CDC NCHS - Provisional COVID-19 Deaths by Week and State
- Rows
- 24,084
- Columns
- 17
- Data range
- 2026-10-29 to 2026-10-29
- Auto-refresh
- Re-checks its source every 7 days
- Last refreshed with new data
- 2026-11-05
Weekly provisional death counts involving COVID-19, pneumonia, and influenza by state of residence.
AI analysis
Dataset Analysis: CDC NCHS Provisional COVID-19 Deaths by Week and State
1. Dataset Overview and Research Value
This dataset, sourced directly from the CDC's public API at https://data.cdc.gov/api/views/r8kw-7aab/rows.csv, represents provisional weekly mortality counts involving COVID-19, pneumonia, and influenza, broken down by U.S. state of residence. The CDC/NCHS provenance signals high institutional reliability, and the date range spanning 2020–2026 (mean year ~2022.72) indicates the dataset captures the full arc of the COVID-19 pandemic through near-present provisional reporting. With 22,626 rows across 54 distinct state/jurisdiction values and 400 distinct start dates, this dataset offers substantial temporal and geographic granularity, making it well-suited for time-series correlation studies, cross-state comparative analysis, and multi-disease mortality trend research. The "provisional" nature of the data is an important caveat — recent weeks are subject to revision as death certificates are processed.
2. Data Quality Observations
The dataset carries notable quality concerns that require careful preprocessing. Across 17 columns and 22,626 rows, there are 62,077 total null cells, implying an average null rate of roughly 16% per cell — but the distribution is highly uneven. The Month column is the most problematic, with 18,468 nulls (81.6%), rendering it nearly unusable as a standalone temporal feature; analysts should rely on Start Date, End Date, or Week Ending Date instead. Influenza Deaths (41.4% nulls) and Pneumonia and COVID-19 Deaths (38.7% nulls) also have severe missingness, likely reflecting suppressed small-count values per CDC data privacy protocols rather than random missingness — a critical distinction that means imputation would be inappropriate and these fields should be treated as structurally censored. COVID-19 Deaths has 6,092 nulls (27%), and Total Deaths has just 1 null, making it the most complete mortality measure. Duplicate row counts are noted as pending recomputation against Parquet, so deduplication validation should be completed before any aggregation work. The Year column has 378 nulls alongside only 7 distinct values (2020–2026), suggesting some rows lack year attribution despite having date fields — a minor inconsistency worth flagging.
3. Key Column Distributions
The mortality columns reveal dramatically skewed distributions that demand attention. COVID-19 Deaths has a mean of 602.9 but a median of only 45, with a skewness of 86.9 and 2,237 outliers — the max value of ~1.25M strongly suggests this column includes aggregated national or cumulative rows (consistent with the Group column having 4 distinct values, likely including "By Total" or cumulative groupings). Total Deaths is even more extreme: mean of 7,325 vs. median of 1,077, skewness of 116, and a max of ~20.6M, confirming multi-level aggregation is present in the dataset. Pneumonia Deaths follows a similar pattern (skew=108, 2,468 outliers). By contrast, Percent of Expected Deaths is comparatively well-behaved — mean 113.0, median 111, σ of 16.9 — indicating most weeks/states ran about 11–13% above baseline mortality expectations, with outlier spikes up to 737% likely corresponding to early pandemic surge weeks. Influenza Deaths has a median of 0 and mean of 35.6 (skew=82.4), reflecting the well-documented collapse of influenza transmission during COVID mitigation periods. Analysts should filter by the Group column first to isolate weekly state-level rows before computing any statistics, as mixing aggregation levels is the primary driver of these extreme distributions.
4. Recommended Join Key Columns
For cross-dataset correlation, the most reliable join strategy uses a composite key of State + Week Ending Date (or equivalently Start Date/End Date), which provides unambiguous geographic and temporal anchoring. State has 54 distinct values (50 states + DC + territories + national aggregate), so joining datasets should align on the same jurisdiction list. MMWR Week + Year offers an alternative temporal join aligned with CDC's standard epidemiological calendar, which is widely used across public health datasets — though the 4,590 nulls in MMWR Week require attention. The Data as of column, while having only 1 distinct value (the extraction snapshot date), serves as a useful metadata tag for tracking data vintage. Avoid using Month as a join key given its 81.6% null rate.
5. Suggested Complementary Datasets for Correlation Analysis
Several dataset categories would pair powerfully with this data. COVID-19 vaccination coverage by state and week (also available via CDC's data portal) would enable direct correlation between vaccination uptake and mortality trends — one of the most policy-relevant analyses possible. U.S. Census Bureau population estimates by state would allow normalization to per-capita death rates, which is essential for meaningful cross-state comparisons given the vast size differences between, say, California and Wyoming. Hospital capacity and ICU utilization data (e.g., HHS Protect datasets) could reveal correlations between healthcare system strain and excess mortality as measured by the Percent of Expected Deaths column. Socioeconomic and demographic data — including age distribution, poverty rates, and comorbidity prevalence by state — would support multivariate regression to identify structural vulnerability factors. Finally, Google or Apple mobility data from the pandemic period could help correlate behavioral changes (lockdowns, reopening events) with the weekly mortality spikes visible in this dataset, providing a behavioral layer to the biological signal.
Columns
- Data as of (date)
- Start Date (date)
- End Date (date)
- Group (string)
- Year (integer)
- Month (string)
- MMWR Week (integer)
- Week Ending Date (date)
- State (string)
- COVID-19 Deaths (integer)
- Total Deaths (integer)
- Percent of Expected Deaths (integer)
- Pneumonia Deaths (integer)
- Pneumonia and COVID-19 Deaths (integer)
- Influenza Deaths (integer)
- Pneumonia, Influenza, or COVID-19 Deaths (integer)
- Footnote (string)