According to the official Alexa website’s Our Data page. The rank math using a ‘combination’ of the average daily unique visitors to the site. In addition to the number of pageviews on the site over the past three months. The site with the highest combination of unique visitors and pageviews rank as #1.
The data is collected from a subset of internet users using one of 25,000 browser extensions for either Google Chrome, Firefox and/or Internet Explorer. An algorithm then ‘corrects’ for various potential biases and attempts to compensate for visitors. Who may not be in Alexa’s measurement panel. A factor it hasn’t always tried to accommodate and normalizes the data based on the geographical location of visitors.
Alexa refers to the data retrieved from browsers as it’s Global Traffic Panel ‘, pulling data from millions of internet users in this way. However, Alexa also sources data directly from third parties to improve the accuracy of the tracking.
Finally, note that Alexa doesn’t provide separate rankings for sub-domains and sub-pages. It does make an exception to this rule for large blog networks, such as WordPress.com.
The general consensus seems to be a reluctant admittance by most (certainly not all). That there does indeed appear to be a rough correlation (seemingly with a LOT of outlying data points). Between a site’s Alexa Rank and traffic for well-established websites that receive over and above a certain level of traffic.
In the article ‘How are Alexa’s traffic rankings set?’ Alexa notes that it doesn’t receive enough data from its sources to make rankings beyond 100,000 statistically meaningful. It also notes that the closer a site gets to #1, the more reliable its rank. Any website that has a rank higher than this 100,000 waypoint can almost be completely left. Because the figure isn’t reliable.
Alexa isn’t able to gather its data from everybody. It obviously doesn’t have access to everybody’s browsing habits. As a result, it can only gather data from a subset of a few million users via certain browser extensions. A little common sense suggests this will almost certainly skew the data right from the get-go. Since the average user presumably isn’t likely to have installed any such browser extensions.