Optimal Citizen Forecasts of Multi-Party Elections

Citizen forecasts are among the most accurate election forecasting approaches.  Murr (Applying Condorcet's jury theorem to forecasting US presidential elections, International Journal of Forecasting 31: 916-929) proposed a method for making them even more accurate in two-party elections: first, select the model that in the previous election best predicted whether the citizen forecast was correct; second, use this model to predict the probability of each citizen to make a correct forecast in the current election; and, finally, optimally weight and delegate the citizen forecasts based on their predicted probability of being correct.  Applying this method to nine US presidential elections, the author showed that it increased the proportion of correctly forecasted states and that it reduced the error of vote share forecasts.  I propose an extension of this method to multi-party elections, applying and evaluating it using eight British General Elections between 1964 and 2015.