Build · PAX2

Name	Publisher	Type	Description	Status
Campaign ad persuasion experiments	Hewitt, Broockman, Coppock et al. · APSR 2024 · Swayable	Experiments	617 real campaign ads, each tested against a control group: measured persuasion effects, hand-coded features for every ad, and 607,450 de-identified responses.	Loaded
PAX2 reference model	PAX2 · built from the persuasion experiments	Models	Per-feature persuasion differences from a random-effects meta-regression, the spread of true effects, and a leave-one-study-out test of how predictable an ad is from its features.	Loaded
Small effects of political advertising	Coppock, Hill & Vavreck · Science Advances 2020	Experiments	59 real-time randomized experiments on 49 presidential ads with 34,000 participants: persuasion effects by context, message, sender and receiver.	Loaded
Do emotional ads persuade?	Replication data · Harvard Dataverse, 2026	Experiments	Real-time campaign advertising experiments testing emotional appeals against measured persuasion.	Loaded
Upworthy headline experiments	Matias et al. · Scientific Data 2021	Experiments	32,487 randomized headline and image tests with click outcomes, 2013–2015: which opening line earns the click.	Loaded
Reform campaigns in Google's political ad library	Google Political Ads Transparency · via GovBlock	Ad libraries	Every open-primary and ranked-choice advertiser's Google ads: creatives, dates, impression bands, spend ranges, targeting, weekly spend and spend by state.	Loaded
Reform campaign shorts	YouTube Data API · Oct 3, 2026	Video	About 30 recent shorts each from Vote YES on 3, Courageous Colorado, Open Primaries and Unite America, with public views, likes, comments and length.	Loaded
Massachusetts House ballots, 2022	Klarner State Legislative Election Returns · via GovBlock	Elections	How many candidates were on the 2022 general-election ballot in each of the 160 Massachusetts House districts.	Loaded
PAX2 takes and video analysis	PAX2	Video	Every take's video, word-level transcript, timestamped video review with captions and text on screen, and measured loudness, pitch and pauses.	Loaded
Polis conversations	The Computational Democracy Project	Opinion	Real Polis conversations with full vote matrices: how opinion groups form and which statements bridge them.	Loaded
Political ad content codebook	Wesleyan Media Project · 2020	Reference	The standard scheme for coding political ads: sponsor, tone, issues, appeals, visuals, music.	Loaded
Persuasion archive reproduction code	active-maintenance-aec on GitHub	Code	R code that downloads, checksums and rebuilds the persuasion archive's analysis.	Loaded
The partner's Ads Manager export	Meta Ads Manager	Paid results	The partner's real flights: spend, impressions, plays, watch time, actions	Stand-in
The partner's videos	The partner	Video	The partner's own takes, in place of the synthetic eleven	Stand-in
Meta Ad Library	Meta	Ad libraries	Every Question 3 ad on Facebook and Instagram	Not connected
Meta Marketing API	Meta	Paid results	Results pulled directly, without a CSV	Not connected
TikTok Research API	TikTok	Video	Reform campaigns' TikTok videos and counts	Not connected

Run	Videos	Estimated Cost	Started	Finished
Analysis 1	11	$0.51	Oct 3, 3:54 PM	Oct 3, 4:21 PM

Real dataStand-in dataReal and stand-in

Remaining

$2.85

before the cap

Per Video

$0.05

about $0.04 a video from the next run on

Biggest Item

Pegasus

the video notes, from the next run on

Workspace

demo-2

the review this browser reads

AWS budget

estimated spend against the cap

7.15 / 10 USD

PAX2 reading

seconds per reading against Amplify's 30-second cut-off

16 / 30 seconds

Worker box

stops itself after this many idle minutes

0 / 30 minutes