Trends and analytics
What the Trends views measure, and what they cannot tell you about the wider literature.
The Trends section aggregates over the corpus rather than over the literature. That distinction governs everything on this page: these views describe what has been collected here, and only approximate the field.
The views
- Trends
- How volume and composition move over time: what is being published, in which themes, and how the balance of directions shifts.
- Methods and evidence
- The distribution of study designs and evidence grades across the corpus. This is where the live figures live, including the share of the corpus using causal or quasi-experimental identification.
- Coverage
- Which themes and outcome categories are well populated and which are thin. Useful mainly for spotting where an apparent consensus is really just a small number of papers.
- Impact
- Citation-weighted views of which work is being picked up.
- Authors and Network
- Who is publishing in this area and who they publish with, as recorded by the corpus. See the caveat in the site guide.
How to read a trend here
A rising line can mean the field is growing, or that our sources indexed more of it, or that the relevance gate happened to admit more of it. These are not distinguishable from inside the chart.
Specific things that will mislead you if you forget them:
- Recent periods are undercounted. Publication and indexing both lag, so the most recent bar is always incomplete and always looks like a decline.
- Composition changes are partly intake changes. Adding a source, or changing what the relevance gate admits, moves these charts without anything changing in the world.
- Direction counts are not effect sizes. Ten weak positive claims and one strong null are not ten-to-one evidence. Weight by evidence grade, which is why Explore sorts on it.
- Citation counts favour older work. Anything recent has had less time to accumulate citations, so impact views are biased towards the past.
What the corpus is
The corpus holds papers that cleared the relevance gate. Papers scoring below the threshold are removed, with an auditable record of the exclusion kept. So the denominator in every one of these charts is "papers judged relevant to AI and economic outcomes", not "papers published". See the intake pipeline.