Not since the failure to predict Truman’s victory over Dewey in the 1948 presidential election have we seen a polling miss like we saw this week. Sure, there have been other misses, but it’s rare that we see such a universal failure. No pollster of any repute picked Donald Trump to win. And, the one poll that had Trump ahead of Hillary Clinton significantly over-estimated his final vote (Clinton won the popular vote, not Trump). So, they got the winner right, but for the wrong reason.
Some of the explanations offered are simply out of touch with how modern political polls are done.
A good example is that voters don’t respond to phone polls so a significant “hidden vote” tipped the election to Trump. Except that a number of the major polls were conducted online, not by phone.
READ MORE: Why did the polls fail to predict a Donald Trump presidency?
What’s important to remember is that pollsters and their media clients want to get it right. They invest significant money and effort to conduct the best polls for just that reason. They are constantly running experiments and refining their methods. So, if something fails, it’s not because they didn’t do their best; it’s because something happened they didn’t expect. And, that’s exactly what happened this time.
The average voter turnout in the four Presidential elections since 2000 has been 59 per cent. Pollsters this time expected turnout to be similar to what it was in the recent past. They built their models based on this assumption.
That’s not what happened; turnout was far lower. Estimates are that it was between 52-55 per cent (votes are still being counted).
READ MORE: U.S. Presidential Election 2016: Voter turnout drops to near 50 per cent
Lower turnout would be fine if there weren’t partisan implications. But, our models for the Reuters-Ipsos poll showed that higher turnout favored Clinton, while lower turnout favored Trump. This was, of course, reported but few bothered to take notice.
In the end, a quick glance at the vote totals won by each candidate tells the story. Even though this was a low turnout election, Trump delivered a greater share of his potential vote on election day. Notably, it was a smaller number of votes than either John McCain or Mitt Romney won when they ran against Barack Obama. This tells us that there was no “hidden Trump vote” missed by the pollsters.
WATCH: Donald Trump’s victory helped by the rust belt revolt
Even though Trump had a lower number of votes than the last two Republican Presidential candidates, he delivered them VERY efficiently. He won a number of states that Clinton lead in the pre-election polls with very small margins. The polls had either given them to Hillary Clinton or had called them as toss ups. If we had a better estimate of the potential voter turnout this wouldn’t have happened.
The lower voter turnout was really about the 6-7 million voters who showed up for Barack Obama, but took a pass on Hillary Clinton. This is what the polls missed. If we had used a turnout factor that was closer to what actually happened, the polling would have been more accurate and the miss less universal.
After the big miss in 1948, pollsters upped their game by introducing a new innovation – random sampling. Random sampling has been responsible for the tremendous record of accuracy for election polls all over the world. In a similar way, I suspect, based on what happened this time, the most important innovation that will emerge will be a new way to model election turnout. After all, we always learn the most from our mistakes.
Darrell Bricker is the CEO of Ipsos Public Affairs.
- Iran fires air defences at military base after suspected Israeli drone attack
- U.S. aid to Ukraine, Israel set to pass. But bills differ in one key area
- Carbon rebate labelling in bank deposits fuelling confusion, minister says
- Conservatives ask interference inquiry judge to rule elections were flawed
Comments