Policy agenda basics
Everything you need before starting to work with Policy Agendas data: what the project measures, the concepts it uses and related terminology, and the codes. No prior knowledge assumed.
The basics
Five questions, answered.
A research network that measures what governments pay attention to, by applying one topic-coding scheme to the records of political activity — legislative hearings, statutes, budgets, party platforms, court decisions, newspaper coverage, public opinion. It began with the US Policy Agendas Project, founded by Frank Baumgartner and Bryan Jones in 1993, and now spans twenty established country and sub-national projects with more in development.
The set of issues actually receiving attention at a given moment, in a given place — a chamber, a court, a front page. It is a measurement, not a statement of intent. Nobody publishes an agenda; it is recovered by coding what was actually said, heard, passed or printed.
Every document gets exactly one major topic and one finer subtopic from a fixed codebook. One document, one code. That constraint is the whole point: because a congressional hearing, a budget line and a newspaper story can all carry the same code, they can be counted on the same axis and compared across decades and countries.
Because attention is scarce and government cannot attend to everything at once. Measuring where attention goes — and how abruptly it moves — is how the project studies why some problems get addressed and others wait. Its central finding is that policy attention does not drift smoothly: it holds still, then lurches.
Yes. The established series are free and public at comparativeagendas.net, with codebooks alongside them. This site adds live, continuously coded series to that ecosystem rather than replacing anything in it.
The vocabulary
9 terms that mean something more specific here than in ordinary usage, each with a real coded example underneath it. Also on the students’ page.
Not a plan or an intention — the set of issues actually receiving attention in a given venue at a given time. An agenda is measured, not declared.
The two levels of the coding scheme. Every document gets exactly one major topic (Health, Defense, Education…) and one finer subtopic beneath it. One document, one code: that constraint is what makes venues comparable.
In the data“Maine patients should not have to wait 12 years for an ME/CFS diagnosis” → major 3 Health, subtopic 398 Health – R&D. “Automated trash collection is coming to York” → major 7 Environment, subtopic 703 Waste Disposal. Both are Portland Press Herald pieces; neither could carry a second code.
The project's central empirical finding: policy attention does not drift smoothly. Most measures barely move for long stretches, then lurch. Plotted as annual change, this gives a sharply peaked, fat-tailed distribution rather than a bell curve.
Coverage of the political contest — polls, endorsements, fundraising, who is ahead — as against coverage of what a candidate would actually do in office. The same running story usually produces both, and the coding follows the article rather than the issue: a piece about the fight over a policy is coded to the campaign, a piece about the policy itself is coded to the policy. Campaign coverage sits at subtopic 2012 under major topic 20, Government Operations — and 20 is not a ‘process’ bucket. It is a substantive topic covering the machinery of government (the postal service, appointments, public property, scandals), and Political Campaigns is only one subtopic beneath it.
In the dataTwo Maine stories, three days apart. “Bobby Charles’ ICE-at-the-Polls Remarks Spark Political Firestorm” (The Maine Wire, 29 August 2026) is the contest — who said what, and who is furious about it. “ICE has arrested more people in Maine this year than in previous 2 years combined, data shows” (Portland Press Herald, 26 August) is the policy — what is actually happening, whoever wins. And the distinction is not a topic: on 19 August “Susan Collins takes cautious line on data centers opposed by most Mainers” and “North Berwick to ask voters about data center ban, comprehensive plan change” carry the same code, 14/1403, from opposite sides of it. The reverse holds inside topic 20 itself: on 3 September “Susan Collins launches 1st attack ads in reelection campaign for Senate” (Portland Press Herald) is 20/2012, the campaign, while “Bangor disrupts struggling healthcare provider’s plan to sell $1.3M building” (Bangor Daily News) is 20/2008, Property Management — the same major topic, and only one of them is about who wins.
Read themBobby Charles’ ICE-at-the-Polls Remarks Spark Political Firestorm The Maine Wire · ICE has arrested more people in Maine this year than in previous 2 years combined, data shows Portland Press Herald · Susan Collins takes cautious line on data centers opposed by most Mainers Bangor Daily News · North Berwick to ask voters about data center ban, comprehensive plan change Portland Press Herald · Susan Collins launches 1st attack ads in reelection campaign for Senate Portland Press Herald · Bangor disrupts struggling healthcare provider’s plan to sell $1.3M building Bangor Daily News
Every step between a signal and a decision costs something — attention, agenda space, a majority, a veto not used — and those costs accumulate along the process. Friction does not just slow change, it reshapes it: where friction is low, attention tracks the signal smoothly; where it is high, pressure builds until it overcomes the cost and the response arrives in a lurch. It is why the data domains are ordered from press coverage to budgets rather than alphabetically, and why the punctuation gets sharper at each step.
In the dataThe Maine press moved from 1.4% to 10.8% Immigration coverage in a single month. No committee calendar, statute book or budget line can move that fast, and none of them did. The signal is the same at every step; what differs is what it costs to register it. That gap between the press series and the output series is the friction, and it is measurable.
Where attention is being measured: a congressional hearing series, a statute, a budget line, a newspaper front page, a poll question. The point of one codebook is that these become comparable.
In the dataCode 14 Housing indexes 367 Maine press articles in those 90 days. The same code 14 indexes hearings in the Congressional Hearings series and stories in the New York Times Front Page series. The number means something different in each — a hearing is not a headline — but because the code is the same, the series can be put on one axis.
A single number for whether two measured series rise and fall together. This site reports Pearson's r, which runs from +1 (they move in perfect step) through 0 (no linear relationship at all) to −1 (one rises exactly as the other falls). It says nothing about cause: two agendas can track each other because one drives the other, because both answer to a third thing, or by coincidence. And because it summarises seventy years in one figure, it hides everything that happened inside them.
In the dataOn the Trends tool, Defense as a share of the front page and as a share of the State of the Union correlate at +0.32 across the 71 years both cover (1946–2016). Square that and you get about 0.10 — roughly a tenth of the year-to-year movement in one series is shared with the other, and nine tenths is not. Across the whole agenda the closest pair of domains is congressional hearings and public laws at +0.73; the weakest link is presidential speeches and the Supreme Court at −0.03, close to no relationship across the years both cover.
The assumption — usually wrong — that a relationship carries along a chain: if A tracks B and B tracks C, then A must track C. Correlation is not transitive, and a grid of correlations quietly invites you to believe it is. Whenever two ends of a chain come apart, something is happening at the step between them.
In the dataPresidential speeches track executive orders at 0.36, and executive orders track public laws at 0.56 — but speeches and laws score just +0.09 with each other. Nine such chains show up among these eight domains, and eight of the nine run across the friction sequence rather than inside one stage. That is progressive friction leaving a fingerprint: each institution an issue passes through charges its own toll, so the two ends of a long chain stop moving together.
A single long-run coded dataset for one venue — for example the New York Times Front Page series, or Congressional Hearings. Each has its own codebook notes, time span and citation.
In the dataThis site's own Maine news collection is one series: five outlets — Portland Press Herald, Bangor Daily News, Lewiston Sun Journal, The Maine Wire, The Maine Monitor — coded with CAP v4. Its span and outlet list are part of the series definition; change either and the numbers stop being comparable with what came before.
The codes
The major topics themselves. Every figure and article list on this site carries these numbers, because the number is what joins a series here to every other Comparative Agendas dataset — label wording drifts between codebook versions, the code does not. Subtopics sit beneath each one; the full scheme is on data & methods.
Where to go next
Three routes on from here, depending on what you came for.
For students and scholars — the same vocabulary with worked examples, plus what to read first: the two books the field is built on, and short explainers of each concept. Teaching carries syllabi from courses that use this material.
Training videos — the project's own ten-module course, from what it measures to coding a document yourself, if you would rather be walked through the ideas than read them.
For researchers — access, license and citation, how the figures are computed, and how to reproduce one. Data domains lists every series the project maintains; search queries across them.