Test Pilot Tracking Protection Graduation Report
Editors Note: This post originally appeared on the Firefox Test Pilot site in February of this year. We’re migrating all of our graduation reports to Medium to make them more discoverable.
UPDATE: 2017–10–12: After this experiment, we ran a follow-up Shield Study to measure breakage under various privacy protections in Firefox, including Tracking Protection. The presentation of those results are publicly available on Google Slides.
The Tracking Protection feature in Private Browsing Mode is popular with Firefox users, but it comes with a few downsides: it can break page layout and multimedia content on websites, and it sometimes blocks content that isn’t tracking you. We’ve collected qualitative evidence of this through user feedback, but since we introduced Tracking Protection, we haven’t been able to quantify the impact of breakage and false positives on the overall user experience.
We launched the Tracking Protection experiment in Test Pilot with a goal of understanding how the block list we use for Tracking Protection in Private Browsing Mode and in Firefox Focus for iOS contributes to website breakage. The experiment was fairly simple by design: we enabled Tracking Protection and let users disable it on specific sites or provide reports about websites that seemed to break as a result of the experiment.
Here’s how it went down
The user interface for the Tracking Protection experiment was simple. When Tracking Protection was enabled, users saw an icon in their browser’s URL bar. Clicking the icon opened a panel where participants could report a problem and turn Tracking Protection off for that domain.
Participants had the option to give more details about the problems they saw.
For this experiment, we only collected data when participants turned Tracking Protection on or off, or when they reported problems with a particular domain. We did not collect data on other browsing activity and we did not record full URLs.
Here’s what we learned
We hypothesized that users would frequently disable Tracking Protection on broken sites. This, it turns out, was not the case. The chart below shows that despite a high number of participants using Firefox Tracking Protection every day, Tracking Protection was disabled on very few sites.
Further, the rate at which users disabled Tracking Protection decreased as the experiment progressed. We think that once users disabled Tracking Protection for the most problematic sites, they didn’t encounter or weren’t bothered by breakage to the point that they would disable it for additional sites they visited later in the experiment. Since we didn’t measure total page views, there are limits to what we can conclude about the number of sites broken by Tracking protection. Nevertheless, this data suggests that breakage may be less widespread than we initially believed, and it’s worth digging deeper into more specific breakage data.
We were also interested in the impact Google Analytics had on site breakage. Google Analytics is a nearly ubiquitous service across the Web (full disclosure: we use it to gather traffic data for testpilot.firefox.com) that is blocked by Tracking Protection. When we began this experiment, some developers suggested that how a site set up Google Analytics could have a large impact on breakage. Blocking Google Analytics shouldn’t cause sites to break, but sometimes site developers write code that ignores the possibility of Google Analytics being blocked. If developers don’t accommodate for this possibility, broader failures in the way that code running on sites can occur because scripts on the site make reference to code that is never loaded.
It turns out that for the most part Google Analytics does not correlate to breakage. In fact, while Google Analytics shows up frequently in our data, it only correlated to breakage reports 25 percent of the time, which is low compared to breakage rates reported for other trackers.
Early on, we noticed that one service — Twimg, Twitter’s remote image hosting service — consistently showed up as a source of reported breakage. By diving into our user feedback, we figured out that this breakage caused images not to render and often lead to broken layout. We communicated our findings to Disconnect (the tracking protection company that provides our block list), and they were able to re-categorize the Twimg domain as a content provider, mitigating the breakage.
In other cases when a known tracker caused site breakages, were less successful in modifying the Tracking Protection experiment to improve user experience. For example, we saw many reports of breakage linking video playback issues to Tealium, a tracker that’s used for marketing analytics. We attempted to create a work around for this breakage by using something called a shim — a fake initial reference to the tracker code that would theoretically let sites continue to run as though it really existed without breaking. Unfortunately, we found that while we could resolve the initial error by faking references with a code shim, secondary errors would crop as sites tried to execute code. Since these errors tended to be unique across different sites, we determined our approach couldn’t account for every error and wouldn’t scale effectively.
Here’s what happens next
What’s next for the Tracking Protection experiment? Now that we’ve announced retirement plans for this add-on, we will no longer maintain it. If you have it enabled, you don’t need to do anything. We’ll automatically uninstall it in the coming days. Tracking Protection will continue as a feature of Private Browsing, however.
The Tracking Protection reporting feature, unlike many Test Pilot experiments, was never designed to graduate into Firefox. We created this experiment to learn how Tracking Protection works in the wild. We also learned a lot about (and are grateful for!) the willingness of our Test Pilot community to provide feedback.
We intend to run more studies based on our initial results. We are launching a similarly structured study in Test Pilot to learn about site performance in Firefox. We’ll use similar feedback mechanisms as the Tracking Protection study to learn about sites that cause Firefox users particular problems so we can address these issues as we work on our next generation Web rendering engine. Look for this study in late February 2017.
We will also conduct some messaging studies to help us understand how Firefox users who aren’t in Test Pilot think about online tracking and how we can best explain the costs and benefits of Tracking Protection and similar blocking services.
Finally, we’re planning a follow-up Tracking Protection Study in Test Pilot based on what we learned this go round. In the next round, we will be adding more fine tuned user control, differential privacy, and an improved user interface. We believe these additions will help us take the next step toward shipping Tracking Protection in Firefox beyond Private Browsing Mode. Look for that study in late 2017. (UPDATE 2017–10–12: The 2017 study is available here.)
Thank you to all the Test Pilots who installed Tracking Protection and reported problems through the add-on!
Want to try a new experiment? Visit https://testpilot.firefox.com.
- Luke Crouch, Privacy Engineer & John Gruen, Test Pilot Product Manager