CDC – Weekly Provisional Counts of Deaths by Sex, Age, and State
- Rows
- 6,552
- Columns
- 8
- Data range
- 2023-06-28 to 2023-06-28
- Auto-refresh
- Re-checks its source every 7 days
Provisional weekly death counts stratified by sex, age group, and U.S. state from 2015 onward. Enables excess mortality estimation and demographic mortality trend analysis at the subnational level.
AI analysis
Dataset Analysis: CDC Weekly Provisional Death Counts by Sex, Age, and State
1. Dataset Overview and Value
This dataset, sourced from the CDC's official data portal (https://data.cdc.gov/api/views/vsak-wrfu/rows.csv?accessType=DOWNLOAD), represents provisional weekly mortality counts stratified by sex, age group, and U.S. state beginning in 2015. The CDC provenance signals high institutional reliability, and the "provisional" label indicates near-real-time reporting with potential for retrospective revision — an important caveat for time-sensitive analyses. With 6,552 rows spanning demographic breakdowns across all U.S. states, this dataset is exceptionally well-suited for excess mortality estimation, COVID-19 mortality attribution, and longitudinal demographic trend analysis. Its subnational granularity makes it particularly valuable for correlating mortality patterns against state-level policy interventions, healthcare infrastructure metrics, or socioeconomic indicators.
2. Data Quality Observations
Data quality here is notably strong. Total null cells across all 8 columns register at zero, which is remarkable for a provisional government dataset of this size and complexity — no imputation or null-handling strategies should be required prior to analysis. The duplicate row count is flagged as pending Phase D Parquet recomputation, which should be resolved before any aggregation work, as undetected duplicates in weekly mortality data could materially inflate death counts. One structural concern worth flagging: "Data as of" and "State" both show distinct=1, meaning the current indexed slice appears to represent a single snapshot date and a single state — this strongly suggests the full dataset has been filtered or that the indexer captured only a partition. Analysts should verify they are working with the complete multi-state, multi-date extract before drawing broad conclusions.
3. Key Column Distributions
Total Deaths is the most analytically rich column, exhibiting a strongly right-skewed distribution (skew=3.15) with a mean of 7,078 but a median of only 1,784, reflecting the wide variance between age-group/sex strata and aggregate "All" category rows. The interquartile range (Q1=235, Q3=7,706) spans nearly 7,500 deaths, and 603 outliers likely correspond to all-ages, all-sex national rollup rows or peak pandemic weeks. COVID-19 Deaths tells an even more extreme story: skew=6.19, median of just 59 versus a mean of 694, and a maximum of 26,028 — with 813 outlier rows, this column captures the severe, episodic nature of pandemic mortality spikes. The near-zero Q1 of 2 confirms that COVID-19 deaths are effectively absent in pre-2020 records and in low-mortality demographic strata. MMWR Week (min=1, max=53, mean=24.8, skew=0.19) distributes nearly uniformly across the calendar year, as expected for a weekly surveillance metric — the rare week 53 (distinct=53) reflects leap-year or calendar-boundary edge cases worth noting.
4. Recommended Join Key Columns
The columns tagged as join keys — "Data as of" (Date), "State" (String), and "MMWR Week" (Integer) — form a natural composite key for cross-dataset linkage. However, "End Week" (Date) with 182 distinct values is likely the more precise temporal anchor for joining to external weekly time-series data, as it represents the actual epidemiological week endpoint rather than a data-refresh timestamp. For demographic joins, Sex (3 categories: Male, Female, All) and Age Group (12 categories) provide stratification handles. A recommended composite join key for most correlation use cases would be: State + End Week + Sex + Age Group, which should approach row-level uniqueness and enable clean many-to-one joins against state-week panel datasets.
5. Suggested Paired Datasets for Correlation Analysis
Several dataset categories would pair powerfully with this data. CDC COVID-19 case and hospitalization counts (also on data.cdc.gov) would allow direct case-fatality rate modeling by demographic stratum and state. U.S. Census Bureau population estimates by state, age, and sex would enable normalization into mortality rates, unlocking comparability across states of vastly different sizes. State-level policy tracking datasets (e.g., Oxford COVID-19 Government Response Tracker or Kaiser Family Foundation policy databases) would support difference-in-differences analyses of mask mandates, vaccination rollouts, and lockdown stringency against excess mortality. CMS Medicare/Medicaid enrollment and claims data could reveal correlations between healthcare access and COVID-19 death rates among elderly cohorts. Finally, CDC Behavioral Risk Factor Surveillance System (BRFSS) data on comorbidity prevalence (obesity, diabetes, smoking) by state would enable multivariate mortality risk modeling across the same geographic and demographic dimensions.
Columns
- Data as of (date)
- State (string)
- MMWR Week (integer)
- End Week (date)
- Sex (string)
- Age Group (string)
- Total Deaths (integer)
- COVID-19 Deaths (integer)