G²LM|LIC - Browsers Don't Lie

The researchers observe every website visit, including the URL (uniform resource locator, i.e., web address) and timestamp. Although browser data can include records from multiple types of electronic devices, most smartphone browser apps do not support internet browser extensions or add-ons, so the PY Insights technology only collects data from personal computers. No information is collected from private browsing or Incognito mode, and personal identifiers are removed prior to analysis.

Each URL has an associated title, which conveys meaningful information, such as a Google search phrase, the headline of a newspaper article, or a YouTube video title. Using the URL, title, and timestamp for each website visit, PY Insights calculates its duration in seconds and provides a detailed categorization scheme for each website domain. The categories are based on Google Cloud Platform’s natural language processing algorithm.

The project used these categories to identify websites as being primarily related to leisure (entertainment) or production (non-recreational). Leisure includes Adults, Arts & Entertainment, Games, Online Communities (including social media), and Shopping. Production includes Business & Industrial, Computers & Electronics, Finance, Internet & Telecom (including e-mail and search engines), Jobs & Education, Law & Government, News, Science, and Reference.

Other Google Cloud categories combined cover 0.8% of our data. Some websites – such as spam webpages – are also labelled as “other”. Median “other” category usage on a day covers 7% of total time use.

Because YouTube represents a sizeable portion of usage and is classified as leisure by PY Insights, also robustness checks were conducted in which YouTube videos as leisure or production related using Google’s YouTube API were re-classified. The researchers collected data on individuals aged 22 to 54 located in 28 states across India, as well as Kenya and Nigeria. Individuals were prevented to use a new browser account or a secondary browser type that is not used regularly from participating by requiring at least 30 days of browser data.

One user was dropped who preferred not to state their gender and took two steps to avoid computer bots: by including an attention test question in the survey and all users with an average of more than 3,000 URL visits per day were manually dropped.