What to Do with Miracle Claims

Miracle claims are all the rage, but what should we actually learn from them?

"The Vision of Saint Hubert" - Peter Paul Rubens (c. 1617-1620)

---

This is a guest post written by an anonymous author.

---

Long-form YouTube videos detailing the miraculous evidence for theism are currently sparking heated debates online. These debates have mainly been fueled by the very public conversion of Joe Schmid, a Princeton University philosophy PhD student and prolific YouTuber. By his own telling, Joe’s conversion to Catholicism depended heavily on miracle claims, and Joe has devoted hours of video on his channel to documenting these claims and to defending theistic explanations of these claims against naturalistic alternatives. As I will argue in this post, Schmid and other apologists pay far too much attention to the details of miracle claims, and reasoning correctly about miracle claims (or really any kind of claim) involves a more holistic mode of evaluation.

When evaluating miracle claims, the central question is not whether the details of any particular miracle claim can be explained naturalistically. The miracle skeptic should fully expect that some miracle claims—even if naturalistically explicable in principle—will not be naturalistically explicable in practice. The central question for both miracle defenders and miracle skeptics is whether the distribution of miracle claims is more likely under one’s preferred metaphysical hypothesis (e.g., theism) than under relevant alternatives (e.g., naturalism). To the extent that the observed distribution is more likely under one’s preferred hypothesis than under relevant alternatives, the evidence of miracles counts in favor of that hypothesis. To the extent that the observed distribution is less likely under one’s preferred hypothesis than under relevant alternatives, the evidence of miracles counts against that hypothesis.

When evaluating the distribution of miracle claims, we might look to a number of factors, but to my mind, the most important things to consider are (i) the number of miracle claims and (ii) the degree to which the facts surrounding those claims resist naturalistic explanation (hereafter, “recalcitrance”).¹ Indeed, we can imagine a histogram representing this distribution of miracle claims, with bins representing levels of recalcitrance and bars representing the number of claims at each level. Each hypothesis under consideration will propose a prior distribution over the frequency of miracle claims and the parameters of the observed distribution of miracle claims (e.g., its mean, variance, skewness, kurtosis). We can then compare the predicted frequency of miracle claims to the actual frequency of miracle claims as well as the probability density of the observed statistics under each prior distribution. This will allow us (i) to compute the ratio of the likelihood of our observations under our preferred hypothesis compared to its alternatives (i.e., its Bayes factor) and (ii) to update our prior probabilities for each hypothesis.

This holistic reasoning is entirely consistent with seemingly contradictory reasoning about specific cases. For example, the theistic defender of miracles is entitled to hold that naturalism better predicts a particular miracle claim (e.g., because the claimed miracle seems to be a fraud).² Similarly, the naturalistic skeptic of miracles is entitled to hold that theism better predicts a particular miracle claim (e.g., because the claimed miracle seems to be a result of divine intervention). This might seem surprising, but it is no more surprising than a doctor admitting that one symptom is best predicted by a particular condition while holding that the overall pattern of symptoms is best predicted by another. Clearly, what matters is the pattern of evidence and not our judgment about any particular case in isolation.³

This simple observation knocks out what I take to be one of the two major argumentative strategies of theistic miracle defenders. This strategy is to present numerous miracle claims in the hope that the audience will (i) agree that at least one of the claims is best predicted by divine intervention and (ii) accept the apparent theistic implications. As we can now see, this strategy invites the audience to draw a theistic conclusion prematurely.⁴ This observation should also be a relief to the tireless naturalist who endeavors in every single case to defend the naturalistic hypothesis as more predictive than the theistic hypothesis. While this work is admirable, and I often agree that the naturalistic hypothesis is more predictive, neither party is obliged to maintain that none of the claims favor their opponent’s hypothesis when considered in isolation.⁵ As I will not tire of saying, it is the overall distribution of miracle claims that is relevant.

All that said, theistic miracle defenders have more sophisticated argumentative strategies available. For example, many of the more thoughtful (and statistically numerate) miracle defenders will argue that the evidence for miracles is part of a cumulative case. As they recount this evidence, they hope that their audience will be moved step-by-step to the view that the theistic hypothesis is more likely than not. This does require that one’s prior probability on the theistic hypothesis not be too low, and for this reason, theistic miracle defenders will package their miracle arguments with reasons for thinking that theism is at least plausible on its other merits. With this non-negligible prior in place, the theist trusts that Bayesian updating will carry the day eventually.

Unfortunately, this argumentative strategy is also misguided, and it rests on a specific misuse of Bayesian inference. A presentation of miracle claims (even a comprehensive or representative one) does not exhaust the relevant evidence. We must also consider all the cases where (i) a miracle might have been claimed but (ii) no miracle was actually claimed. Some of these will be more relevant than others. For example, theism plausibly places a higher likelihood on a miracle claim when a devout believer offers intercessory prayer than when no one prays at all. But all of these cases matter, because the evidential significance of miracle claims depends on the rate at which they occur, not their raw number. To put the point another way: we cannot weigh the significance of any particular number of miracle claims unless we have some sense of how many opportunities there were for such claims to be made. Without that denominator, we cannot say whether the number of claims at each level of recalcitrance exceeds what naturalism would predict.

What matters is (again) whether or not the overall distribution of miracle claims is more likely under one’s preferred metaphysical hypothesis. Are there more or fewer claims than we might expect? Are they more or less recalcitrant than we might expect? Are there more or fewer claims than we might expect at each level of recalcitrance? Answering these questions requires us to undertake a holistic evaluation of the pattern of miracles claimed (and not claimed) before being at all confident that the evidence of miracle claims favors our preferred view. As I will argue in the rest of this post, this kind of holistic evaluation is extremely challenging and unlikely to render a decisive answer either way.

To be clear, the challenge here is not that the cases under consideration are very numerous (though they are). When confronting an overwhelming body of data (e.g., all events where a miracle might have been claimed), it is perfectly reasonable to rely on a sample of that data. However, that sample must be unbiased for it to support any direct statistical inferences. We might forgive miracle defenders for presenting only miracle claims and not presenting a representative sample which includes cases where no miracle is claimed. We might even forgive miracle defenders for presenting only the most recalcitrant miracle claims. Nevertheless, unless we understand (i) how many candidate claims there were and (ii) how the claims presented were selected, we can’t learn much from them.

If we do know (i) and (ii), there are multiple ways we can learn from a biased sample. For example, we might expect upstanding politicians to have one or two plausible allegations of corruption. There are all kinds of ways for proper behavior to appear improper, and we can fully expect a politician’s rivals to find and publicize any apparent improprieties they can through oppo research. Does this mean we should never take these allegations seriously? Of course not. There will be some set of allegations (weighted by their number, severity, and evidential basis) which the filtering effect of oppo research cannot adequately explain.

In this case, it doesn’t matter that the rivals will only release negative information. We know that only a relatively small number of a politician’s actions might appear corrupt, and the sample provided by the rival places a lower bound on the number of apparently corrupt actions. At some point, there will simply be too many apparently corrupt actions released for the actual number of apparently corrupt actions to be an acceptably small subset of the politician’s past behavior.

The preceding example leaned heavily on our knowing the size of the population from which the biased sample was drawn. To see a case where the precise nature of the selection mechanism plays a more central role, suppose we are trying to estimate the ratio of black to white balls in a large urn using a small sample. Our assistant draws balls one at a time out of view and puts back every other white ball. If we know this, we can just double the number of white balls reported to recover an accurate sample.

So, do we know (i) how many candidate claims there were and (ii) how the claims were selected when evaluating a long apologetics presentation with many miracle claims? Not well, but we can roughly characterize them. With respect to (i), it seems likely that the number of cases in which a miracle claim might or might not have been made is staggeringly large, but even narrowing it down to the right order of magnitude seems challenging. We could restrict this considerably by asking how many cases of serious disease occur each year (in the case of medical miracles), or how many hosts are consecrated each year (in the case of Eucharistic miracles). Even in these restricted cases, the numbers are almost certainly extremely large, likely in the hundreds of millions each year.

With respect to (ii), it depends greatly on the source of the reported miracle claims. We might hear a report from a family member who claims to have experienced a miracle. We might hear one from an apologist specializing in miracle research. The selection effects here will be very different in each case. However, given the recent surge of interest in miracle claims presented by apologists, I will focus on those here. In this case, the selection effect seems extremely strong—if the apologists’ own reports are to be believed. Here is a quote from miracle researcher Ethan Muse:

“The vast majority of miracle claims are false. So therefore, the mere fact of a claim, without the additional substantiation and corroboration like we’re talking about here, the fact that you have a claim [by itself] is not good. Even sometimes testimony—lots of multiply attested testimony, even from people who were firsthand—you would not believe in researching the number of examples of that quality of testimony that I found that I can then show was spurious. It just turns out retrospective memory is unreliable, especially when people are giving testimony a lot later. There’s a certain fraction of people that will tell lies that will seem implausible or surprising to us. There’s a natural tendency for legendary embellishments and exaggerations to occur. There are misinterpretations that happen all the time. Sometimes there are coincidences. All these things do happen, and they happen with great frequency, which is why you need to be as forensic and analytical about verifying each case as we’ve been here and not just super credulous...”

It seems clear that Ethan is keen to show that he only reports on miracle claims he believes have passed his careful vetting and failed to admit mundane explanations (e.g., fraud, unreliable memory, exaggeration, etc.). Further, he is keen to show that the fraction surviving this vetting is very small. This constitutes a very strong selection bias in favor of recalcitrant miracle claims. This is perfectly reasonable if you believe that some miracle claims are genuine and you want to report as few spurious miracle claims as possible. In fact, the real selection effect here is probably stronger than Ethan suggests because he primarily investigates miracle claims that have risen to his attention—a process that requires individuals and the people around them to be sufficiently impressed by a claim that they amplify it.

While this kind of filtering may be admirable under certain assumptions, this is actually quite problematic for the miracle defender when their audience wants to evaluate miracle evidence as a whole. Given the prevalence of religious belief (along with the incentives Ethan cites above) and the many opportunities for miracle claims to be asserted (discussed above), we can safely conclude that the number of reported miracles is extremely large. This means that even if the naturalist predicts a small rate of inexplicable miracle claims, the number that exist will be significant. If even one in a million miracle claims is inexplicable, then there will almost certainly be quite a significant number of such claims. Further, given the filter applied by researchers like Ethan, we should expect their presentations to almost exclusively feature claims of this kind.

The problem for the miracle defender here is straightforward: This expectation should hold whether theism is true or naturalism is true. Both hypotheses predict with high probability that apologists will be able to fill long videos with impressive miracle claims. Charitably, the theistic hypothesis might place a slightly higher likelihood on this observation, but I suspect we are dealing with likelihoods very near 1 in both cases. Hence, apologetics videos of this kind should in practice convince no one.

While I have much more to say on this subject (and hope to say it in future posts), what I have said so far cuts both ways. If it is extremely hard to get a good sense of the distribution of miracle claims given the available evidence (i.e., heavily selected samples predicted similarly well by rival hypotheses), then the evidence of miracles simply does not favor either side. While this is equally true for both sides, it seems likely to disappoint miracle defenders far more than naturalists. The naturalist is generally playing defense and sees miracle claims as a challenge to be defused. Neutralizing miracle claims, as I have done here, will seem like a victory in this context. On the other hand, miracle defenders such as Ethan Muse and, more recently, Joe Schmid place extraordinary importance on miracle claims in support of their theological positions. Both routinely use miracle claims, for example, to justify the acceptance of positions that (at least in the case of Joe) they previously found deeply morally questionable.⁶

For example, Joe has said that the argument from evolutionary animal suffering is a very powerful version of the argument from evil and that he had grave concerns about Catholic teachings on gay sex and hell (understood as eternal conscious torment).⁷ Despite these objections, he subjected his intellect and will to Church teaching primarily on the basis of the strength of miracle claims presented to him by Ethan and others. As he states:

“I didn’t have a satisfactory response to all these problems until after I was just utterly flooded with overwhelming evidence for miracles.”

While he also offers direct responses to these objections, Joe is explicit that these responses only aim to blunt their force, and that this is acceptable since the miraculous evidence outweighs whatever force those objections retain. Joe also reports that his intuitions have begun to change following his conversion, but by his own admission, we might expect this as a result of cognitive penetration.

If the argument in this post succeeds, Joe must reassess whether the total evidence favors Catholicism. Perhaps he will discover arguments that do more than merely blunt the force of his objections and, instead, make Catholic teachings plausible. Perhaps he will not, and his objections will lead him to once again reject Catholicism. Naturally, this project will depend a great deal on the kind of philosophical reasoning that Joe now regards with suspicion, and I will leave it to him to wrestle with the implications of his meta-philosophical concerns. Fundamentally, my worry is that the argument from miracles has become a thought-stopping mechanism for Joe—one that shields against theological doubts and offers refuge in the face of philosophical uncertainty.

---

This is the first in a series of posts on arguments for theism based on miracles. In the next post, I will consider two major objections: (i) the possibility of a single miracle so well-evidenced that it alone supports belief in God and (ii) the possibility of a set of recalcitrant miracle claims so large that it overwhelms the problems introduced by selection. In the post after that, I will discuss the value of constructing naturalistic explanations of miracle claims and what those efforts might tell us about theism and naturalism.

1. I assume we are comparing generic versions of theism and naturalism. If, for example, one wanted to evaluate a particular version of theism—e.g., Christian theism or Catholic Christian theism—we would also need to consider the consistency of each miracle claim with relevant theological commitments.

2. Whenever I say “predicts a particular miracle claim” or something similar, I am using this as shorthand for “predicts the facts surrounding a miracle claim.” This includes, but is not limited to, the claim itself.

3. I need to reiterate that this kind of holistic evaluation is how we should always approach evidence—as the medical diagnosis case illustrates. We cannot evaluate a scientific hypothesis without looking at the pattern of evidence—pro, con, or neutral.

4. This isn’t to say a single case couldn’t be so compelling that it should convince us of theism. It is only to say that merely being better predicted by theism isn’t enough.

5. I want to emphasize how valuable I think this research is, and I hope to give reasons for its value in a future blog post. That said, if this argument motivates the tireless naturalist to have one more game night with their family, watch one more great film, or drink one more bottle of great wine with their friends, I wouldn’t object.

6. It is important to note that Joe and Ethan rely on more than just the recalcitrance of miracle claims. They also rely on the distinctively Catholic coding of what they regard as the most recalcitrant miracle claims with theological content. They argue that such miracles specifically vindicate Catholic Christian theism. This is an argument I may return to in a later post, but for now, I will simply note that the moderately surprising fact (if it is one) that the most recalcitrant miracles with theological content are almost all Catholic-coded doesn’t do nearly as much work against naturalism as the miracle claims themselves.

7. In fairness, Joe has described being more open to Catholic responses to these concerns now that he has returned to the Catholic Church, but he seems to regard an extended conversation with Ethan on miracles as the most significant tipping point.