Friday, November 7, 2014

Why the 2014 Election was Hard to Prognosticate

My 2014 election model projections were as follows:

U.S. House: Actual: 248 Republicans, Model: 250 Republicans, Delta: +2

U.S. Senate: Actual: 54 Republicans, Model: 51 Republicans, Delta: -3

Governorships: Actual: 30 Republicans, Model: 28 Republicans, Delta: -2

The Polls

The Polls were skewed towards Democrats by average of 4.95 points in governor races and 3.5 points for Senate races. In my personal House measurement, it favored Democrats by 6 points. Sometimes we are only as good as the data we receive. That being said, I was one of the only people to project at least 15 house seat gains. Besides, it is hard to prognosticate an outcome such as the Republicans defeating four incumbent Senators when they have only defeated 2 sitting Senators the past 25 years (not a trend). And it is even more difficult to find a trend that would lead us to believe that Republicans would win Governorships in 3 of the most liberal states in the Union (Illinois, Massachusetts, and Maryland) in the same year. So this year was an anomaly and it could be the start of new trend.

To better understand the 2014 election results I ran a few models. The independent variable was the final polls, actual results, and the difference between the two. The x variables were the Partisan Voter Index (PVI – an index that defines the ideology makeup of a state or house district – for instance, R5, indicates the state or district leans Republican by 5 percentage points and D5 would be the opposite), Incumbency, and the final partisan makeup of congress and the governorships. An evaluation of the poll data model and its constant and parameter coefficients revealed:

· The House generic ballot would favor Republicans by 1.385 points

· The Senate generic ballot would favor Republicans by 1.236 points

· The Governor generic ballot would favor Republicans by 2.582 points

The final or actual results told a much a different story:

· The House generic ballot favored Republicans by 3.796 points

· The Senate generic ballot favored Republicans by 0.819 points

· The Governor generic ballot favored Republicans by 5.802 points

The final House Generic Ballot poll was 2.2 points in favor of Republicans when it was actually 3.8. In fact, if the above 3 results are averaged the overall generic ballot favored Republicans by over 3.5 points when the final polls indicated a House generic ballot of only 1.4 and an average of only 1.7 points overall (far below the 2.2 in the polls – you can read my blog on the dichotomy of two polls). In the course of history a 2.2 or 1.4 margin for Republicans is huge let alone 3.5 or 3.8 points! This certainly can explain account for the huge Democratic bias in the polls.

The above results can be explained further by evaluating the PVI (Partisan Voting Index). Republicans won 9 seats in the Senate, but the gains were in traditional red states: Alaska, Arkansas, Louisiana, Georgia, North Carolina, South Dakota, Montana, and West Virginia and in purple Iowa. The data shows people were much more partisan with their Senate vote. And these results also explain why polls were more accurate for Senate races as well as why the generic ballot only favored Republicans by about 1 percentage point in these races.

The actual results showed people were significantly less partisan with their votes especially in the House and in governor races. Republicans picked up multiple house seats in Democratic strongholds in states such as California, Illinois, and New York. Republicans won governorships in deep blue states such as Massachusetts, Illinois, Michigan, New Mexico, Nevada, Maine, Wisconsin, and Maryland. And they won a few of these races by huge margins. They nearly won huge upsets in states like Vermont, Rhode Island, and Connecticut. Governor races were also much tighter in states like New York and Oregon than suspected. This explains the partisanship and why Republicans had a huge generic ballot advantage.

Based on polling, the models also indicated that following:

· House incumbents had a 3.8 point advantage with an actual PVI coefficient of 0.38

· Senate incumbents had a 7.1 point advantage with an actual PVI coefficient of 1.5

· Governor incumbents had a 5.3 point advantage with an actual PVI coefficient of 0.53

My model includes open seats in the incumbent data (in other words, whomever held the seat last Republican or Democrat is considered the incumbent). Historically these incumbent advantage numbers are low. We know incumbents have an advantage going into elections because incumbents are hard to beat. Since most Senate races were in Red states and these races had voters vote along party lines (high PVI), it lead to the gain of 9 seats in the Senate even though the Republican advantage in the generic poll was not that high. That being said, I am willing to believe the Senate generic ballot was much higher than 1%. And even though the house and governor races saw voters be less partisan, the huge generic ballot in the Republicans favor helped them overcome this and win many seats. In fact, it shows more independents and Democrats were voting for Republican candidates. This explains why Republicans did well in New York, Illinois, Massachusetts, California and Maryland. Massachusetts and Illinois are good example because they had both a governor and senator races. The Republican won the Governor races, but were destroyed in the Senate races (30 point difference in Massachusetts and 15 points in Illinois). This explains exactly what the models show: Senate races vote along partisan lines and the governor races did not. About 20% of Senate races changed hands, 15% of governor races, and 5% of house races. And this would lead me to believe the low PVI played an important role in this for house races. Since house races are far less partisan getting seats to flip is harder despite a low advantage and high generic ballot. Besides, Republicans were expected to have a fairly high generic ballot in house races because they held nearly 60% of the seats to begin with.

No comments:

Post a Comment