[saeranv]

Both 538 and the Economist forecasting models model errors as correlated - each state's vote share is not considered an independent event.