Trendency vs. 538, Explained

The rise in popularity of public polling during the last few campaign cycles has driven a more data-centric approach of politics. Polling aggregators like Real Clear Politics or HuffPollster provide a valuable service by providing a clearinghouse of polling data, making it easier to find long-term trends. The plethora of public data has also led to the rise in electoral models that look to predict the outcome of elections in important states like Iowa and New Hampshire.

Recently, Nate Silver of published his predictions for various of these early state votes. Silver offers two scenarios for each state, one model builds upon poll results only and one that includes several other variables, including endorsements and the national polling average. Additionally, for the New Hampshire model, 20 percent of the model is based upon the predicted results in Iowa.

Since both models are heavily reliant on polling data, you would think there would only be small adjustments between the two, but that is not the case. In Iowa, 538 gives Hillary Clinton a 65 percent chance of winning the caucus based on polling only, a 30-point lead over Bernie Sanders. Under their “Polls Plus” model, Hillary’s advantage jumps to 60 points (80-20 percent).

In New Hampshire, the difference between the models is even more apparent. There, when only looking at the polling data, Sanders has a 73 percent chance of winning the primary a week after Iowa. But once the other factors of Silver’s model are included, Clinton actually becomes the favored candidate, 53-47 percent (this data was pulled from this past weekend and may have been updated since this article was written).

At face value, there appear to be some relatively strong assumptions being made under the Polling Plus model. Not many people would give Clinton an 80 percent chance to take Iowa (although she may well end up winning there), and certainly fewer would look at recent polls and give her the edge in New Hampshire.

Using our Trendency Research data, we don’t run predictive models based on endorsements or other factors that are harder to quantify accurately. Instead we follow the trends of voter opinion as well as movement within their strength of support for a candidate. We have been looking at states like Iowa and New Hampshire since the middle of 2015 and we maintain a different opinion than Silver about the direction New Hampshire is headed.

In our surveys, we don’t ask for binary responses to questions. Instead, we allow users to allocate their choice of candidate on a sliding scale. They can apportion all their support to one person, or divide it among several if they haven’t made up their mind. Voters who log-in to take our surveys online do so on a regular basis over the course of several months and the site registers the changes (or lack thereof) in their opinions on each visit.

In the analysis, Trendency algorithms utilize Threshold Analysis to separate strong supporters from weak. Voters at higher Thresholds are more likely to cast a ballot for that candidate and less likely to switch their allegiance. As you move to lower Thresholds, voters begin to divide their support among more candidates who they may or may not end up voting for.

In the end, this allows for a more nuanced view of the horserace than simply topline results. For example, a voter logging into Trendency may give Sanders 100 percent support one month and then only 75 percent support the next. Under the conditions of a standard survey, this voter would still say they were voting for Bernie, even though the strength of their support has dipped. Trendency reads this movement and inputs it into the Threshold Analysis.

In New Hampshire, Bernie Sanders has enjoyed strong support among voters on Trendency Research. In fact, his likely success was apparent months ago. The reason behind this confidence lies with the strength of his support. At the 90 Threshold (strong, consistent voters), Sanders takes a whopping 80 percent of the support, and at the 75 Threshold, this support drops to a mere 69 percent. Each Threshold does not account for all people who will vote on February 7th, but it does show clearly that Sanders has a built in mechanism of core supporters who most likely will propel him to a decisive victory in New Hampshire, regardless of national polls and what happens in Iowa.

We are not in the business of predicting election results the same way Silver is, instead, we look at survey data both as a snapshot and over time. As it currently stands, our data is right in line with Silver’s polling only model: if the election were held today, Sanders would win handily. He has enough strong supporters willing to cast their votes for him, and in far greater numbers than those willing to do so for Clinton, that no number of endorsements or national polling stature can knock Bernie off his perch.