The 2016 presidential election changed the landscape of the American presidency forever and 2020 was close to bringing an encore performance. Joe Biden entered the election as a 90% favorite according to Nate Silver’s statistical models with the outcome being decidedly closer than most polls and pundits anticipated. To make matters surrounding the outcome even more confusing, Donald Trump injected disorder into the electoral process by proclaiming victory, announcing that mail-in votes were fraudulent, and demanding that the count be stopped. These actions all run counter to the democratic values our nation’s forefathers stood by and place a shroud of doubt over the ideals that the United States has championed since its inception.
The tension of the situation is heightened by the fact that our incumbent president is the one making these assertions, but if he is confused over his lack of mail-in votes, he has only himself to blame – his attempts to discredit mail-in voting cast doubt into a contingent of voters that played a decisive role in determining this race. Notably, Trump’s rhetoric pertaining to the outcome of the election has reinforced Us versus Them alignment in political spheres. This attempt to make Americans doubt our democratic process socializes the risk of losing the election to citizens as it instigates our existing political divide, creating a new issue for Americans to take a side on [1]. Consequently, determining the legitimacy of this election has become one of many polarizing political issues, but Trump cannot be singled out as the sole perpetrator of the polarization we witness today. Indeed, social media also plays an incredibly large role, for it has an incentive to increase polarization for the sake of engagement, while our biological response to social media reinforces the Us Versus Them divide.
Us Versus Them: The Creation of Social Hierarchies
Most of you reading this post know what is like to identify with a group; I assume most readers are supporters of sports teams. There are moments when identifying with a team is so strong that, when coupled with alcohol, fights break out at sporting events involving fans from opposing sides. Ridiculously, the only basis these individuals have for fighting is quite literally the name of the team they elect to support! Similarly, we are all Americans, yet find ourselves divided over different political views and this occurs even when alcohol is not involved.
In his masterpiece on human behavior Behave, Robert Sapolsky discusses the nuances of the Us Versus Them divide. Here is some of his exposition to introduce the idea:
Us/Them-ing typically involves inflating the merits of Us concerning core values – we are more correct, wise, moral, and worthy when it comes to knowing what the gods want/running the economy/raising kids/ fighting this war. Us-ness also involves inflating the merits of our arbitrary markers, and that can take some work – rationalizing why our food is tastier, our music more moving, our language more logical or poetic… [Conversely, we view] Them as threatening, angry, and untrustworthy… But Thems do not solely evoke a sense of menace; sometimes it’s disgust.
It is no wonder then that political tensions are so high – humans have a positive appraisal of Us and a negative appraisal of Them. But how are these dichotomies created, and more importantly, what factors are at play?
Social Media and the Great Divide
Social media platforms are central to the manifestation of Us and Them fractures. Platforms such as Facebook, Twitter, and Instagram, affect us biologically: “in 10 minutes of social media time, oxytocin levels can rise as much as 13%, a hormonal spike equivalent to some people on their wedding day [2].” Additionally, as Sapolsky states “[oxytocin] prompts trust, generosity, and cooperation toward Us but crappier behavior toward Them – more preemptive aggression in economic play, more advocacy of sacrificing Them (but not Us) for the greater good. Oxytocin exaggerates Us/Them-ing.” This line of logic reveals that each time you receive a like on a picture or get feedback on a post and are rewarded with an oxytocin hit, you are biologically primed to feel an affinity for Us and distaste toward Them!
Additionally, we are drawn to mimicry subconsciously. Mimicry is flattering, and we feel positively toward those who act, think, speak, talk, and dress like us. “An unconscious Us-ness is born from someone slouching in a chair like you do,” writes Sapolsky. Consequently, our oxytocin-induced affinity for Us is reinforced by the individuals you interact with on social media – you have a biological disposition to construct social graphs with those similar to you. As a counterpoint: how many of your friends on Facebook share different political views? I would imagine these individuals to be in the minority of many social graphs, and this is by design.
Social media companies generate profit by keeping users engaged on their platforms. To do this successfully, users must enjoy the experience they are receiving – if users have a strong distaste for the content that is appearing in their feeds, they will simply stop engaging. Consequently, social media companies invest resources to determine how to increase user engagement [3]. This is accomplished by implementing statistical methods that analyze what users might be interested in, based on a comparable set of users. Thus, conditional probabilities are the engine driving user engagement.
For instance, suppose that I tweet oranges are the best fruit. Twitter – leveraging the information I have freely offered – will look at tweets and advertisements that other orange loving users engage with and begin to populate my feed accordingly. Consequently, I will very quickly start to see advertisements for vitamin C and orange juice in my feed, as these advertisements had the highest probability of piquing my interest [4]. Additionally, I might see the tweets of other staunch orange supporters on Twitter and consequently adopt some of their views and opinions because they believe in the purity of the orange just as I do – mimicry is a powerful phenomenon.
To illustrate the discussion above, consider an imaginary social media platform named Juice and the following statistics on Juice’s 200 users:
The chart above conveys that 100 Juice users believe oranges and apples are the best fruit, 51 and 49 users believe orange juice and apple juice are the best drinks respectively, and 100 users have no preference when it comes to choosing a drink. But, notably, every Juice user has a preference as to what they believe the best fruit is. Now, suppose that on any given day there is a 1% chance that a Juice user who has no preference for a certain type of drink develops a preference for apple juice if they like apples, or orange juice if they like oranges. Furthermore, we are assuming that this conversion occurs due to Juice’s algorithms maximizing engagement, coupled with our biological disposition to form groups of Us and Them on social media. Logically, all of Juice’s users will eventually be evenly divided between apple juice and orange juice over time. The question then becomes, how long does this process take?
As shown in the graph above, there is meaningful dispersion in the number of years it takes for all Juice members to develop a preference for, well, juice. In my simulation, 50% of the time users were polarized in 15 years or less, but this also took as few as 11 years and as many as 22 years to play out.
The real world is far more complex than our idealized example, but the point remains that polarization takes time. People flip back and forth from one preference to another and there is no fixed probability of transition [5]. Indeed, given the efficacy of social media engagement algorithms, coupled with our biological disposition toward forming groups of Us and Them, moderate views are likely to be marginalized over time. Moderates are outcasts of both Us and Them in the political landscape. Belonging to an Us in the political sphere necessitates choosing a side. Avoiding this would require an abundance of mental fortitude and intellectual independence, especially when multiple forces are pulling you in different directions.
Yet another unique feature about Us versus Them dynamics is that “sometimes you help Us by directly helping Us, sometimes by hurting them,” per Sapolsky. This notion has manifested quite clearly in political spheres – many would state that the overall state of politics in America is in disarray, and this is largely a function of how winning is determined in zero-sum games – Their loss is Our gain [6]. This is a frightening situation; one political party “outperforming” another is irrelevant if the state of the nation politicians serve is deteriorating. In Sapolsky’s more light-hearted manner, “[it’s] not a great mindset to think you’ve won World War III if afterward Us have two mud huts and three fire sticks and They have only one of each.” We need to find a way to make the game of politics positive-sum for the citizens that depend on the outcome. This starts with holding leaders accountable – the time for dogmatism is over. We the people demand results.
Changing the game of politics, and combating dogmatism, can only occur alongside a detente of ideologies. We are a long way from this currently and it may only get worse; there are a large number of individuals who unconditionally adopt the values of the party they support, making it difficult for nuance to exist in the landscape of politics. To that end, “an arbitrary symbol of an Us core value gradually takes on a life and power of its own, becoming the signified instead of the signifier [7].” Political platforms and branding facilitate Us/Theming, making it easy to identify and partition Us and Thems on social media. Elephants and donkeys; BLM and MAGA – this is all shorthand for communicating your perceived position to Us’ and is certain to incite anger in Them’s, irrespective of any nuance which may exist in your true beliefs. In the world of Us versus Them, there is no room on the margin.
The Statistically Silenced
America’s tolerance for intolerance reached its breaking point in May of this year, culminating in a supernova of unprecedented social activism under the backdrop of lockdown. Now, perhaps more than ever, being perceived as an individual who is against social and racial progress is unacceptable. Indeed, we witnessed the consequences of this perception play out regarding Amy Cooper who was fired the day after her infamous video surfaced. Consequently, any individual who is aware of the Amy Cooper incident, but may sympathize with political groups that are perceived as anti-social and anti-racial justice, is unlikely to reveal their true stance because they are aware of the downside that comes with that perception. This is why social media discourse is largely filled with progressive discourse (or perhaps this is the bias of my own social graph shining through) – no rational person wants to be perceived as being indifferent or against social and racial progress because they are aware of the potential consequences.
For pollsters, pundits, and statisticians relying on traditional polling mechanisms to drive insights behind election behavior, questions regarding one’s beliefs on racial and social equality are subject to sampling bias. To that end, data-mining social media is unlikely to yield meaningful insights for the same reason mentioned at the end of the preceding paragraph. This presents serious issues for statisticians and pollsters looking to accurately assess the pulse of America’s political landscape in the days leading up to an election, as we saw in 2016 and as we witnessed play out in 2020.
To substantiate the claims above, I will compare the New York Times’ final pre-election poll for the swing states Arizona, Florida, Pennsylvania, and Wisconsin, to the observed election results. Below is a breakdown of New York Times’ poll and our focus will be on the Spread and Margin of Error (MOE) columns. The Spread is simply the difference between Biden’s polling results and Trump’s, where a Spread greater than 0 implies a Biden win – more care is required in defining the Spread’s MOE.
The margin of error is computed using elementary statistics. Given a 95% confidence interval, the MOE implies that if the true election day spread falls outside of the Spread +/- MOE, such an outcome had a 5% probability of happening by chance alone [8]. Consequently, if the observed election day spread falls outside of our computed lower and upper bound spreads, then we should reject the null hypothesis that our sample Spread (and distribution assumptions) accurately reflect the election day (population) Spread. The table below shows our 95% Spread confidence intervals for each state, along with the observed election day spreads and their computed probabilities:
These results are fairly encouraging – the observed election day spread’s for AZ, FL, and PA are contained within our confidence intervals, but we would reject our null hypothesis for FL and PA under a 90% confidence level. As a result, we might be slightly skeptical that our sample for these states accurately reflects the population of election day voters. Conversely, under our assumption that the Wisconsin poll accurately reflected the election day population, there was only a .08% chance of observing the state’s election day spread of 0.62 [9]! Consequently, the polls almost missed the mark in FL and PA, and completely missed the mark in WI – but why?
In the first gray sampling table, the Undecided column reflects the percent of those sampled who had had no preference between Trump and Biden or who were voting for a different candidate. Using the true election day results and our sampled Undecided group, we can estimate the proportion of the poll-implied Undecided group that truly did have a preference but abstained from sharing that opinion. If our assertion that sampling bias is created because no rational person wants to be perceived as being indifferent or against social and racial progress is true, then we would expect Trump to capture a larger share of Undecided voters as implied by the New York Times polls strictly due to the lack of nuance that exists when regarding Us versus Them; the views of the individual are always secondary to the overall perception of Them [10]. A contingent of Trump supporters may support social and racial progress, but simply weigh other criteria more heavily and consequently do not want to incur the downside of being perceived as anti-progressive, keeping their support private. With this narrative crafted, here are the results:
The table above implies that a majority of the poll-implied Undecided group were Trump supporters who were potentially disincentivized to share their true opinions due to the perceived risks of being labeled a Them (anti-progressive) and incurring the downside that comes with that perception. Additionally, Wisconsin’s results also make sense. Since the Undecided column could not explain all of the votes Trump obtained, this implies that our poll sample for WI was a poor representation of the population – there were more Trump supporters than the poll captured. Another feasible explanation is that the individuals in the poll flipped their stance come election day, for “[t]here can be striking discrepancies in Us/Them relations between what people claim they believe and how they act – consider differences between election poll results and election results [11].”
The plot above shows what our theoretical election day population may have been assuming it was normally distributed. Given the share of implied undecided votes that were cast for Trump in the states mentioned above, it was challenging to gauge the pulse of the nation heading into this election because of the imbalance in risk associated with supporting Trump instead of Biden. Due to this inherent risk, the left tail of our election day population does not easily surface under polling or data-mining. In a world predicated on signal, this is a source of noise that we will have to address unless we can reverse the course of polarization moving forward.
We the People
In Biden’s first speech after becoming president-elect, he made it clear that he did not see Us and Them, simply We the People:
I pledge to be a president who seeks not to divide, but unify. Who doesn’t see red states and blue states, only sees the United States… It is time to put away the harsh rhetoric, lower the temperature, see each other again, listen to each other again, and to make progress, we have to stop treating our opponents as an enemy. They are not our enemies: They are Americans — they are Americans. The Bible tells us, to everything there is a season, a time to build, a time to reap and a time to sow and a time to heal. This is the time to heal in America.
This is the speech of a man who understands the fracture that exists in our nation today. As I stated earlier, there is no single individual or entity at fault for our polarized state. The rhetoric used in Biden’s speech is an attempt to heal the divide that exists in our nation, for this fracture has seeped into areas that simply should not be politicized. Consider wearing a mask, which “has become a statement about masculinity, politics, or social freedoms depending on the context,” to borrow from The Arrival of Entropy. The politicization of actions such as mask-wearing has created zero-sum games that could leave Us with three sticks, and Them with one. Again, Biden understands this, tweeting: “Wearing a mask isn’t a political statement — it’s a patriotic duty.”
Our president-elect is doing his best to address our fractures with his rhetoric. Since many working-class individuals view the Democratic party as “[a party] of coastal urban elites who are any more concerned about policing various cultural issues than improving their way of life that has been declining for years,” Biden must speak in this manner [12]. There is no more room for Us and Them – strictly Americans.
I echo his sentiment by stating that it is time to come together. This piece was meant to reveal the dynamics of Us Versus Them so that we as a nation can do our best to combat the divide it creates. This requires an exercise in empathy, an understanding of what drives this divide, and a bit of intellectual independence. It is impossible to win a battle against biology but being aware of our biases can help limit biology’s margin of victory. Hopefully, four years from now, our differences have been reconciled enough where statistics can finally cut through the noise, finding the pulse of a nation that has been reinvigorated.
□
[1] This idea is developed in far more depth here https://notesfromdisgraceland.wordpress.com/2020/07/19/violence-power/.
[2] https://buffer.com/resources/psychology-of-social-media/
[3] TikTok has been remarkably successful to this end, altering the engagement model for social media companies everywhere. Its ability to assess what users enjoy is echelons above competitors because TikTok curates content for users without the need for social graphs (i.e. followers). Just by examining a user’s browsing history and the few videos they have been engaged with, its sophisticated algorithms generate tailor-made feeds. Indeed, no two TikTok users’ feeds are the same, and that is precisely the point.
[4] By highest probability, I mean the highest probability from a Bayesian perspective. This probability would be the conditional probability of engagement Twitter estimated based on my comparable set of users. This prior would be adjusted after they receive feedback on my engagement. Additionally, you could see how this mechanism could impact one’s political views.
[5] Our simulation is a non-ergodic Markov Chain. Ergodicity cannot be ignored in the real world, and furthermore the probability of conversion is certainly dynamic.
[6] A zero-sum game is a game in which the resources the winner receives were obtained from the loser.
[7] Sapolsky, again.
[8] This is simply the interpretation of a 95% confidence interval – if we conducted an infinite number of polls and constructed confidence intervals from their results, 95% of our CI’s would contain the true population parameter. Also, pardon my use of “5% probability,” it is easier on the eyes than .05.
[9] Again, to be more precise: the chance of observing an outcome at least as extreme as Wisconsin’s election day spread of 0.62 is .08% and we would reject the null hypothesis that our sample Spread accurately reflects the population at the 90%, 95%, and 99% confidence levels.
[10] My remarks here are derived from media commentary that paints Trump supporters as anti-progressive; take this op-ed here, or again here. Additionally, here is another piece (that takes a view counter to the first two links) discussing an idea similar to the one I share in this post.
[11] ibid.
[12] Attributed to Andrew Yang.