NYTimes COVID-19 US Data - Counties
- Rows
- 2,502,832
- Columns
- 6
- Data range
- 2020-01-21 to 2022-05-13
- Auto-refresh
- Re-checks its source daily
US county-level daily cases and deaths time-series. Large detailed dataset.
AI analysis
NYTimes COVID-19 US Counties Dataset — Analyst Brief
1. Dataset Overview & Value This dataset, sourced from the New York Times GitHub repository (https://raw.githubusercontent.com/nytimes/covid-19-data/master/us-counties.csv), represents one of the most authoritative and granular public records of COVID-19 spread across the United States. It captures daily cumulative case and death counts at the county level, spanning 844 distinct dates and 3,220 unique FIPS-coded geographies across 56 state/territory designations. The NYT provenance signals high editorial rigor and consistent methodology, making this dataset particularly well-suited for longitudinal correlation studies — examining how case trajectories interacted with policy, demographics, healthcare capacity, or economic conditions over time. Its sheer scale (2.5M+ rows) enables both broad national trend analysis and fine-grained local comparisons.
---
2. Data Quality Observations Overall quality is strong, but two columns warrant attention. Deaths carries the most significant null burden at 57,605 nulls (~2.3%), which likely reflects early pandemic reporting gaps, counties with suppressed small-number death counts for privacy, or jurisdictions that lagged death attribution. FIPS codes show 23,678 nulls (~0.9%), a meaningful concern because FIPS is the standard geographic join key for US county-level cross-dataset work — rows missing FIPS cannot be reliably linked to external sources like Census data or hospital records without fuzzy county/state name matching. No type mismatches are flagged, and the zero-null counts on date, county, state, and cases are reassuring. Duplicate row counts are noted as pending recomputation against the Parquet layer, so deduplication should be validated before any aggregation or modeling work proceeds.
---
3. Key Column Distributions The cases column tells a dramatic story of inequality in outbreak severity. With a mean of ~10,034 but a median of only 1,773, the distribution is sharply right-skewed (skew = 25.2), and the interquartile range (Q1=382, Q3=5,884) confirms that the vast majority of counties experienced relatively modest cumulative counts while a small number of large urban counties drove enormous totals (max ~2.9M). The 317,894 flagged outliers represent rows from high-density counties like Los Angeles or Cook County that statistically dominate the distribution. Deaths mirrors this pattern even more acutely (skew = 26.4, median=33, mean=162, max=40,267, ~292,666 outliers), reinforcing that case fatality burden was extraordinarily concentrated. The date column's 844 distinct values, combined with 3,220 FIPS codes, implies near-complete daily coverage across roughly 2.3 years of the pandemic — a rich time-series structure ideal for lag-correlation and event-study analyses.
---
4. Recommended Join Keys - fips (Integer) is the gold-standard geographic join key for linking to any US government or academic county-level dataset (Census, CDC, CMS, BLS). However, given ~23,678 null FIPS rows, a composite fallback key of county + state should be maintained for rows where FIPS is absent. - date (Date) serves as the temporal join key and should be normalized to ISO 8601 format before joining to policy timelines, vaccination records, or mobility data. - state (String) enables state-level aggregation joins, though with only 56 distinct values it is better used as a grouping/filter dimension than a primary key.
---
5. Recommended Companion Datasets for Correlation Analysis
| Dataset / Topic | Join Keys | Potential Insight | |---|---|---| | US Census ACS (demographics) | FIPS | Correlate case/death rates with age, income, density, race | | CDC Vaccination Coverage by County | FIPS + date | Measure vaccination impact on case trajectory | | CMS Hospital Capacity / Beds | FIPS or state | Link death rates to local healthcare resource availability | | BLS Unemployment / Employment Data | FIPS + date | Explore economic disruption alongside outbreak severity | | Oxford COVID Policy Tracker (OxCGRT) | State + date | Correlate lockdown stringency with case suppression | | Google / Apple Mobility Reports | State/county + date | Test behavioral response to rising case counts | | Weather / Climate Data (NOAA) | FIPS + date | Investigate seasonality effects on transmission |
The combination of granular geography, daily resolution, and a trusted publisher makes this one of the most versatile anchor datasets available for pandemic-era correlation research — provided the FIPS null gap is addressed and deduplication is confirmed.
Columns
- date (date)
- fips (integer)
- county (string)
- state (string)
- cases (integer)
- deaths (integer)