Construction of the Tranco list
We designed the standard configuration of the Tranco list to improve agreement on the popularity of domains and stability over time, using the rankings from the four studied providers as our source data. To achieve this, we average ranks over the lists of all four providers over the past 30 days. We apply the Dowdall rule (scoring items with 1, 1/2, ..., 1/(N-1), 1/N points) to calculate the new score on which domains are ranked.
Properties of source lists
Alexa, a subsidiary of Amazon, publishes a daily updated list consisting of one million websites. Usually only pay-level domains are ranked, except for subdomains of certain sites that provide "personal home pages or blogs" (e.g. tmall.com, wordpress.com).
The ranks calculated by Alexa are based on traffic data from a "global data panel", with domains being ranked on a proprietary measure of unique visitors and page views, where one visitor can have at most one page view count towards the page views of a URL. Alexa states that it applies "data normalization" to account for biases in their user panel. The panel is claimed to consist of millions of users, who have installed one of "many different" browser extensions that include Alexa's measurement code. However, through a crawl of all available extensions for Google Chrome and Firefox, we found only Alexa's own extension ('Alexa Traffic Rank') to report traffic data. Moreover, this extension is only available for the desktop version of these two browsers. Chrome's extension is reported to have around 570,000 users; no user statistics are known for Firefox, but extrapolation based on browser usage suggests at most one million users for two extensions, far less than Alexa's claim.
In addition, sites can install an 'Alexa Certify' tracking script that collects traffic data for all visitors; the rank can then be based on these actual traffic counts instead of on estimates from the extension. This service is estimated to be used by 1.06% of the top one million and 4% of the top 10,000.
The rank shown in a domain's profile on Alexa's website is based on data over three months, while the downloadable list was based on data over one month. However, since January 30, 2018 the list is based on data for one day; this was confirmed to us by Alexa but was otherwise unannounced.
Alexa's data collection method leads to a focus on sites that are visited in the top-level browsing context of a web browser (i.e. HTTP traffic). They also indicate that ranks worse than 100,000 are not statistically meaningful, and that for these sites small changes in measured traffic may cause large rank changes, negatively affecting the stability of the list.
Alexa is by far the most popular list used in recent security and network measurement studies: we found 133 papers from the four top-tier security conferences over the past four years to use Alexa's list, while only 3 other papers used another list and this always in conjunction with Alexa's list.
Cisco Umbrella publishes a daily updated list consisting of one million entries. Any domain name may be included, with it being ranked on the aggregated traffic counts of itself and all its subdomains.
The ranks calculated by Cisco Umbrella are based on DNS traffic to its two DNS resolvers (marketed as OpenDNS), claimed to amount to over 100 billion daily requests from 65 million users. Domains are ranked on the number of unique IPs issuing DNS queries for them Not all traffic is said to be used: instead the DNS data is sampled and "data normalization methodologies" are applied to reduce biases, taking the distribution of client IPs into account. Umbrella's data collection method means that non-browser-based traffic is also accounted for. A side-effect is that invalid domains are also included (e.g. internal domains such as *.ec2.internal for Amazon EC2 instances, or typos such as google.conm).
Majestic publishes the daily updated 'Majestic Million' list consisting of one million websites. The list comprises mostly pay-level domains, but includes subdomains for certain very popular sites (e.g. plus.google.com, en.wikipedia.org).
The ranks calculated by Majestic are based on backlinks to websites, obtained by a crawl of around 450 billion URLs over 120 days. Sites are ranked on the number of class C (IPv4 /24) subnets that refer to the site at least once. Majestic's data collection method means only domains linked to from other websites are considered, implying a bias towards browser-based traffic, however without counting actual page visits. Similarly to search engines, the completeness of their data is affected by how their crawler discovers websites.
Quantcast publishes a list of the websites visited the most in the United States. The size of the list varies daily, but usually is around 520,000 mostly pay-level domains; subdomains reflect sites that publish user content (e.g. blogspot.com, github.io). The list also includes 'hidden profiles', where sites are ranked but the domain is hidden.
The ranks calculated by Quantcast are based on the number of people visiting a site within the previous month, and comprises 'quantified' sites where Quantcast directly measures traffic through a tracking script as well as sites where Quantcast estimates traffic based on data from 'ISPs and toolbar providers'. These estimates are only calculated for traffic in the United States, with only quantified sites being ranked in other countries; the list of top sites also only considers US traffic. Moreover, while quantified sites see their visit count updated daily, estimated counts are only updated monthly, which may inflate the stability of the list.