| Term | Definition | Section |
|---|---|---|
Resonance score Shown as “Resonance” | A 0–100 score for how well a paid video earned and kept attention and got people to act, per person reached. 50 is the average of the first round of results. It is built from Engage, Interest and Act, weighted 25%, 40% and 35%. | Score |
Engage rate Shown as “Engage” | 3-second plays divided by impressions: the share of people shown the video whose scroll it stopped. | Score |
Interest rate Shown as “Interest” | Average watch time divided by the video's length: how much of the video the average viewer watched. | Score |
Act rate Shown as “Act” | Shares, saves and comments divided by 3-second plays: how often viewers did something with the video. Reactions and link clicks are not counted. | Score |
KPIs behind the Resonance score Shown as “KPIs” | The working behind Resonance, one column each for Engage, Interest and Act: the counts from the ad export, the rate they make, the reference average it is measured against, how far from it the rate sits, the component score, and the weight that component carries. | Score |
Forecast | What PAX2 expects a video to score before its results come in, given as a likely range, the way a weather forecast gives a chance of rain. When the results arrive they are set beside the forecast as the actuals, and the gap between the two is how the forecasting improves. | Score |
Counts from the record Shown as “Counts” | The platform numbers each rate is built from, as recorded in the ads export: the part over the whole. | Score |
Observed rate Shown as “Rate” | The count divided by its base, before any adjustment. | Score |
Reference average Shown as “Ref. avg” | The pooled rate across every scorable video in the first round of results. It is frozen when that round loads, so later rounds are measured against the same yardstick. | Score |
Evidence-weighted rate Shown as “Weighted” | (count + k × reference average) ÷ (base + k). A rate built on little evidence is pulled toward the reference average, so a small budget cannot produce an extreme score. k is 2,000 impressions for Engage, 300 viewers for Interest and 800 plays for Act. | Score |
Distance from the reference Shown as “Distance” | How many reference spreads the weighted rate sits above or below the reference average, measured on the log-odds scale. | Score |
Component score Shown as “Score” | The distance turned into 0–100: the share of a normal curve that lies below it. 50 is average; 84 is one spread above. | Score |
Component weight Shown as “Weight” | How much each component counts toward Resonance: Engage 25%, Interest 40%, Act 35%. A design choice, tested against alternatives on Scoring. | Score |
Score range Shown as “Range” | The span the score is likely to fall in given how much evidence stands behind it: the 10th to 90th percentile of 400 draws consistent with the counts. | Score |
Limited evidence Shown as “Thin data” | Fewer than 15,000 impressions. The score is shown but its range is wide. Under 1,000 impressions a video is not scored at all. | Score |
Impressions | How many times the video was shown in a feed. | Delivery |
People reached Shown as “Reach” | How many different people saw the video at least once. | Delivery |
Average frequency Shown as “Frequency” | Impressions divided by reach: how many times the average person saw it. | Delivery |
Amount spent Shown as “Spend” | Dollars spent on the flight. A delivery condition, shown beside the score and never inside it. | Delivery |
Cost per 1,000 impressions Shown as “CPM” | Spend divided by impressions, times 1,000. | Delivery |
3-second video plays Shown as “3s plays” | Times the video played for at least 3 seconds. | Engagement |
Average watch time Shown as “Avg. watch” | Seconds the average viewer watched. | Engagement |
ThruPlays | Plays to the end, or to 15 seconds for longer videos (Meta's definition). | Engagement |
Paid run Shown as “Paid” | The numbers from a video's paid run, as the ad platform exports them: what was spent, how many saw it, how far they watched (3 seconds, 25%, 50%, 75%, to the end), and what they did after (shares, saves, comments, reactions, clicks), with the cost of each. | Campaigns |
Hits and Misses | Every short PAX2 tracks, as a dot: publish date across, and views over the channel's usual (its median) up, on a log scale. Green dots broke out well above their channel's usual, red ones fell well below it, grey landed near it. Each dot opens its video. | Campaigns |
Feature lift formula Shown as “Formula” | How PAX2 measures whether a trait helps videos break out. Within each channel, the share of videos with the trait that broke out is compared with the share of those without it; each channel counts in proportion to how many videos it has on both sides; the channels are pooled into one difference in points; and a 90% range comes from redrawing the channels a thousand times. | Campaigns |
Swayable persuasion tests Shown as “Swayable” | Swayable showed 617 political ads from 2018 and 2020 to one randomly chosen group of people and not to another, then asked both groups how they would vote. The difference is how much the ad changed minds. Each ad was also coded on a checklist of traits, so the effect of a trait is the shift for ads with it against ads without it (Hewitt, Broockman, Coppock and others, APSR 2024). | Campaigns |
Wesleyan Media Project Shown as “Wesleyan” | Wesleyan tracks and codes nearly every political ad that runs: on TV since 2006, and on Meta and Google since 2018. For 2022 it labels each Meta ad's goal (persuade voters, raise money, or both), its tone and the party it favors, beside what Meta reports it spent and how often it was shown. It shows what every other campaign ran, not whether any of it worked. | Campaigns |
Upworthy Research Archive Shown as “Upworthy” | From 2013 to 2015 Upworthy ran 32,487 tests showing the same story under different headlines to readers at random, and recorded which headline drew more clicks. It measures packaging: what a title does to whether people open the story. Released by Matias and Munger, 2021. | Campaigns |
Internal wins above replacement Shown as “iWAR” | A video's views divided by its own channel's median views. It takes channel size out, so a small channel's hit counts as much as a big one's: 3× or more is a hit, half or less a miss. A channel growing fast understates its new videos, because its median lags behind it. | Campaigns |
Wins above replacement Shown as “WAR” | A video's views divided by the median views of every short across all 65 followed channels together. It measures raw reach in the whole field, so big channels lead it simply by being big. | Campaigns |
Views per day since posting Shown as “Views/day” | Public YouTube views divided by the days since the video was posted. Organic, not paid, and not comparable to Resonance. | Campaigns |
Median video length Shown as “Median length” | The middle length across a channel's recent shorts. | Campaigns |