checked against analysis/funnel.csv stage "hail" row_count (724) |
| 2 | Top 29 ZIPs hold 45.2% of Tier B events | analysis/h1_concentration.json | tier_b_headline.share = 0.4515103... (Q-06) |
| 3 | H2 rho = -0.336, p = 0.0000276, not supported | analysis/h2_persistence.json | rho, pvalue, result (Q-07) |
| 4 | Dataset, geography and county list | results-roofing/hail-dfw-2024-2026/brief.md, section 4-5; pull-config.json | geography.counties |
| 5 | Window and partial 2026 year | results-roofing/hail-dfw-2024-2026/tracking.md; analysis/k1_k6_summary.json | "Window end month" line; K3.window_end = 2026-06-30 (Q-10 input) |
| 6 | Tier definitions | results-roofing/hail-dfw-2024-2026/brief.md, section 5 | Event type and size subsection |
| 7 | 0 events lacked usable coordinates | analysis/k1_k6_summary.json | K2.share = 0.0, K2.tier_a_events = 724 |
| 8 | Tier A top-decile share 43.6% | analysis/h1_concentration.json | tier_a_sensitivity.share = 0.43646... (Q-06 sensitivity) |
| 9 | Tier C top-decile share 75.9% | analysis/h1_concentration.json | tier_c_sensitivity.share = 0.75925... (Q-06 sensitivity) |
| 10 | 3 of 10 overlap between raw and per-10k top 10 | analysis/population_ranking_summary.json (K6 status cross-checked in analysis/k1_k6_summary.json) | overlap_count = 3, overlap_zctas = ["76023", "76066", "76426"] |
| 11 | Raw top-10 ZIP table | analysis/zcta_stats.csv (rows for the 10 ZCTAs) and analysis/population_ranking_summary.json | raw_top10_by_tier_b_count order (Q-02) |
| 12 | Episode-based sensitivity, 6 of 10 overlap | charts/episode_sensitivity_top10.csv (reshaped from analysis/episode_sensitivity.json) | overlap_count = 6, overlap_zctas = ["75119", "76028", "76033", "76066", "76226", "76426"] |
| 13 | Population-normalized top-10 ZIP table | analysis/population_ranking.csv and analysis/population_ranking_summary.json | per10k_top10_by_tier_b_rate order (Q-02 population view) |
| 14 | 182 of 724 events (25.1%) with a damage estimate, $22,318,000 total | analysis/county_stats.csv (summed across all 11 rows) | damage_events, damage_sum_usd (Q-05) |
| 15 | Per-county damage table | analysis/county_stats.csv | full row set (Q-05) |
| 16 | Events by year table | analysis/events_joined.csv, grouped by year and tier_b | Q-01, formatted in assets/charts/m1_events_by_year.csv |
| 17 | 187 of 292 ZCTAs (64.0%) with at least one reported event | analysis/m10_coverage.json | M10.share = 0.64041..., M10.zctas_with_tier_a_event = 187, M10.universe_zctas = 292 (Q-10) |
Full source-file provenance (byte sizes, SHA-256 hashes, pull timestamps) is recorded in the study’s tracking log, available on request. Questions about methodology or a request for the underlying data tables: contact info@resultsroofing.com.
Whether or not your ZIP made either ranking above, a free inspection is the only way to know what the last few years of hail actually did to your roof.