Far more accurate then the pollsters...
The following is a general description from their website of the methodology used:
Election Prediction Project attempt to predict the result of elections, through:
Making prediction of the overall result of an election by predicting the result of individual electoral districts.
Making prediction of the result of individual electoral districts through analysis of factual information and punditry specific to the electoral districts.
The project solicits information and opinion submission from the general public prior to an election. You may contribute your insights to a specific electoral district through submission forms for specific elections. (Current Project: Ontario, Quebec) Only submissions addressing a specific electoral district will be posted on the website. Election Prediction Project has only conducted prediction on parliamentary elections, where formations of executive/government are based on gaining control of the legislature. There is currently no plan to conduct predictions on elections that generate separate legislative and executive powers, such as the US congressional/presidential elections, or municipal elections.
Submissions are collected, processed, and posted on a regular basis (more "regular" during writ period, and understandably as most of us are university students, less during university exam period). The evaluation panel, consists of individuals from various locations, exchanges their analysis on the submissions and concludes with a prediction.
The evaluation process is entirely subjective. Members of the evaluation panel have very different political background and often disagree. Predictions are undoubtedly influenced by panelists' specific insights, experiences, or biases about particular contests. However, we strive to come up with predictions that are most sensible as we see them.
The panel gives consideration for the following:
Integrity of the source - More weight are given to submissions with more information about the author, and particularly those where real names and email address are posted. (Please kindly report any non-functional email addresses posted so that we may correctly evaluated the postings in question.)
Objectivity of the source (1) - Less weight is given to submissions from a particular participant who demonstrated consistent bias. Submissions from individuals that demonstrated clear political motive or blatant bias are generally ignored.
Objectivity of the source (2) - Obviously, special treatment is given to submissions from individual who are politically related, such as political staffers, party organizers, student/union leaders or government employees. This in no way discourages you to submit and identify yourself as such. We always appreciate insider tips.
Quality of the posting (1) - Weight is given to submissions that are backed by intelligent and thoughtful reasons. Detailed reasons provide others viewing the site with evidence to back up your claim.
Quality of the posting (2) - Malicious, inaccurate, slanderous or misleading information are generally ignored, and will seriously reduce the weight given to other submissions from the source in question. Some members of the project have a particular tendency of appending satirical and sometimes demeaning "Editor Notes" to submissions that are just plain ignorant or obviously out of touch with the reality.
General trends - The panelists are all politically minded individuals who pay close attentions to news and issues. We also share polling information and gossips, which may influence the prediction.
Ipsos-Reid, Ekos, Strategic Council and all those other professional pollsters could learn a heck of a lot from their unconventional approach to providing election predictions. It would be interesting to see if the more traditional polling methods could be combined with an approach like this to provide even more accurate projections.