COVID-19 Data Hub Raw Data Level 2
- Rows
- 980
- Columns
- 47
- Data range
- 2020-01-12 to 2022-09-17
- Auto-refresh
- Re-checks its source every 30 days
Raw state-level COVID-19 time-series data.
AI analysis
COVID-19 Data Hub – Raw State-Level Data Analysis
1. Dataset Overview & Analytical Value
This dataset, sourced from the COVID-19 Data Hub API (https://storage.covid19datahub.io/rawdata-2.csv), represents a single administrative unit's (Level 2) COVID-19 time-series spanning 980 daily observations. The geographic constants — latitude 15.818°N, longitude 104.275°E, ISO alpha-3 code 764, and population of 487,976 — strongly identify this as a Thai provincial-level record (ISO 764 = Thailand), likely a single province tracked continuously over time. The dataset's primary analytical value lies in its pairing of epidemiological outcomes (confirmed cases, deaths) with policy intervention indices (stringency, containment, government response), making it well-suited for studying the lag-effects of non-pharmaceutical interventions (NPIs) on case trajectories within a single jurisdiction.
---
2. Data Quality Observations
Data quality is significantly compromised across several dimensions. Most critically, 12 columns carry 100% null rates (980/980 nulls each), including recovered, tests, vaccines, peoplevaccinated, peoplefullyvaccinated, hosp, icu, vent, keygooglemobility, keyjhucsse, keynuts, and administrativearealevel3. These columns are entirely absent and contribute to a total null cell count of 11,760 across 47 columns — roughly 25.5% of all cells being null. Several columns (administrativearealevel, latitude, longitude, population, isonumeric, keylocal) carry zero variance (σ = 0, distinct = 1), meaning they are constants that add no analytical signal. Notably, recovered and tests being fully null is a significant limitation: vaccine uptake, testing rates, and recovery cannot be analyzed, which are essential for computing derived metrics like positivity rate or case fatality rate. The id, administrativearealevel1, administrativearealevel2, isoalpha3, isoalpha2, and isocurrency columns all show distinct = 1, further confirming this is a single-entity dataset with no cross-region variation embedded.
---
3. Key Column Distributions & Notable Statistics
The confirmed column (mean = 4,882.81, median = 105.5, σ = 7,042.73, skew = 1.25, Q1 = 1, Q3 = 6,609.5, outliers = 144) reveals a highly right-skewed cumulative case count, with the extreme divergence between mean and median indicating prolonged low-count early periods followed by exponential surge phases — classic pandemic wave behavior. deaths mirrors this pattern (mean = 33.37, median = 1, σ = 48.03, skew = 1.39, outliers = 90), with the median of 1 confirming that the majority of the time-series predates significant mortality events. Among the policy index columns, governmentresponseindex (mean = −51.00, median = −55.47, range = −73.18 to 0, skew = 1.5, outliers = 64) and stringencyindex (mean = −47.73, median = −50, skew = 0.927) are encoded as negative deviations from a baseline — a non-standard encoding that must be accounted for before interpretation. The economicsupportindex is particularly notable: despite only 5 distinct values, it has 444 flagged outliers, suggesting a bimodal or step-function distribution concentrated at −75 and −62.5 (Q1 = Q3 bounds), pointing to highly discrete, infrequent policy changes. facialcoverings (range = −3 to 4, σ = 2.70, Q1 = −3) shows the widest spread among NPI variables, suggesting multiple policy reversals or encoding of both mandatory and voluntary regimes.
---
4. Recommended Join Key Columns
| Column | Role | Rationale | |---|---|---| | date | Primary join key | 980 distinct values, zero nulls — ideal temporal alignment key | | isoalpha3 (= "THA") | Geographic join | Standard ISO code for merging with WHO, World Bank, or OWID datasets | | isoalpha2 | Secondary geo key | Useful for APIs using 2-letter country/region codes | | keyapplemobility | Mobility join | Non-null, distinct = 1 — links to Apple Mobility Trends Reports | | keygadm | Spatial join | Non-null — links to GADM administrative boundary datasets for GIS work | | population | Normalization key | Constant 487,976 — use to compute per-capita rates before joining |
The date column is the critical join spine for any time-series correlation; all policy, mobility, and outcome datasets should be aligned on this axis before computing lagged correlations.
---
5. Recommended Paired Datasets for Correlation Discovery
Given this dataset's strengths (daily NPI indices + case/death outcomes for a Thai province), the following pairings would unlock the most analytical value:
- Apple/Google Mobility Reports — keyapplemobility is non-null and ready to join; mobility trends in retail, transit, and residential zones would directly correlate with stringencyindex and stayhomerestrictions, enabling NPI compliance analysis - Thailand MoPH Provincial Health Data — matching province-level hospitalization, ICU occupancy, and testing data would fill the 100%-null hosp, icu, and tests columns that are critically absent here - Oxford OxCGRT Full Dataset — the policy indices here appear to be OxCGRT-derived; joining the full OxCGRT dataset would resolve the negative-encoding anomaly and provide flag variables (whether measures are targeted vs. general) - Vaccination rollout data (OWID or WHO) — given vaccines, peoplevaccinated, and peoplefullyvaccinated are all null, overlaying Our World in Data's Thailand vaccination time-series would enable pre/post-vaccine outcome comparison - Climate/Weather Data (e.g., ERA5 or OpenWeatherMap) — latitude/longitude constants enable easy weather joins; temperature, humidity, and seasonal indices are established correlates of respiratory virus transmission rates in tropical settings like this province
Bottom line: This is a clean temporal backbone for a single Thai province with excellent NPI policy coverage, but it requires external data joins on at least 4–5 dimensions (testing, hospitalization, vaccination, mobility, and weather) before robust causal or correlation modeling is feasible. Treat it as a policy index time-series joined to sparse outcome data, not a comprehensive epidemiological record.
Columns
- id (string)
- date (date)
- confirmed (integer)
- deaths (integer)
- recovered (string)
- tests (string)
- vaccines (string)
- people_vaccinated (string)
- people_fully_vaccinated (string)
- hosp (string)
- icu (string)
- vent (string)
- school_closing (integer)
- workplace_closing (integer)
- cancel_events (integer)
- gatherings_restrictions (integer)
- transport_closing (integer)
- stay_home_restrictions (integer)
- internal_movement_restrictions (integer)
- international_movement_restrictions (integer)
- information_campaigns (integer)
- testing_policy (integer)
- contact_tracing (integer)
- facial_coverings (integer)
- vaccination_policy (integer)
- elderly_people_protection (integer)
- government_response_index (decimal)
- stringency_index (decimal)
- containment_health_index (decimal)
- economic_support_index (decimal)
- administrative_area_level (integer)
- administrative_area_level_1 (string)
- administrative_area_level_2 (string)
- administrative_area_level_3 (string)
- latitude (decimal)
- longitude (decimal)
- population (integer)
- iso_alpha_3 (string)
- iso_alpha_2 (string)
- iso_numeric (integer)
- iso_currency (string)
- key_local (integer)
- key_google_mobility (string)
- key_apple_mobility (string)
- key_jhu_csse (string)
- key_nuts (string)
- key_gadm (string)