Testing voting transfer

Two reliable survey data sources are already available. Taking into account the relevant literature and forecasting practices, we are switching from general fixed probabilities to a model where likelihood to vote is based both on actually casted votes and surveyed changes in the intention to vote. Memories of past voting are unreliable though, with a surprising number of voters stating to have voted options different than those stated prior to vote, with parties perceived as winners being remembered much more than voted. Still, we think this approach adds more information than the previos model, so we have updated the algorithm.

Our estimation still does not account for circumscription bounding (the case of Democracia Ourensana). We hope to be able to solve this issue, because this party is starting to be increasingly recognized as a relevant one.

Weighted probability