One Year Later: On Presidential, Senate, and Gubernatorial Polling, 2004-18

 

A little over a year ago, I wrote a post about the 2018 election and tried to make predictions based on what I was seeing days before the election. As a politics junkie, election years are like a prolonged sports season for me and I feel a rush as Election Day comes ever closer and I wish to revisit this subject. I revisit my predictions and expand my investigations into polling. On my five predictions, I proved correct on all of them. The best resource to look at is poll averages, but I’ve noticed there have been some upsets. I have defined an “upset” as the result being the opposite of what was predicted through RealClearPolitics poll averages. Given the increase in upsets over the years, I have become interested in measuring average poll biases. Since RCP records with averages go back to 2004, I am looking back to this point to measure and compare average Senate and gubernatorial polls to results.

I have not been able to find if there were any RCP polling averages for gubernatorial races in 2004, but I found four tossup Senate races and of course, Bush vs. Kerry. Overall, there was an average Democratic poll bias of 0.9% among the five races, with the one upset being Republican Mel Martinez’s Senate victory in Florida. 0.9% was also the average poll bias towards Democratic nominee John Kerry. 2004 wasn’t really a year of surprises except for people who didn’t pay attention to poll averages.

2006 and 2008 were quite predictable years in the major tossup races, with there being no upsets at all in 2006 and only one in 2008, when Al Franken defeated Norm Coleman for reelection in Minnesota by a hair. In truth, as memory recalls, little was surprising about 2008. While Democrats faced a nightmare election in the House in 2010, they pulled off a few favorable upsets in the Senate, which could be attributed to the hardliner candidates the GOP nominated in Colorado’s Ken Buck and Nevada’s Sharron Angle, both of whom were projected in RCP averages to defeat the incumbents. The average polling error for the latter election was 8.3%. While 2010 was a blowout year in the House, it was merely a good year in the Senate. The following election cycle would prove a highly ironic one on the polling front.

I remember in 2012 that given some worrisome polling results and publicity, Republicans were eager to dismiss bad polls and there was even a blogger who wrote about poll skewing and was releasing “unskewed polls” that showed victories for the GOP. It turns out this guy was right…there was poll skewing but he got the direction wrong. In a most bitter irony for the Republican Party that was eager to believe in “skewed” polls against them (which I sheepishly admit I was one), the 2012 election constituted the height of Republican bias in polling, with an average bias of 3.8 points in tossup races. Montana and North Dakota proved the upsets in the cycle as Democrats kept these seats, with North Dakota being the biggest shocker. If that poor blogger had only chosen to write one election season later, however, he would have been right.

In 2014, the Republicans were on an upswing and that was the first year I worked on a successful state election campaign. I remembered 2012 so I was, in a manner of speaking, conservative in my enthusiasm. After all, although I didn’t know this at the time, the GOP had historically underperformed their polling numbers. That year’s Election Night party was triumphant and jubilant, unlike two years before. The GOP had outperformed average polling, and in this case it was by 2 points. This was also the year of the absolute worst average Senate polling blunder, which predicted that Kansas would elect Independent Greg Orman by a little under a point, but Republican Pat Roberts won reelection by almost 11 points.

2016 was the pinnacle of Democratic poll bias with an average of 3.4 points and in eleven of the thirteen tossup races there was an average polling bias to the Democrats. There were also not two, but three upsets in the Senate. Republicans Pat Toomey of Pennsylvania and Ron Johnson of Wisconsin won reelection while Democrat Maggie Hassan narrowly defeated New Hampshire Republican incumbent Kelly Ayotte. Easily the greatest upset of the three was Ron Johnson, as many pundits had been writing his political obituary earlier in the year and I thought he was certainly going to lose. Of course, the greatest upset of the year, indeed, the greatest upset in modern political history, was the election of President Donald Trump. Polling was biased by 1.1% towards Clinton and she won the popular vote by 2.1%, but she had not paid enough attention to certain critical states that Trump unexpectedly won, such as Michigan, Pennsylvania, and Wisconsin. The one upset in the gubernatorial races was the Republicans keeping Indiana in their column.

2018 was seemingly a year of contradictions: Democrats gained 41 seats in the House, but lost two in the Senate and won less gubernatorial races than expected. Indeed, this year there were six upsets, three in the Senate and three gubernatorial, with four of the six being Republican victories. Were Ohio to have an RCP average tabulated, it would have been there too as Democrat Richard Cordray was leading in polls over Republican Mike DeWine. Despite the record number of upsets in an election year, average polling bias was at its second lowest this year in tossup races, only 1.2% more Democratic than the actual vote.

In summation of biases:

2018 –  D +1.2

2016 – D +3.4

2014 – D +2

2012 – R +3.8

2010 – R +2.9

2008 – R +1.4

2006 – R +1.7

2004* – D +0.9

*- No gubernatorial races with RCP averages for 2004.

 

Leave a Reply

Fill in your details below or click an icon to log in:

WordPress.com Logo

You are commenting using your WordPress.com account. Log Out /  Change )

Google photo

You are commenting using your Google account. Log Out /  Change )

Twitter picture

You are commenting using your Twitter account. Log Out /  Change )

Facebook photo

You are commenting using your Facebook account. Log Out /  Change )

Connecting to %s