Racism and Underdetermination by Evidence

This week, Starbucks will be shutting down 8000 of its stores for one day. Employees at these locations will undergo anti-discrimination training, including arguably dubious efforts to combat implicit bias. And all of this is a response to the recent arrest of Rashon Nelson and Donte Robinson—both black men in their twenties—at a Philadelphia Starbucks, which triggered widespread condemnation and accusations that a culture of anti-black prejudice pervades the coffee chain.

Slightly different accounts of the incident have been given by different news outlets, but something like the following sequence of events seems to have taken place. Upon arriving at the Starbucks in Rittenhouse Square, Mr Nelson asked to use the restroom. Permission was refused by the manager, who told him that the facilities were for paying customers only. Mr Nelson and Mr Robinson then took a seat at a table. The manager asked them if she could bring them drinks or water, and they declined, saying they were waiting to meet someone. Mr Nelson and Mr Robinson were then asked to leave by the manager, on the grounds that they had not bought anything. They refused. The manager then phoned the police and reported that there were “two gentlemen in my cafe that are refusing to make a purchase or leave.” Police officers arrived soon afterwards and asked Mr Nelson and Mr Robinson to leave. Again, they refused. The police officers continued to speak to them while they waited for backup to arrive. During this time, Mr Nelson and Mr Robinson’s contact arrived and questioned the police, accusing them of discrimination. More police officers arrived, and Mr Nelson and Mr Robinson were arrested on suspicion of trespassing. Some of the incident was captured in a video, subsequently posted online. The unappetising image of two black men—who had remained civil and calm throughout the process—being led away in handcuffs for an apparent triviality, soon went viral.

At which point the Frappuccino hit the fan: a number of protests at Starbucks have been staged, with participants chanting slogans such as “Starbucks’ coffee is anti-black”; calls have been made to boycott the chain; the city of Philadelphia—in discussion with Mr Nelson and Mr Robinson—has agreed to provide a $200,000 grant to fund a programme supporting state high school students to become entrepreneurs; and the Philadelphia Police Commissioner Richard Ross apologised unreservedly (abrogating a previous defence of the police officers’ actions) and announced new guidelines for officers in similar situations. Within the organisation itself, Starbucks’s executive chairman and chief executive, Howard Schultz and Kevin Johnson, met with Mr Nelson and Mr Robinson to apologise personally. Starbucks is now discussing a number of changes to policy in response to the incident, and has reached an undisclosed financial settlement with the two men.

Despite the raft of measures taken in the wake of the incident, opinion is divided over whether this was really a case of racially motivated discrimination. For every person decrying the incident as evidence of systemic racism, both in Starbucks and society at large, there are others who are unable to find evidence of racial bias lying under the surface events. In the sprawling comment threads below myriad online articles about the incident, many have been quick to generalize from their own lives. Whites who have been asked to buy-or-leave sometimes appeal to their own experiences to dismiss any suggestion of impropriety. Conversely, the whites who have sat undisturbed in cafes without buying anything sometimes appeal to their experience to draw the opposite conclusion. Who is right? And, importantly, how can we tell?

In an interesting way, the Starbucks incident parallels one of the historical cruces of the scientific revolution. In 1613, when Galileo published his most important findings on planetary motion, they included a compelling piece of evidence for heliocentrism: the phases of Venus. The Galilean hypothesis that the Earth orbited the sun made it possible to explain and predict the different distributions of light observable on the planet at different times. But when Galileo presented these findings to the Catholic church (who, for various reasons, preferred the geocentric view) shortly after his publication, a number of astronomers pointed out that the phases of Venus observed by Galileo were consistent with an alternative geocentric view advanced by Tycho Brahe. Tycho believed, roughly, that all the planets orbited the sun but that the sun orbited the earth. Tycho’s astronomical view will sound strange to the modern ear, but it was a very popular theory of planetary motion in its time. This was, in part, because it made successful predictions regarding the movement of heavenly bodies—especially, for example, the seemingly backwards motion of planets.

What this example shows is that the same evidence used to support one theory can also be used to support a competing theory. Philosophers call this the problem of contrastive underdetermination. In other words, the theory is underdetermined in the sense that the evidence equally supports two competing hypotheses. In this case, the views of both Galileo and Tycho were supported empirically by the phases of Venus. Heliocentrism was, of course, eventually decisively vindicated, but not every hypothesis we might entertain is so unequivocal in this regard. Sometimes competing hypotheses will predict all the same observations or events, or, at least, all the observations and events we are in a position to make or witness. When this happens, it isn’t clear how, or if, the evidence could confirm one hypothesis over the other.

Does Racism Explain Black Disadvantage?

Sydney. London. Toronto.

QuilletteG. E. Zuriff

This problem is not merely academic. Nor is it restricted to issues in science. Underdetermination can get personal. Last year, one of the authors (Anthony) took a trip to Costco. He had to use the restroom and tried to enter via the exit (the bathrooms are located near the exit). As faithful Costco members know, for some locations, entering via the exit is a cardinal sin. He had just watched an older white man enter through the exit and thought he would try his luck. He was quickly stopped. Anthony is of Middle Eastern decent and had an unruly beard at the time, and he asked if he’d been stopped “because I’m not white?” The Costco employee did not take this well and some brief shouting took place. At the time, Anthony believed that the evidence supported the view that he had been stopped because he wasn’t white. After all, there is some evidence for this. A white man had entered through the exit and wasn’t stopped.

Faced with evidence like this, how can we go about weighing its strength? Though it wasn’t available in Galileo’s time, today there is a satisfyingly elegant mathematical theory of confirmation formulated by the eighteenth century mathematician and Presbyterian minister Thomas Bayes. It allows us to calculate the degree of support, or confirmation, a hypothesis has as a result of some evidence, and our overall knowledge. To do this, we need to know:

To what extent a hypothesis predicts the evidence offered in its support.
How likely the hypothesis was in the first place (i.e. prior to gaining this evidence).
How likely the evidence was to turn up anyway (i.e. even if the hypothesis didn’t hold).

The more likely the hypothesis in the first place based on independent background knowledge, and the more it predicts the evidence, the more probable the hypothesis given the evidence. Conversely, if the evidence was likely to turn up irrespective of whether or not the hypothesis is correct, then its explanatory power diminishes. Here is how this plays out in the Costco incident:

Evidence₁: A non-white male was prevented from entering through the exit at Costco.
Evidence₂: A white male was not prevented from entering through the same exit moments before.
Hypothesis₁: The Costco employee who stopped the non-white male is biased against non-whites.

Here hypothesis 1 predicts both of the pieces of evidence. As such, the evidence provides a degree of confirmation for the hypothesis. So far, so good. The problem, though, is that the hypothesis is underdetermined by the evidence. There are competing theories equally supported by the evidence, some of which seem just as likely in the first place as the initial conclusion. Here is an equally plausible hypothesis which could explain the evidence:

Evidence₁: A non-white male was prevented from entering through the exit at Costco.
Evidence₂: A white male was not prevented from entering through the same exit moments before.
Hypothesis₂: The Costco employee aims to stop all customers from entering though the exit, but did not observe the white male.

Again, the hypothesis predicts both pieces of evidence, and so both pieces of evidence provide a degree of confirmation for the hypothesis. Here’s another:

Evidence₁: A non-white male was prevented from entering through the exit at Costco.
Evidence₂: A white male was not prevented from entering through the same exit moments before.
Hypothesis₃: The Costco employee aims to stop all customers from entering though the exit, but the white male is an employee.

And so on. All three hypotheses equally predict and are equally supported by the available evidence. Nor does any one of these hypotheses seem obviously more likely in the first place than any other. A quick examination of the Starbucks incident reveals something similar: competing hypotheses are underdetermined by the available evidence.

Evidence₁: The manager at Starbucks refused to give Rashon Nelson and Donte Robinson the bathroom code.
Evidence₂: The manager at Starbucks asked Rashon Nelson and Donte Robinson to leave, and called the police when they did not.
Hypothesis₁: The manager at Starbucks did these things because she was biased against non-whites.

Now, for many, arriving at this hypothesis seems like a perfectly good bit of reasoning. Moreover, it’s certainly possible that the conclusion is correct. But how probable is the conclusion and are there equally plausible alternatives that might be reasonably drawn given the available evidence? As it happens, the evidence in the Starbucks case seems to be equally well predicted by at least one alternative hypothesis (no doubt there are others). Consider, for example, that the Starbucks outlet at the centre of the dispute had staff guidelines stating that non-paying customers must be asked to leave, and that the police should be called if they refuse. This additional information lends support to a different hypothesis:

Evidence₁: The manager at Starbucks refused to give Rashon Nelson and Donte Robinson the bathroom code.
Evidence₂: The manager at Starbucks asked Rashon Nelson and Donte Robinson to leave.
Hypothesis₂: The manager at Starbucks took these decisions because they are required by store policy and she enforces store policy regardless of race.

Evidence₁and Evidence₂also support Hypothesis₂. Given the available evidence, both theories are underdetermined. So what of parts two and three of Bayes’s theorem? How likely was the racism hypothesis prior to gaining the evidence, and how likely was the evidence to turn up even if the racism hypothesis doesn’t hold? If you take it for granted that racism is both widespread and of the sort that leads to discrimination in action, then, given those background beliefs, the likelihood that the incident was a product of racism will be quite high. If, on the other hand, you think that the sort of racism that results in active discrimination is not widespread, or if you think it likely that Starbucks managers generally and consistently enforce store policy towards non-paying customers, then, given those background beliefs, the likelihood that the incident was a product of racism will be low. The most obvious next step is to gather more evidence, as Galileo and other heliocentric advocates had to do. An obvious place to start is to investigate whether or not the manager inflexibly enforces store policy regardless of race. This could break the predictive symmetry and support one hypothesis over the other.

However, it is more likely that no-one will bother with this kind of humdrum investigative work, and the case will remain underdetermined. What then? We want to encourage two things: epistemic humility and epistemic grace.

Epistemic humility involves resisting the temptation to draw a conclusion when that conclusion is underdetermined by the evidence. That can be hard. There is discomfort in admitting that you don’t know. But doing so is a necessary first step in honest inquiry and authentic attempts to seek truths. We need to be cognisant of the background beliefs on which our judgements depend, and honest about when those background beliefs aren’t justified. Too often, pundits and news outlets are happy to proffer confident takes on events without doing anything like the groundwork needed to support their conclusions. While it’s unlikely that the chattering classes will cultivate epistemic humility any time soon, as consumers of information we can guard against inheriting unjustified beliefs from them by being conscious of when the claims are underdetermined by the available evidence.

Epistemic humility leads to epistemic grace. We should not be irrationally uncharitable. In the absence of evidence, we need to resist the temptation to interpret every negative interaction between a white person and non-white person as a case of racism. This unfortunate assumption is one that is all too common today; it breeds paranoia and further alienates an already divided culture. It’s also reinforced by the notion that taking racism seriously requires us to interpret events as racist whenever possible, even when the evidence doesn’t support that conclusion above others. Racist discrimination is real, and there are plenty of cases where the evidence points unambiguously to this. But when it’s genuinely underdetermined (or, worse, when the evidence points the other way) chalking someone’s behavior up to racism isn’t a mark of moral purity, it is just a failure to investigate the facts. Carefully seeking the truth and fighting racism aren’t at odds with one another. This should be a platitude. Unfortunately, in 2018 it is not. Similarly, when our beliefs are underdetermined, we, as a society and as individuals, should grant those with whom we have a disagreement a reasonable amount of grace. For one thing, both those who see racism in the Starbucks incident, and those who do not, can be at least conditionally rational. The depth, breadth and character of racism in American society is a hotly debated topic. Your interlocutor may be unduly hasty, or just plain wrong, but it would be rash to simply assume stupidity or disingenuousness on the part of those who read the situation differently.

The antidote to polarisation and mutual incomprehension is epistemic humility, epistemic grace, and open discussion. By employing these tools, we can dig down and uncover the presuppositions that drive others’ thinking, and our own. It’s only when these presuppositions are excavated that we have a chance of meaningful dialogue: the kind of meaningful dialogue that can actually get people to change their minds in light of evidence, work their way towards the truth, but accept when the facts lie out of reach. Maybe one day enough pieces of the puzzle will be in place for us to make a credible guess at the motivations of the Rittenhouse Square Starbucks manager. Until then, we should be prepared to be charitable in our judgements of others and find a way to live with the remaining uncertainty.