An independent estimate of the Google PageRank

Using various continuously updated information available using different search engines, it is possible to estimate the Google PageRank of a site (i.e., the Page Rank of a site's main webpage) without using Google's PageRank servers (as is done at this site, or this site for example).

The estimate itself uses the size of the site, the number of back links to it in yahoo, MSN and altavista, as well as the number of internal links within the site, i.e., without using any google data. The estimate PageRank "formula" was obtained by carrying out a χ² fit between a predicted PR, modeled using the aforementioned variables, and the actual PR for 120 or so randomly chosen websites.

The result was a fitting formula that gave an average error of 0.9 between the predicted and real PR. In just over 50% of the cases, the PR was right on the mark, while in less than 10% of the cases, it is more than one PR point off (i.e., in just over 90% of the cases, the predicted PR is the correct or ±1.)

Note that large variations between the estimate and actual google Page Rank can exist because:

• The site has large variations, in which case the Google PR is not "up to date" with the current status of the site: The backlink information updates on a time scale of days, while Google's PR on months. In otherwords, the estimated PR can be used as a prediction or forecast.
• A site can be banned. In such a case, the real google PR would be 0, while the estimated one larger.
• Various other anomalies. For example, your site could have a few but very high quality links. In such a case, the estimated PR would underestimate the real PR.

Since the PR estimate is not accurate, a better use of the tool would be to follow changes over large periods. e.g., if you see a large increase over a several month period, expect your real Google PR to increase as well.

You can read more about the analysis in my blog entry on the Google PR.

A typical query time is about 5 seconds.

This is a beta version. The fitting formula will be updated after the next PR update to remove systematic errors (which may be important for PR<=2 pages)