Purpose
The main idea behind this site was to make my prediction aggregation and model in a more accessible form. For previous elections I have been able to accurately precinct the results by averaging averages and accounting for state and nationwide polling errors.
By looking at multiple sources it helps to eliminate the bias and mistakes that one aggregator or model might have.
Averages of Averages
Each source is given equal weight when making the predictions. After an average is calculated an adjustment will be applied where applicable. Not all sources can be used at all times and they will be listed as unavailable if there is no data.
Current Calculations
The averages currently use VoteHub, Silver Bulletin, CNN, Race to WH, and FiftyPlusOne. The New York Times is unavailable for the generic ballot because there is no usable national generic ballot average available from that source.
The generic ballot is also modified by my algorithm in an attempt to better account for polling errors. I will update to show what amount the adjustment is on the main page.
For the House, Senate, and Governor predictions I take the modified generic ballot and apply that to the current polling numbers from each state accounting for past performances of the various pollsters that made polls for each state.
As much as I would like to put the math for each Congressional and gubernatorial race on the site it would be a pain in my ass to do, if you have any questions about any calculations I would be happy to answer though.
Algorithm adjustment
The algorithm adjustment comes from looking at polling errors from previous elections and accounting for that as well as accounting for voting propensity of various groups in midterm elections.