Our World in Data – COVID-19 Testing Time Series
- Rows
- 997
- Columns
- 14
- Data range
- 2020-02-25 to 2022-06-23
Daily COVID-19 testing data by country including total tests, daily new tests, tests per thousand, positive rate, and testing units (PCR, antigen, etc.). Collected from official national sources. Coverage from early 2020 through June 2022 when OWID stopped updating the testing dataset.
AI analysis
COVID-19 Testing Time Series – Dataset Analysis
1. Dataset Overview and Research Value
This dataset, sourced from Our World in Data's GitHub repository (raw.githubusercontent.com/owid/covid-19-data), represents a curated time series of COVID-19 testing activity drawn from official national health authorities. It spans early 2020 through June 2022, capturing the full arc of the pandemic's first two years across what appears to be a tightly scoped slice of 4 countries (distinct ISO codes = 4, distinct entities = 4). This narrow country coverage is a critical finding — despite 997 rows and 850 distinct dates, the dataset is not a global panel but rather a deep longitudinal record for a small country cohort, making it ideal for within-country trend analysis and time-series correlation studies rather than broad cross-national comparisons. Its greatest analytical value lies in pairing testing intensity metrics against case outcomes, hospitalizations, or vaccination rollout timelines.
---
2. Data Quality Observations
Overall data quality is moderate but manageable. Total null cells across 14 columns reach 2,719, representing roughly 19.5% of all possible values — a non-trivial missingness burden. The most severe issue is the Notes column, which is 100% null (997/997 rows), rendering it entirely useless and a candidate for immediate column removal. Source URL and Source label each carry 270 nulls (27.1%), which undermines source traceability for more than a quarter of records — a concern for audit-grade analysis. The two daily-change columns (Daily change in cumulative total and its per-thousand counterpart) share 277 nulls (27.8%) and are also typed as String rather than numeric, which is a type mismatch that will block any direct arithmetic or regression without a parsing step. The Short-term positive rate and Short-term tests per case columns have a comparatively minor 28 nulls (2.8%) each. No duplicate row count was available from Phase D recomputation, so deduplication should be verified before analysis.
---
3. Key Column Distributions
The most analytically rich columns are Short-term positive rate and Short-term tests per case, and both show notable distributional characteristics. The positive rate (mean = 0.184, median = 0.169, σ = 0.140) is moderately right-skewed (skew = 1.35) with 45 outliers, suggesting episodic surges where testing capacity was overwhelmed — values approaching the maximum of 0.967 indicate near-total positivity at certain time points, likely corresponding to Omicron-era under-testing. The tests per case column is dramatically right-skewed (skew = 5.78) with a mean of 35.3 versus a median of just 5.9 and 141 outliers — the extreme max of 1,067.6 reflects periods of very low case prevalence relative to testing volume, possibly early suppression phases. The 7-day smoothed daily change (mean = 1,783, median = 1,301, skew = 0.889, 7 outliers) offers the cleanest signal for trend analysis given its smoothing design and low null count of only 16 rows. Cumulative total per thousand (mean = 195.1, median = 145.1, Q1 = 18.9, Q3 = 346.4) shows wide dispersion across the time range, reflecting ramp-up from near-zero early in the pandemic.
---
4. Recommended Join Key Columns
The natural join keys are ISO code (standardized 3-letter country identifiers) and Date (zero nulls, 850 distinct values, daily granularity). Together they form a compound key suitable for joining against any dataset using ISO 3166-1 alpha-3 codes and calendar dates. Entity (country name string) can serve as a secondary human-readable join key but is less reliable for programmatic joins due to potential naming inconsistencies across sources. Note that with only 4 distinct countries, any cross-dataset join must ensure the partner dataset also contains those same 4 entities or the merge will be severely limited in scope.
---
5. Recommended Pairing Datasets for Correlation Analysis
Several dataset categories would pair powerfully with this data:
- COVID-19 Case & Death Time Series (also from OWID or Johns Hopkins) — the most direct pairing, enabling correlation between testing rates and confirmed case/fatality trajectories, and validating whether positive rate spikes precede reported death surges with a predictable lag - COVID-19 Vaccination Rollout Data — joining on ISO code + date would allow analysis of whether vaccination coverage correlates with declining positive rates or reduced testing demand - Hospital & ICU Capacity Data — linking testing intensity to healthcare system strain could reveal whether high tests-per-case periods coincide with system relief - Government Response Stringency Index (Oxford COVID-19 Government Response Tracker) — correlating lockdown/restriction levels with testing volume changes could illuminate policy effectiveness - Mobility Data (Google or Apple COVID Mobility Reports) — pairing with the 7-day smoothed testing change could reveal whether population movement predicts testing demand with a short lead time
Columns
- Date (date)
- ISO code (string)
- Entity (string)
- Source URL (string)
- Source label (string)
- Notes (string)
- Cumulative total (integer)
- Daily change in cumulative total (string)
- Cumulative total per thousand (decimal)
- Daily change in cumulative total per thousand (string)
- 7-day smoothed daily change (integer)
- 7-day smoothed daily change per thousand (decimal)
- Short-term positive rate (decimal)
- Short-term tests per case (decimal)