Abstract
CrowdTangle closed in August 2024, but many archived exports of Facebook posts remain in use, and each row carries an "Overperforming Score" that compares a post's interactions with an expected value. I examine what this score measures in 856,699 posts from 186 party and leader pages in 20 European countries, 2010 to 2020, exported in October 2020. The score is present for every post and follows the documented sign convention. But the expected value it uses does not reflect the time of posting. Inverting the documented formula, 95% of posts have an implied benchmark within 2% of a single value for their page and post type, and within each page and type the score ranks posts almost exactly as raw interactions do (median rank correlation above 0.999). A constant built from each page's last 100 posts before the export fits the score closely (correlation 0.97 on the log scale), whereas benchmarks built from the posts before each post do not (0.52 to 0.56). As a result, older posts look like heavy underperformers: the median score is −25 for 2010 and −1.1 for 2020. The score removes differences between pages, so populist and leader advantages in raw interactions disappear, while comparisons within pages simply reproduce raw counts. In archived exports the score is best read as raw interactions rescaled by a page and type constant set around the time of export, so researchers should work with raw interaction counts, compared within page and period, or with baselines they build and report themselves.
Questions this answers
Cite
Hotham, T. (2026). Count the Interactions, Not the Score: What CrowdTangle's Overperforming Score Measures in Archived Facebook Data. Preprint, Zenodo. https://doi.org/10.5281/zenodo.23418605