The final polling projection for the 2016 Iowa Democratic Caucus is below; 87 total polls are included, 58 of which released gender sub-sample data:

Each projection above uses the default parameter settings for Local Regression with an alpha of 80% and a degree of 2. What this intuitively means is that the calculation of a given point uses 80% of preceding data points and weights the values to favor recency.

The overall projection incorporates every single poll available. The gender specific projections include only polls which released gender sub-samples. The re-weighting row, depicted last, utilizes the Turnout Projection concept to apply the 2008 exit poll gender breakdown of 43% male and 57% female. The overall projection and the gender re-weighting differ by about 1% in favor of Hillary Clinton.

The table above depicts the popular support associated with a given candidate. This is not how the results will be reported tonight. The results are reported in delegates as allocated by each participating precinct caucus. The allocation of delegates is complex and previously covered in more detail; there are however two allocation related items I want to discuss further.

In general a candidate must reach a certain threshold of support before delegates will be allocated. For precincts awarding 4 or more delegates this threshold is 15%. The threshold increases as fewer delegates are awarded. This means that O'Malley will likely receive close to zero delegates overall and his support will be absorbed into the two other candidates. The most recent Quinnipiac Poll suggests that O'Malley's supporters will slightly favor Clinton 50-47 while Public Policy Polling gives the edge to Bernie Sanders at 57-27. The projections for O'Malley, are smaller than the overall margin, so the absorption rate will not be decisive.

Secondly, the allocation of delegates is imprecise and varies depending on the group sizes formed by each candidate's supporters. Based on a modeling of the delegate selection process, both leading candidate's support falls within the range where group sizing can affect the result; at the 49% threshold, an additional 4.91% of delegates could be allocated if group sizes are optimized. This number is slightly more than the projected margin.

The general takeaway is that while the polling projections are likely to be accurate, there are variables not measured by the polls which could wholly affect the final outcome.

When the results are known, a pollster rating will be published that incorporates entrance polling to empirically compare the accuracy of polled samples with that of the actual outcome.