Voidly
All findingsPublished research

Lead/lag cross-correlation: which countries’ censorship precedes others’

Where /atlas/correlation-matrix shows simultaneous co-movement and /atlas/cohorts shows shape similarity, this surface adds the time-shifted axis: for every pair of the 50 most-censored countries, the daily confirmed-censorship rate is cross-correlated at lags of -30 to +30 days, and the lag with…

Read the full published summary

Where /atlas/correlation-matrix shows simultaneous co-movement and /atlas/cohorts shows shape similarity, this surface adds the time-shifted axis: for every pair of the 50 most-censored countries, the daily confirmed-censorship rate is cross-correlated at lags of -30 to +30 days, and the lag with maximum Pearson r is recorded. 129 pairs clear the bar of |r| >= 0.4 with a Benjamini-Hochberg FDR-corrected p < 0.05 (1,225 raw pairs evaluated; 30 of 50 countries kept after a sparse-country filter that drops degenerate r=1.0 artifacts). The strongest single signal is Saudi Arabia leading Thailand by 1 day (r = +0.91); the densest sub-graph is a Middle East -> Central Asia chain (UZ -> PK -> AZ at +-1-2 days, IQ -> KZ at +26 days, MA -> IR at +15 days). Lag-zero is intentionally excluded (it is covered by the correlation matrix). Honest caveats baked into every response: cross-correlation is descriptive, not causal — a lead/lag pair can reflect a shared regional driver rather than imitation; the FDR correction controls false discoveries but 7-day-centered smoothing inflates short-lag correlations; IODA disruption rows are excluded from the source signal so the series is confirmed-censorship only.

What we found

The strongest lead/lag pairs in the past 365 days are concentrated in two regions: the Gulf states (SA, AE, QA, BH cross-correlate at ±1 day, suggesting near-simultaneous regional patterns just under the lag-zero floor) and a Middle East → Central Asia chain (AE → SA → TH → KZ pattern at +1 day, IQ → KZ at +26 days, MA → IR at +15 days). The Asian cluster (UZ → PK → AZ at ±1–2 days) is the densest sub-graph: Uzbekistan’s daily censorship pattern leads Pakistan by 1 day with r=+0.89, and Pakistan in turn leads Tanzania by up to 30 days with r=+0.72.

What this means — and what it does not

Cross-correlation is descriptive, not causal. Three plausible explanations for any significant lead/lag pair:

  • Direct influence: country A’s state apparatus or ISPs literally follow country B’s playbook (e.g., a shared censorship vendor pushing the same block list).
  • Shared driver: both countries respond to a common upstream event (regional election cycle, sanctions package, transit-provider outage, common upstream CDN block).
  • Measurement artifact: overlapping ISP infrastructure, the same OONI probe burst sampling both countries, or a single news event triggering coordinated test runs across multiple probes.

Distinguishing these three is what the Atlas case studies pipeline is for — each significant pair here is a candidate for a deeper write-up that pulls in incident metadata, ASN overlap, and the political event calendar.

How to read the table

A row UZ → PK lag=1d r=+0.89 reads: when Uzbekistan’s daily citable-censorship count spikes, Pakistan’s count spikes one day later, and the Pearson correlation between the two smoothed series (UZ today vs. PK tomorrow) is 0.89 over the overlap window. The FDR-adjusted p-value tells you how unlikely that pattern is under the null of two independent random walks given that we tested 1,225 pairs × 60 lags = 73,500 hypotheses.

Where this fits in the Atlas

/atlas/correlation-matrix answers “which countries co-move today?” /atlas/cohorts answers “which countries have similarly shaped 365-day patterns under time warping?” This page answers the third question: “which country’s pattern reliably precedes another’s by N days?” — the predictive-signal axis that would let an analyst, on seeing a spike in country A, raise the alert level for the countries that historically follow it.

The source record.

Original source links and labels. Linked responses may have changed since publication.