Home      Subscribe (free)    All Articles

The Box Travels

Polling
Polls misfire when predicting presidential elections

Polling Hocus Pocus

In the Senate Impeachment trial last week, lawyers for the prosecution cited polling data from the spring and summer of 2019 showing Joe Biden with a strong lead over President Trump in a hypothetical matchup for the 2020 election. These polls were given so much credence, they were presented as evidence supporting the allegation that President Trump improperly asked the President of Ukraine to investigate the unusual activities of Joe Biden and his son in Ukraine. Putting aside for the moment the fact that Biden would have to get past more than 20 competitors from his own party before he would ever face Trump in the 2020 election, there is a math problem with this logic.  

If polling is done objectively and scientifically, it can accurately reflect the majority opinion of voters. However, Presidential elections in the US do not use a majority vote to decide the winner. The electoral college system is used instead. This is one big reason why national polls are notoriously wrong at predicting the outcome of Presidential elections.

 It is a simple math problem that exists regardless of party affiliation or any bias that may be present in the poll questions. You have only to look back to the last presidential election in 2016 to see this effect. In the week before election day in 2016, Hillary Clinton was leading Donald Trump in the Real Clear Politics average of national polls by 3.3 percentage points. When the final vote was counted, Clinton ended up winning the popular vote by 2.1 percentage points – not exactly what the polls had predicted, but well within the margin of error, validating the accuracy of the pre-election poll.

Of course, Clinton lost the election because she lost the electoral college vote. Had there been a scientific poll devised to reflect the outcome of the electoral college voting system rather than a simple majority opinion, the pre-election poll results would’ve been different. Trump’s margin of victory in the electoral college turned out to be an overwhelming 15 percentage points, proving the peril of relying on presidential polls based on simple majority opinion.

 Even though the mechanism of the electoral college system is relatively easy to understand, it is extraordinarily difficult to fashion a scientific poll that accurately reflects the model. To do so would require doing fifty separate polls simultaneously (one in each state) and then weighting those results according to the number of electoral votes allotted to each state before adding them all up. Thus, it is highly improbable that President Trump, knowing better than anyone how meaningless presidential polls are at predicting a winner, would feel the least bit imperiled by a poll taken 18 months before the election showing Joe Biden as the winner. This is just another weak argument by the impeachment prosecutors that further devalues their already very flimsy case.

The outcome of the popular vote for President in any election is indifferent to how the voters are geographically distributed throughout the country. There is nothing inherently wrong with this unless a few large states become disproportionately populated with voters of a single political ideology. Should this occur, the possibility arises that just those few states could determine the election outcome by themselves. This results in a phenomenon akin to “mob-rule”1 dictating the outcome of national elections and is exactly what the authors of the Constitution sought to avoid.

They rightly believed it unfair that in a country composed of many sovereign states, the ideological philosophy of only a few states could potentially be imposed on a majority of all the states. The main purpose of the electoral college voting system was to guard against this. It forces candidates to achieve broad-based support across the entire country, as opposed to having overwhelming support in only a few densely populated states.

The 2016 election provided a perfect example of how this works. In California, the most populous state in the country, Hillary Clinton defeated Donald Trump by a whopping 4.3 million votes. California alone more than accounted for her national popular vote margin of victory of 2.9 million votes (all states combined). However, California’s electoral vote total is capped at 55 (53 House reps + 2 Senators) regardless of how big the raw vote total was for the winning candidate. Mrs. Clinton would’ve performed much better in the electoral college if her excess support in California was spread more evenly around the country into many of the smaller states where she lost. Her highly disproportionate support in California (in addition to a few other large states) is what led to her defeat in the presidential race. The electoral college system did precisely what it was designed to do, prevent the large states from determining alone, for the whole country, who the next president would be.

Another excellent example of how national polling data is poorly understood and wrongly used is a recent CNN poll that purported to show how leading democrat candidates for president would perform against Donald Trump in the general election. The poll result indicated that every one of the top six candidates would defeat President Trump. The candidates themselves seized on this information in their campaign speeches and on the debate stage as if it was reality. It isn’t. The most optimistic meaning of this poll is that if a national election were held today, each of these candidates might win the popular vote against President Trump. Just as in 2016, these results are meaningless for predicting who the actual winner will be unless the poll is somehow modeled on the geographical and mathematical underpinnings of the electoral college voting system.

We are convinced this simple mathematical fact also leads to incorrect polling data assumptions about the general thinking of the country on other questions not related to candidates. For example, the anti-Trump news media are all too willing to cite polling data showing nationwide support for Trump’s impeachment and removal from office. They want you to conclude that this poll result indirectly proves Trump has lost significant support among his voters. However, even if this poll result is accurate, it is equally likely that Trump’s support simply dropped even further than it was in the large population centers but remained steady (or even grew) broadly across the country. Thus, his standing on the electoral map would be unchanged. This isn’t an “alternate” reality as many media pundits like to describe, it is reality.

In addition to the opinion polls’ difficult, if not impossible, task of accurately predicting the outcome of an electoral college vote, there are other more subjective areas of polling inaccuracy, especially this far in advance of an election. There can be intentional bias in the wording of the questions, over-sampling of one party or the other, the effect of the tone of a live human voice vs. a recorded or synthesized voice asking the questions, calling cell phone numbers vs. only landline numbers, etc. Then there is a problem in the way that media outlets typically report the poll results. They usually cherry-pick one or two questions out of a much larger set and “spin” the result to make an eye-popping headline. For example, a headline resulting from the CNN poll mentioned earlier might say that the economy was extremely important to only 40% of those who planned to vote in the next election. But in this poll, the question about the importance of the economy gave respondents a choice of “extremely important”, “very important”, “moderately important”, and “not that important” as possible answers. If you add up all the degrees of positive response from the poll question (extremely-40% + very-40% + somewhat-18%), the poll reveals that the economy is important in some degree to 98% of voters! That’s a far cry from what the headline implies. The possibilities for introducing error, intentional or not, are almost endless. Looking at averages of polls (instead of individual polls) helps to a great degree, but even then, you must consider that the polls being averaged likely did not ask the exact same question in the exact same way.

We strongly believe the Democrats know that the public polling data they are using to assess how their candidates hypothetically perform against Trump is invalid and poorly supported. This is why, in our view, they are so intent on trying to remove Trump from office now via impeachment. Given the extremely long odds of that happening, we suspect they have already conceded Trump will be reelected. The only reason for them continuing to spin these recent polls as reality would be to support their own specious accusation in the impeachment trial that Trump has “rigged” the 2020 election. If Trump does get re-elected (as we believe they know will happen) they can point back to all these poll results as proof that he somehow cheated. Unfortunately, there will be millions of Americans that fall for this fiction.

What will be fascinating (again) in the aftermath of the upcoming election is how the Democrats and the news media will try to reconcile that an undisciplined, stupid, racist, and misogynistic Donald Trump was able to pull off a brilliant and undetectable massive cheating scheme to win an election all the polls say he should have lost. Will they replay the Russian story again or try something new?

1Thomas Jefferson and his fellow Founding Fathers deeply feared what they called “mob rule” that would result from a pure democracy. Jefferson stated – “A democracy is nothing more than mob rule, where fifty-one percent of the people may take away the rights of the other forty-nine.”

4 Responses

  1. Again, well done! In addition…………..

    Trump consistently under polled leading up to the 2016 election. It happened for most of the primaries and in the general election. In today’s polarized and toxic political environment, I believe there are many “closet” Trump supporters that don’t admit it publicly or to pollsters. It seems that we are seeing this phenomenon continue to play out and it will through the upcoming election cycle.

    During the 2012 election cycle, I displayed yard signs for Both Romney and McCrory. For the 2016 election, I did not display any yard signs despite my support for Trump. I think there are a lot of people out there with the same mentality.

    As with the Electoral College conundrum you mention, the democrat establishment is well aware of this and they are very concerned.

    1. Ray – you are absolutely right about Trump being under polled. This is an excellent point and something we should’ve pointed out in the article. I suspect there are even more “closet” supporters now then there were in 2016. The attendance at his recent rallies, even in Democrat controlled states, is a big clue.

      I think the constant leaking of information that normally would be private leads people to believe that polls aren’t really anonymous anymore. The suspicion that your poll answers might potentially be used against you somehow is keeping an unknown number of people from responding. Let’s hope those same people don’t have a similar fear of the ballot box.

  2. Another great blog topic! And I’m so happy to hear that the pollsters don’t (can’t) guess on electoral votes.
    I just wish they were not called polls. ‘Polls’ give legitimacy to nothing more than guesswork. I would argue there is nothing scientific about asking people questions on anything related to what they like or what they will do, unless all people were honest but they are not. Polling is done because people like the drama of… anything and everything. People are not computers that respond based on 1s and 0s. People are all innately programed to lie for defensive or acceptance reasons and they do. With any potentially consequential answer, people will say one thing when sitting with their mother, another if with their spouse, another with their personal buddies, another with work buddies and then…. will change as needed in each of those same scenarios if needed. For instance, you were with your mother and her closest personal friend – you would probably change your answer to avoid confrontation that would affect your mother’s friendship. But but but we should believe they would be truthful with a pollster. Pollsters guess. The entire % difference in political differences nationwide is a few percent – that we know from actual voting. Stop the drama and outlaw polls that only masquerade as something of meaning.

    1. Johnny – I agree about the human nature problem with polling. In addition to those you mention who change their tune depending on who they’re with, there is also the category of wiseacres who, if called by a pollster, simply want to play games and intentionally answer the questions falsely just to “mess up” the poll.

      Unfortunately, polls can’t be outlawed. But we can learn to take them, as well as the “spin” about them, with a big grain of salt.

Leave a Reply

Your email address will not be published. Required fields are marked *