On an especially lazy day in Florida, I peer up at the partly cloudy sky. Most of the visible sky shines blue with only the occasional patch of clouds. Like most people who observe the cloud formations, I eventually see different shapes that resemble real-life objects. The monochromatic shapes metamorphosize from one object to the next. They aren’t the actual objects, of course. The clouds are merely collections of humidity suspended in the atmosphere. Their resemblance to any real object is merely a coincidence.
However, our minds fool us. We desperately impose order in shapes that occur strictly randomly. If you should see a woman’s face or an AT-AT Walker in a cloud formation, it merely reflects what currently occupies your mind. The objects that we see have no correlation to real life. If anything, what you perceive reflects your mind, not the image that you gaze upon.
Our natural instinct imposes order on random things. We cleverly come up with mnemonics to remember a class of information. Does anyone else remember the phrase, “Every Good Boy/Bird Does Fine/Fly.”?
Imposing order on randomness
Psychologists developed a test with random black and white inkblots; these were known as Rorschach tests. Even being a psychology minor, I never took a Rorschach test. In fact, I know of no one who has taken one. While it’s certainly true that what we see reflects what we think, I don’t believe any of these tests were conducted in a way to produce useful results. Where we able to find the next Ted Bundy from one of these tests? I doubt it.
However, when we gaze upon and ponder either a cloud formation or inkblot, we implicitly know that it represents nothing. Any conclusions that we reach merely reflect our own ‘order’. If you should see a woman’s face and I see an AT-AT Walker, neither of us is implicitly right nor wrong. We simply differ, and that’s okay.
Unfortunately, we dismiss other conclusions as biased based on our life experiences. We generalize that our perceptions are indeed the norm, and other conclusions that deviate from ours are wrong. Occasionally, we even rationalize through counterexamples. We believe that our order is the order.
Dealing with outliers
As we cross into adulthood, we realize that rarely is anything either always true or always false. In order to simplify our lives, we ignore certain key details in order to simplify the problems. When I first studied Physics in college, we ignored certain rules. We joked about point masses, frictionless surfaces, and perfectly inelastic collisions. However, we still solved each problem, and each such answer lied closely enough. In theory, an object in motion stays in motion. However, if you were to put your car in neutral and coast, it will inevitably come to a stop… so much for frictionless surfaces.
Unfortunately, we make similar generalizations about human nature. We yearn for a certain order to our world and want to perceive the world according to those rules. Even when confronted with concrete counterexamples, we rationalize this particular instance is the exception to the rule. Furthermore, given a large enough population, there’ll always be an exception.
Mentally, we’ll concede this particular example doesn’t fit our mental model, so we’ll dismiss it as insignificant. Statistically, there’s a term for this. Such a data point is called an outlier. The purpose of a good model lies in its ability to predict. While any model will have its share of outliers, if there are enough outliers, by definition, it’s not a good model.
‘A few bad apples’
Suppose that a man drives down the road, five miles under the legal speed limit, and a police officer stops him. After a series of questions, the police officer issues a citation for precisely that, driving five miles under the legal speed limit. And he advises the driver that he can simply toss that citation in the trash. The police officer mentions that his driving under the speed limit is suspicious. Did I mention that the driver happens to be black?
We’ll often hear the term ‘a few bad apples’ to describe the exception to those rules. We generally want to believe that the police don’t disproportionately stop black motorists without reasonable suspicion. Naturally, this isn’t always true, there are ‘a few bad apples’. They are the outliers; they do not fit the model. This particular example is an outlier; there is no such thing as systemic racism. Again, the purpose of a good model lies in its ability to predict.
However, we need to collectively ponder about what is the threshold of ‘a few bad apples’. At what point do these ‘bad apple’ outliers become statistically significant enough that we need to reevaluate the nature of our model. What if we were to take an objective look at the numbers? Say that we take statistics from over 100 million traffic stops and find that, per capita, black motorists are 10% more likely to be pulled over than white drivers.
Well, research found that they’re not stopped 10% more than white drivers. Per capita, black drivers are 20% more likely to be stopped than white drivers.
Change the model
Let’s reflect upon that police officer in that last section, would he have perceived the driver as suspicious if he were doing the exact same thing, but he was white? No? I won’t go as far as asserting that police officer concluded the driver behaved suspiciously because he was black. I don’t believe that it was anything that nefarious. However, if you treat two people differently based strictly on the color of their skin, isn’t that the very definition of racism? What if we find that the best model for predicting traffic stops includes the driver’s race as well as their behavior? Maybe systemic racism still exists?
Isn’t it time we start to look at the data empirically? Orthogonally, I’ve been on enough threads on Twitter where people assert without foundation that transgender people are groomers; that their intent is to sexually molest children. Though as it happens, someone has been tabulating who has made the news for sex crimes involving children. Interestingly, people from religious employment are one-hundred twenty-five times (500 versus 4) more likely than a transgender person to commit a sex crime against a child.
Does that not fit your mental model? No? Maybe your model needs to change.