- Active hours: Total time the user spent interacting with the browser - Session length: Total browser session time - Number of subsessions: Total Telemetry reports - Total tab open events: Number of tabs opened - Total URI count: Total number of pages loaded - Unique URI count: Number of unique websites (TLD+1) visited - Address bar searches: Number of searches initiated from the address bar - Search bar searches: Number of searches initiated from the dedicated search bar - Total Yahoo searches: Number of searches made using the Yahoo search engine - Total Google searches: Number of searches made using the Google search engine - Total searches: Total number of searches overall initiated from one of the Firefox search bars - Number of bookmarks: Total bookmarks added - Length of history: Total number of pages visited
So everyone with an adblocker had normal activity levels at the start (compared to the non-adblock-user they are 'matched' with by the propensity scoring), but after installation, then the adblocker people saw large increases in activity that their non-adblock counterparts, who were web browsing just as much before hand, did not experience.
This largely rules out the reverse causation of 'browsing -> adblock installation'. It could still be driven by people forecasting that they will be doing a lot of web browsing in the future and installing adblock before they begin doing all the extra browsing, but this feels a lot less plausible to me.
I'm struck by the almost numerical identity of their adblock correlate on page views with the long-term causaleffect on total site traffic I estimated in my banner ad A/B test ( https://www.gwern.net/Ads ) of (15% vs my 14%, 13-16% CI). The unit of analysis isn't the same but they're arguably equivalent.
This is also interestingly parallel to Pandora's giant A/B test: Huang et al 2018's https://davidreiley.com/papers/PandoraListenerDemandCurve.pd... "Measuring Consumer Sensitivity to Audio Advertising: A Field Experiment on Pandora Internet Radio".