Measuring third party tracker power across web and mobile

Reuben Binns, Jun Zhao, Max Van Kleek, Nigel ShadboltACM Trans. Internet Techn 2018 · DOI · arXiv

TL;DR

This paper introduces a novel metric, PROWISH-HHI, to measure the market power of third-party trackers by considering their prominence and corporate structure, not just their prevalence. The study reveals that the web and mobile tracking ecosystems are partially overlapping but distinct, and that traditional market share measures are insufficient for understanding tracker power, especially when considering mergers and acquisitions.

What Problem Does This Paper Address?

In today's digital economy, third-party trackers collect vast amounts of user data across websites and mobile applications. This data collection has significant implications for user privacy and competition among companies. Regulators are increasingly scrutinizing the consolidation of tracking capabilities within a few firms, but traditional measures of market power, like revenue or unit sales, fail to capture the true "tracking capability" of these entities. This is because a tracker's power stems from its ability to gather, analyze, and monetize data from a wide array of sources, which may not directly correlate with its financial performance. Furthermore, the distinct nature of web and mobile ecosystems makes it challenging to define a unified market for tracking and to assess how firms operate across these platforms. This paper aims to bridge this gap by proposing a new approach to measure tracking concentration that accounts for these complexities.

Methods

The authors propose a novel metric called PROWISH-HHI (Prominence Weighted Integration Share - Herfindahl-Hirschman Index). This metric builds upon existing methods by:

  • Prominence Weighting: Instead of just counting the number of websites or apps a tracker is integrated into (prevalence), PROWISH weights this by the popularity of those sites/apps. This better reflects the actual number of users a tracker can potentially reach.
  • Parent-Subsidiary Relationships: The analysis incorporates corporate structures, aggregating tracking capabilities of subsidiaries under their parent companies. This provides a more accurate picture of the market power of larger entities.
  • Herfindahl-Hirschman Index (HHI): This standard economic measure of market concentration is adapted to use the PROWISH metric as the basis for market share.

The paper also details a comprehensive data collection methodology involving:

  • Web Crawling: Utilizing the OpenWPM project and the Webcensus dataset to analyze the top 5,000 websites.
  • Mobile App Analysis: Employing a combination of static (APK analysis) and dynamic (network traffic monitoring) methods on the top 5,000 Android applications from the Google Play Store.
  • Cross-Platform Analysis: Comparing trackers found on equivalent web and mobile services to understand ecosystem overlap.

Datasets

  • Web Data: The study used a subset of the Webcensus dataset (part of the Web Transparency project), consisting of a stateful crawl of 100,000 websites. This was further filtered to analyze the top 5,000 websites based on Alexa rankings.
  • Mobile Data: The study analyzed the top 5,000 Android applications from the Google Play Store.
  • Equivalent Services: A subset of 358 genuinely equivalent web-app pairs were manually validated to assess platform overlap more precisely.
  • Knowledge Base: The authors constructed a knowledge base of tracker firms, including their jurisdiction, founding date, and parent-subsidiary structures, compiled from public sources like Crunchbase, OpenCorporates, and Wikipedia.

Key Insights and Ideas

  • Prominence Matters More Than Just Presence: A tracker's true power isn't just how many sites it's on, but how popular those sites are. A tracker on a few highly popular sites can reach more users than one on many less popular ones.
  • Corporate Structure is Crucial: Aggregating trackers under their parent companies significantly alters market share rankings, revealing that large conglomerates wield more power than their individual subsidiaries might suggest.
  • Web and Mobile Aren't Separate Worlds: While distinct, the web and mobile tracking ecosystems have a partial overlap. This suggests that regulators might need to consider them as a single, albeit complex, market.
  • Traditional Metrics Fall Short: Standard market share calculations based on revenue or prevalence are inadequate for assessing tracker power, especially when considering the impact of mergers and acquisitions.
  • Mergers Can Dramatically Increase Concentration: The study models the impact of past acquisitions, showing that some mergers (like Google and Doubleclick) significantly increase market concentration, potentially warranting regulatory attention.

Key Results

  • Web Tracking Concentration: Using the PROWISH-HHI metric, the web tracking market shows a concentration of 0.12 at the parent level, which is six times higher than the 0.02 HHI calculated using simple integration share. This level is significant enough to merit attention under EU competition measures.
  • Mobile Tracking Concentration: The mobile tracking ecosystem shows a PROWISH-HHI of 0.12 at the parent level, similar to the web.
  • Ecosystem Overlap: Analysis of equivalent web-app pairs revealed that, on average, only 15-17% of trackers overlap between a service's web and mobile versions, suggesting platform-specific tracking strategies.
  • Impact of Acquisitions: The merger of Google and Doubleclick, for instance, accounts for a PROWISH-HHI difference of 0.039, which would meet the proposed threshold for intervention under EU law.

Limitations and Future Work

  • Web Crawling Limitations: The order of website visits might influence detected third-party content, and only the first page of sites was crawled. App analysis was limited to 30-second sessions.
  • Assumptions on Relationships: The study did not account for potential binding agreements between parent firms not to merge tracking capabilities or for cross-network data sharing agreements.
  • Data Sensitivity Not Considered: The prominence and prevalence metrics do not account for the sensitivity of the data collected by trackers.
  • Popularity as Proxy: Prominence relies on website rank as a proxy for audience size, which may not perfectly reflect actual user numbers.
  • Future Work: The authors suggest incorporating cross-network data sharing detection, investigating the impact of counter-transparency efforts (like proxy services and obfuscation), and exploring the influence of regulatory environments on tracking behavior. They also propose examining geographical differences and the trade-offs faced by developers.

Summary updated

BibTeX
@article{binns2018measuring,
  author = {Reuben Binns and Jun Zhao and Max Van Kleek and Nigel Shadbolt},
  title = {{Measuring third party tracker power across web and mobile}},
  journal = {ACM Trans. Internet Techn.},
  year = {2018},
  doi = {10.1145/3176246},
}

Loading related content...

Figure