Polls can’t be Trumped

 

Note – The views expressed here are my personal views. 

All market research has an element of “margin of error” with one exception – elections. There is no margin of error associated with polls and hence they can’t be trumped.

Psephology is a super specialized subset of market research that attempts to forecast election results. I don’t envy psephologists since they don’t have the cover of “margin of errors” to hide behind. A lot has been written about  predictions from polls having been wrong in the US (yet again!!) though at least one large agency has claimed its predictions were accurate.

With due humility I must say that I predicted correctly that Trump would win. In January 2024 an ex-colleague, Bill Marshall and I predicted that Trump would win the election. Through the course of the year, I obsessively read through every detail available on the polls and listened to as many podcasts as I could. Armed with my knowledge, in mid-August, I made the same prediction to a bunch of friends. In October 24, I upset a colleague of mine, Lucy Hovey, over dinner with the statement that “my professional reading of the poll results is that Trump will win”. I continued to pore over opinion poll data published and on the eve of the election I upset another American colleague over breakfast with the prediction that Trump would win.

My objective here is not to critique the opinion polls since I do not have the expertise to do so. Neither is my objective to provide a political view of the results. The intent of this blog is to share my learning’s from having gotten my prediction right and lessons for the market research industry.

There are three big learning’s for me

“Signals always come with noise: It is trying to separate out the two that makes the subject interesting.”- David Spiegelhalter

1. Identify Isotropic Signals – In January 2024, Bill and I made the prediction that Trump would win the election based on election results in 2023.

In October 2023, the center left Labor party was defeated in New Zealand in the elections. In November 2023, in the Netherlands the right-wing populist Party for Freedom won the most number of seats and all four parties of the incumbent coalition government suffered losses. Geert Willers, the leader of the Party of Freedom is a known ally of Trump. In November 2023, Javier Miller won the Presidential election in Argentina. It was a sign of discontent with status quo in the face of inflation of over 140% and 40% poverty.

The signals I take from these are a rise in anti-incumbency because of the cost-of-living crisis and an increased acceptance of right wing policies. Both these signals point to a Trump victory. The signal also played out in the UK elections where anti-incumbency led to the Conservative party being voted out after 14 years in power and the right wing Reform Party got 14% of the votes.

In August, I made the prediction by looking at the poll of polls and comparing the results to 2016. In August 2024, Kamala Harris was leading by 2% (Source: FiveThirtyEight Interactives). In August 2016, Hillary Clinton was leading by 7.5%. 

Looking at the past poll data with an anchor to real life events that followed allowed me to read the poll results alongside the biases in response survey patterns. The signal I took out was that the margin of lead of Kamala Harris was insufficient to lead to a win as the same biases of 2016 should be expected to play out again.

In October, the poll of polls showed, Kamala Harris still in the lead though the lead had dropped to 1.7%. I was confident in my prediction of a Trump victory because the poll lead was narrower than in 2016 and, the prediction market had diverged. By Mid-October on Polymarket, Trump had a 60% chance of winning vs. 40% for Harris.

The three signals – consistent pattern around the world, consistent pattern in the US over time and inconsistent pattern between polls and polymarket – are all isotropic and pointed to Trump winning. The signals came from looking beyond the poll data and connecting the dots across data sources.

Too often in market research, we are satisfied looking at a single source of information. In a dynamic, multi-faceted environment that is dangerous, naïve and potentially misleading.

“I can’t change the direction of the wind, but I can adjust my sails to always reach my destination. – Jimmy Dean”

2. Focus on what matters – As per Statista, the most important issues for voters in the US were inflation / prices (24%), immigration (13%), jobs & the economy (11%), abortion (9%), healthcare (8%) and climate and environment (8%).

It is unsurprising that the most important factor, by a large margin, was prices, as the US had seen an inflation of 4.1% in 2023 on the back of 8% in 2022. On this dimension, Kamala Harris had the burden of incumbency that plays against her. The quotes below from a New York Times article basis a focus group they did in October 24 is emblematic.

“Teri, 44, Wisconsin, independent, white, systems analyst - Even though we have well-paying jobs, we’re definitely cutting back a little more than we’re accustomed to — thinking twice about a vacation, trying to limit grocery costs. Seeing the price of things makes you wonder how families that aren’t as fortunate are making ends meet.”

“Gaylin, 31, Georgia, independent, white, stay-at-home mom - Kamala Harris is the current administration. She’s already a part of the problem that we’re having now.”

With jobs and economy again the anti-incumbency hamstrung Kamala Harris. Immigration has been a strong point for Trump.

Abortion is an interesting point to look at. It was on the ballot in 10 states and won in 7 of the 10 (In Florida a majority voted in favour but it did not meet the threshold of 60% for constitutional change in the state). It even won in four Republican states – Arizona, Nevada, Montana and Missouri. In all four of these states Trump won showing that it is not what matters most to people. I quote again from the New York Times article

“Traci, 54, Michigan, independent, white, case analyst - I think that Harris will continue to make abortions available for women in this country, which I think is a good thing. But abortions, to me, are not so much a political issue. They’re a social issue. I appreciate that Trump gave that right back to the states to deal with. For me, it’s not really a political issue or a voting issue.”

Too often in market research we focus on overall opinion and even worse we are often reductionist in using “archetypes” in describing results. In my opinion that is the lazy way out. The devil is in the detail which we must not gloss over. We must focus on what matters to people and not artificial constructs of “overall opinions and archetypes”.

“There is no such thing as public opinion. There is only published opinion.” Winston Churchill

3. Beware of factoids – The Oxford English Dictionary defines a factoid as “an item of unreliable information that is reported and repeated so often that it becomes accepted as fact”.

“Celebrity influencers” are a big craze in the world of marketing. If one went by the number of celebrities supporting candidates Kamala Harris, she would have won. After all she had all the A listers endorsing her -   Oprah Winfrey, Megan Thee Stallion, George Clooney, Leonardo DiCaprio, Bruce Springsteen, Winfrey, Ricky Martin, Beyonce, Lady Gaga, Jennifer Lopez, George Clooney, Lizzo, Katy Perry, Jon Bon Jovi, Cardi B and the biggest cultural icon of 2024 Taylor Swift. But people were not choosing which chocolate brand to buy but rather who they trusted to bring down prices. Indeed, the endorsement by celebrities lent an aura of being “out of touch” with the realities of ordinary people.

Then there is the debate that Trump won because he used “social media’ by going on a “podcast”. Well yes Trump did go on the Joe Rogan podcast (which got 40 million views on YouTube in just 3 days) but similarly Kamala Harris went on Call her Daddy with Alex Cooper. According to a USA Today / Suffolk University poll, of those who saw the Kamala Harris podcast only 34% were more likely to vote for her and 51% were less likely to vote for her. The equivalent numbers for Trump were 50% more likely to vote for him and only 28% less likely to vote for him. The issue is not the participation in social media and podcasts but rather appeal of the content. 

As researchers, we must try to gauge the mood of citizens by poring over media articles and social media. But we need to be aware of our biases and biases in media and social media. I quote from the New York Times article again

“Gaylin, 31, Georgia, independent, white, stay-at-home mom -I follow a lot of people on Instagram. You see all these celebrities posting their support for Kamala Harris. And then when you look at the comments, it’s “Go, Trump,” “Trump 2024,” all of this stuff. It seems more people are voting for Trump than you would believe. But then it seems like they want you to believe that Kamala Harris is in the lead. If it doesn’t go the way that it seems like it’s going to go, then it makes you think that there’s going to be something off with it.”

In summary – I offer three lessons for the market research industry

1. Don’t take a myopic view via one source of information. Look for patterns from multiple sources across time. Look for weak signals, particularly contradictory signals to help complete the jigsaw.

2. Look at the data holistically. Avoid the temptation of going with headline summary measures but rather cut to what really matters.

3. As “truth tellers” we need to be extra cautious about our biases and not look for “confirmation of our biases”. We need to challenge status quo but we need to be cautious about getting swept up by the latest fads. Do not “smell the exhaust” or “drink the kool aid” depending on which side of the Atlantic you are on.