Cross-corpus signals: which topics the press is leading or following, where coverage and discussion diverge, and which stories are under-reported (or over-reported) relative to grassroots commentary.
Each dot is one balanced topic, plotted by how its selected period’s coverage compares against the 12 weeks that preceded it. 0 means a typical week; +1 means roughly double the typical volume; −1 means silent. Press volume on the horizontal axis, discourse on the vertical — the four quadrants tell you whether the two corpora are tracking each other or pulling apart.
For each balanced topic we line up the two weekly-volume series and slide one against the other from −4 to +4 weeks, picking the offset where the two move most tightly together (or most exactly apart). The strongest twelve are shown.
| Topic | Direction | Lag | r |
|---|---|---|---|
| Election Satire discourse view → | press leads | 4w | +1.00 |
| Public Spending Priorities discourse view → | discourse leads | 4w | +1.00 |
| Public Engagement discourse view → | press leads | 4w | -1.00 |
Why we filter to balanced topics. A topic that shows up heavily in news but not at all in discourse (or vice versa) tells us nothing about how the two are related — the comparison is undefined when one side is silent. To make the cross-corpus signals meaningful, this page restricts to topics with at least 5 items in EACH corpus over the trailing 12 calendar weeks. That puts every dot on the quadrant in the same analytical frame: "both sides have enough data to take a reading".
The trade-off is that newer topics, niche topics, and corpus-asymmetric topics drop out. If a topic only ever appears in press releases or only in editorial blogs, you won't see it here — you'll see it on /press or /discourse individually. The per-topic page links from each row let you drill in with no balance filter applied.
Z-scores, not raw counts. Each axis is the corpus's deviation from its own 12-week mean. A topic with consistent low coverage and a topic with consistent high coverage end up with similar z-values when both are running at their typical level — the quadrant rewards change, not volume.
Lead-lag is per-topic. The cross-correlation runs on that topic's own weekly series in both corpora. We pick the lag with maximum |r| and report it. With only 12 weekly samples, |r| values below 0.25 are likely noise; we hide them.
Spotted something wrong on this page? Report a correction.
