A Miracle Is Not An Explanation

And Neither Is “Supernatural”

I had some positive conversations with miracle believers in my previous posts that culminated into some salient disagreements. I want to elaborate on them here. Though it sounds like a slogan, I sincerely believe that calling something a miracle or supernatural is not an explanation. Here’s why:

The False Super/Natural Dichotomy

I don’t like the supernatural/natural dichotomy because I have no idea how a supernatural cause (or effect) differs from a natural one. I’m agnostic on the supernatural, but I’m inclined to disbelieve in the supernatural because every argument I see in favor of it is pretty bad.

I don’t have any bias against the supernatural in principle. I could imagine an example of observing and being confident of something “supernatural” through a Humean methodology. The problem with doing this is that such an event would cease to be considered mysterious or supernatural because of the observed regularity.

A Supernatural Humean Healing:

There is a prophet of a God who has immense healing powers. He can heal any ailment because (he says) of his connection to the almighty. Whenever he is asked to heal, or merely chooses to heal, he does. This pattern is so consistent that scientists can design experiments to test his healing powers. They have the best tools in the world that can perceive down to the quantum level. When they run those experiments, all they see are lights that descend on the injured or sick tissue. When they go deeper, they observe the light descending on the cells, the molecules, the atoms, and so on. These pieces of matter act in a manner that can only be called “healing.” The lights are then observed and studied, and when they are, it’s established that they’re just lights. The prophet does this healing on command, and at all times of observation, the result is the same.

In that situation, I’ll gladly say that this prophet has healing powers. Where they come from or how they work, I don’t know, but I can be confident that he has them. These powers, no matter their origin, could be considered “supernatural” in the sense that they defy current explanation. At the same time, they would possess a quality that makes them less “supernatural” compared to other things we consider supernatural. Namely, they could be tested, act in a uniform way, and lead to regular, predictable outcomes.

Contrast that with our actual experience of the purported supernatural. The results are always mixed, inconsistent, and never replicate a consistent result. That may sound like an overstatement, but if you look into the efforts to demonstrate paranormal things like ghosts, demons, psychics and so on, it’s really not. Whenever anything resembling a peer reviewed study is produced, the effects aren’t statistically significant and the results fail to replicate.

For this reason, skeptics reject the supernatural; it isn’t out of circular logic or an a priori commitment against supernatural origin, but because such phenomena fail to be tested or studied.

We need a solid body of evidence to draw reliable inferences or formulation probabilities about anything, and don’t have that for purported supernatural causes. And so, when skeptics say they don’t have enough evidence for God or the supernatural or what have you, this is what they mean: The evidence we have is of such a poor quality that the best we can do is suspend judgment.

Hypothesis Testing In Philosophy of Religion Is A Farce

By my estimate, the best arguments for miraculous and supernatural explanations are due to the tools of process of elimination and induction to the best explanation. That is, when we exhaust naturalistic explanations, the supernatural ones win.

This is obviously bad reasoning. I’m talking about “introductory to philosophy” level bad reasoning. I am aware that this sounds mean and polemical, and I don’t mean to insult people who disagree with me, many of which have PhDs. But hear me out.

This bad reasoning derives from the misapplication of hypothesis testing, which has grown in popular philosophy of religion discourse like cancer (okay that one was definitely polemical, but as I’m criticizing a method and not people, I will stand by it).

I’m not going to go into the full explanation of hypothesis testing here because there are others who do that better elsewhere. For our purposes, what you need to know is that the application of hypothesis testing in philosophy of religion is mistaken because philosophy is not a science. Philosophy does not run experiments or collect data.

For science, hypothesis testing relies heavily on practices of testing theories and replication. Scientists pondering whether a hypothesis fits the data they observe are trying to retrodict that data (that is, predict it backwards) in an effort to refine their theories to better make future predictions. In this way, it’s not enough that a scientific theory fits existing data, it must predict future data.

And so, philosophy of religion that relies heavily on hypothesis testing is psuedo-scientific. It doesn’t uncover new facts about the universe or make predictions about the future, or at least predictions that we can test. Instead, it merely tries to fit all already-observed data to sound compelling. In statistics, that’s called overfitting, and its costs are high. Among them, overfit theories are bad at reliably predicting future data.

Hypothesis testing methodologies exist in fields outside of science, like criminal investigations, but those too are disanalogous to philosophy of religion.

Criminal investigations involve events that we have abundant experience with. Not only do we have abundant data of people committing crimes in various ways, but evidence in criminal investigations often depends on patterns of human behavior and physical evidence that we are quite familiar with. Unlike philosophy of religion’s speculative causal relationships, criminal investigations deal with highly probable, strongly evidenced causal relationships.

Polycausality is Good, Actually

Almost a decade ago, I had to take a research methods class in graduate school. It was the most fun class I took in grad school because it enhanced my personal philosophy.

Among the many things I learned was that what’s called monocausality (that is, one thing causing an entire phenomenon) is extremely rare, especially in social science. Often we are dealing with multiple causal factors that together compound and account for various outcomes.

Let’s say we’re trying to predict someone’s political beliefs based on a variety of variables we have access to, from age to religion and so on. Through long term rigorous analysis, we can draw correlations between certain variables to the point where they are predictive. So, for example, an older, white, male evangelical Christian living in Alabama is very likely to vote for a Republican. Not all evangelicals who fit that demographic profile will vote Republican, but it’s a probable inference because each of those variables are independently predictive.

And so, we would say that there is polycausality at work on this person’s voting behavior, as there are multiple causal factors that increase the likelihood and probability of them voting republican. As helpful as this message is, those causal factors will likely not accumulate to being 100% predictive. Our theory needs a margin of error, or else we run the risk of overfitting.

More interesting is how we account for outlier cases, which all miracles are. By definition, our best scientific theories won’t predict a miracle with 100% specificity, for the same reason our theories can’t predict all outliers with specificity.

Indeed, there’s probably at least one older, white, male evangelical Christian in Alabama who is a social, well off what our modeling says he should be. Our best theories would say it’s unlikely that this person is real, but also that it’s not impossible. To the extent we can understand how this outlier case came to be, we’re going to have to find details specific to that outlier case. Such methods will always come off as speculative and improbable (ad hoc as people say), but so is the existence of outlier cases in the first place.

Supernaturalists often throw jabs at non-supernaturalists, saying their explanations are ad hoc or amount to conspiracy theories. But this misunderstands the probability calculi of these situations. If you have a tremendously improbable outcome that isn’t impossible, the best explanations will appear extremely unlikely because our most generalizable theories can’t accommodate all of the data perfectly while retaining predictability.

For example, if I were to tell you that a high school football team scrimmaged the University of Alabama and won, the explanation of “because they were better” is very simple but also not very helpful. That outcome itself is extremely improbable to the point of near impossibility, and so the component causal variables will be improbable. Those improbable variables could be that they played the scrimmage at all, that it was a scrimmage between first year walk-on players for Bama against multiple five star prospects on the high school team (say it was IMG Academy). It could be that the scrimmage went into 8 overtimes. And so on.

It would be nice if we could explain every crazy, unlikely thing that happens with a simple theory, but as we’ve seen, that would sacrifice predictability. We should thus expect seemingly ad hoc, improbable explanations for improbable outcomes, at least sometimes. Supernaturalists who tell you otherwise are not practicing good statistical thinking.

It’s Not Enough To Just Eliminate Rival Hypotheses

Not only are supernaturalists deceptive with their hypothesis testing rhetoric and dismissal of polycausality, they also employ tactics no one would think reasonable in any other context. That is, they believe that it’s enough to prove their hypothesis by merely eliminating rival hypotheses.

This is silly because in all other contexts, when you’re trying to push a theory of causality of something, it’s not enough to eliminate a rival hypothesis, you have to construct your own. It’s not enough to say “Your theories don’t account for all of the data and therefore I win by default.” No!

If we’re trying to determine who took a cookie from the cookie jar, the elimination of Jack as a suspect may increase the likelihood of Jill being the culprit, but it doesn’t mean Jill is guilty by default. If you’re going to prove she’s guilty, you must have a causal explanation supported by data.

And this returns to where metaphors of criminal investigation fail to be analogous to other causal explanations. That is, the typical suspects in a crime are a set number of people that investigators can eliminate, and almost deductively prove who committed a crime. If Jack and Jill’s single mother comes home to their rural house with no neighbors to find the cookie jar empty, and Jack can be confidently eliminated as a suspect, Jill is very likely guilty.

But when we’re talking about mysterious polycausal events, our candidate explanations are almost never so narrow. Many of those explanations may be outside of our perception or comprehension. And so, unlike criminal investigations or investigations with only a handful of likely hypotheses that can be seemingly deductively eliminated, mysterious events will not have that same level of certainty around candidate hypotheses. Sometimes we’ll have no good candidate hypotheses. In those cases, suspension of belief is warranted.

Three More Problems With Miraculous or Supernatural Explanations

Up to this point, I’ve been critical of miraculous and supernatural explanations because of methodological mistakes people make in coming to them (bad hypothesis testing, misunderstanding the importance of polycausality and outliers, and “win by default” logic). In this section, I want to examine the flaws in these explanations that are specific to the explanations.

First, these explanations fail to meet sufficient explanatory standards. If I were to say “Matt punched Joe in the face” that sentence embeds all sorts of explanatory content. We could continue to rewrite that sentence, usually into multiple sentences, expanding the explanatory content to be more detailed. We could say “Matt balled up his fist, took a step forward, lifted his arm, balled his fist, and made contact with Joe’s face.” In theory, we could break down each component part of this sentence into component causal parts, down to the molecular or quantum level. Obviously, we don’t do this because the sentence “Matt punched Joe in the face” captures all the relevant information we want to know.

This is much different from miracle claims. When we say “Ryan miraculously healed Judy” there is no additional causal explanation, no way of expanding on that sentence. This is very weird and unique to supernatural or miraculous claims; the appearance is just taken at a given, with no further investigation.

I doubt we would accept this low standard of explanation for other important, consequential questions. Why make an exception for miracles?

Second, we don’t have enough data on supernatural causes to be confident about their causal nature. Remember how I said, for the Humean, there’s really no difference between supernatural and natural events and causes, other than that supernatural events and causes seem to lack consistent evidence? Well, if we have such poor evidence for these events, in that we can’t be certain which causes are supernatural or not, we can’t really be confident in any inference or probability we form about the nature of supernatural events.

Keep in mind the story of the healing prophet: We can be confident that his powers are real, but we can’t be confident of their causal origin. Some may immediately jump to supernatural origins, but what evidence do we have of supernatural origins to feel confident in that conclusion? It could just as well be advanced science or the work of non-Godly beings. Supernaturalists don’t like this reluctance to accept divine origins, but the point here is not that I would believe a specific alternative spooky explanation, but that I can’t confidently accept an explanation that doesn’t have sufficient evidence.

Third, what data we do have seemingly disconfirms their causal power. When people try to create a science or probability of the supernatural, it destroys them as explanations. Take the healing waters of Lourdes for instance. In about 150 years of purported visits, the Catholic church considers about 7,000 people miraculously healed. That sounds like a lot of people, until you realize at least 100,000 people come to Lourdes seeking healing every year (but possibly more). If the causal theory is that Lourdes’ waters or devotion at Lourdes causes people to be healed, why is it that so few people are actually healed?

Let’s assume that the 100,000 number is accurate going back a century. That means that 10,000,000 people visited Lourdes looking for healing, and only 7,000 were healed. That’s 7 in 10,000 people were healed, or .07% healing rate. How can we say that the waters of Lourdes heals people when the number is that low?

For such high visitation, you could consider those healings random statistical noise. If we did a clinical trial of a drug and it only healed .07% more people than a control group, we would not consider that effect statistically significant. We most certainly would not say that it treats the illness. Why make an exception for Lourdes?

Final Thoughts

Supernatural and miraculous explanations should be rejected because they are the result of bad methodology and a form of special pleading.

The people who propose these explanations advocate for fallacious reasoning, including pseudo-scientific hypothesis testing, rejecting polycausality, and “win by default” logic. Supernatural and miraculous explanations themselves fail to meet an adequate standard of explanation, are insufficiently supported relative to naturalistic explanations, and the data that is collected on them disconfirms their causality.

We would not accept such meager explanatory standards in any other field of explanation.This is not because of the purported spiritual causal origins of supernatural explanations, but because of their poor quality and reasoning.

Until someone can demonstrate to me that these objections are wrongheaded, I will continue to say that calling something a miracle or supernatural is not an explanation.