So everyone and their brother has an election model these days I figured I should try and put one together too.

It's a fairly simple model and it samples polls via the

Huffington Post Pollster API. I use a median filter moving average to estimate the likely spread between Democratic and Republican (D-R) percentages for each state for which polling is available.

There are many states for which no polling is available yet. Default D-R spreads for each unknown state are assigned using a heuristic: Deep South states have a strong (20 point) bias in favor of Republicans...Unpolled New England states have a 5 point bias toward Democrats. Each unpolled state has an implicit variance in this estimate derived from the average margin of error in polled states.

For states with polling, state variance is estimated from the average variance of the most recent polls from that state (usually between 3 and 5 polls). If there are no polls in the current monthly interval, we estimate a trend from older data and calculate the likely D-R spread based on this trendline.

Electoral votes are assigned to Democrats for positive D-R speads, Republicans for negative spreads.

The model is run 10000 times, with random noise added to each spreads based on the variances estimated above.

Today's model run predicts an

**80.6% probability of a D win** and

**19.4% probability of an R win**, with an

**average of 290 electoral votes for the Democrat and a median and mode of 292 electoral votes.**
An histogram of outcomes for Democratic Electoral Votes - Republican Electoral Votes is shown below. Outcomes above zero are scenarios where Democrats win the electoral vote.