Concept

Bayesian Probability

Intro

How do you change your mind in a sane way when new evidence comes in? That is what Bayesian probability is for. It is a math rule for updating belief honestly.

The rule comes from an 18th-century English minister named Thomas Bayes. The idea, stripped of math, is this. Start with how likely something is before you see new evidence (the prior). Ask how well that something would explain the new evidence compared to alternatives (the likelihood ratio). Multiply them. You end up with how likely it is after the evidence (the posterior).

Concrete example. You hear a noise outside. Could be the wind. Could be a burglar. Burglars are rare, so your prior probability of a burglar is low. But if the noise sounds like glass breaking and the wind is calm, the burglar hypothesis explains the evidence better than the wind hypothesis does. Your belief shifts toward burglar. You did not jump to certainty; you updated.

For apologetics this framework matters a lot. The strongest modern case for Christian theism is cumulative: each line of evidence (the universe began, the universe is fine-tuned, moral truths exist, the resurrection has historical evidence behind it, religious experience is widespread) is a separate update. Each one nudges the posterior probability of theism up a little. Stack them all and the cumulative effect can be very large, even if no single argument is a knockout. The same framework also defangs David Hume's old attack on miracles, because in Bayesian terms you do not get to stipulate that miracles are impossible. You have to weigh the actual evidence against the actual alternatives.

In full

Bayesian probability is the formal framework for updating belief in light of evidence. It treats probability as a measure of rational credence, degrees of belief held by an agent given background information, and provides a mathematically precise rule (Bayes' theorem) for revising those credences when new evidence arrives. Apologetically the framework is load-bearing: the most rigorous contemporary case for Christian theism is a cumulative Bayesian case in which independent lines of evidence, cosmological, teleological, moral, historical, experiential, each raise the posterior probability of theism above the posterior of naturalism. Bayes is also the framework in which the Humean attack on miracles is best answered (the prior against a miracle is calibrated against the posterior given the totality of evidence, not stipulated to one). Treating apologetics as Bayesian is itself a methodological claim: claims are evaluated by probabilistic warrant given all the evidence, not by deductive demonstration in isolation.

Bayes' theorem

The theorem (Thomas Bayes, posthumous publication 1763; rediscovered and generalized by Pierre-Simon Laplace, Théorie analytique des probabilités, 1812):

P(H | E) = P(E | H) × P(H) / P(E)

Where:

P(H | E), the posterior probability of hypothesis H given evidence E. What you want to compute: how likely is the hypothesis after taking the evidence into account?
P(H), the prior probability of H. How likely was the hypothesis before the evidence?
P(E | H), the likelihood of evidence E given hypothesis H. If H were true, how likely would we observe E?
P(E), the marginal probability of E across all hypotheses. Normalizes so posteriors sum to one.

The intuition: evidence raises the probability of a hypothesis to the extent that the hypothesis predicts the evidence better than the alternatives. P(E|H) > P(E|~H) means E confirms H; P(E|H) = P(E|~H) means E is neutral; P(E|H) < P(E|~H) means E disconfirms H. The ratio P(E|H) / P(E|~H) is the Bayes factor or likelihood ratio, the strength of the evidence.

A compact alternative form, useful for cumulative cases:

Posterior odds = Prior odds × Likelihood ratio

Iterating this across multiple independent (or near-independent) lines of evidence is the formal engine of the cumulative case.

Key thinkers

Thomas Bayes (1701-1761), Presbyterian minister and mathematician; "An Essay towards solving a Problem in the Doctrine of Chances" (posthumous, 1763) introduced the theorem in its original form. The fact that the founding figure was a clergyman writing in a context of natural theology is historically piquant but logically incidental.

Pierre-Simon Laplace (1749-1827), Théorie analytique des probabilités (1812); generalized Bayes' theorem and applied it to astronomy, demography, and inverse-probability problems. The "rule of succession" derives from his work.

Frank Ramsey (1903-1930), "Truth and Probability" (1926); developed the subjective interpretation of probability as rational degree of belief, defensible via Dutch-book arguments (an agent whose credences violate the probability axioms is exploitable by a clever bookie).

Bruno de Finetti (1906-1985), Theory of Probability (1970/74); independently developed subjective Bayesianism and the representation theorem connecting exchangeable random sequences to mixtures of i.i.d. distributions.

Richard Swinburne (b. 1934), Oxford philosopher of religion; the foundational contemporary Bayesian theist. The Existence of God (1979; 2nd ed. 2004) is the canonical treatment: each traditional theistic argument (cosmological, teleological, consciousness, providence, religious experience) treated as a probabilistic confirmation; the cumulative posterior of theism rendered "more probable than not" given the full evidence. Popular companion: Is There a God? (1996). The Bayesian apparatus is also deployed for Christology in The Resurrection of God Incarnate (2003).

Tim McGrew & Lydia McGrew, analytic philosophers of religion. "The Argument from Miracles: A Cumulative Case for the Resurrection" in The Blackwell Companion to Natural Theology (Craig & Moreland, eds., 2009), a detailed Bayesian analysis of the resurrection evidence yielding a posterior overwhelmingly favoring resurrection over naturalistic alternatives. Notable for treating individual minimal facts as quasi-independent evidence-streams and multiplying Bayes factors.

John Earman (b. 1942), Pittsburgh philosopher of science, not a theist but a sharp Bayesian critic of Hume's argument against miracles. Hume's Abject Failure: The Argument Against Miracles (2000) shows that Hume's argument either trivially proves too much (no testimony can establish any novel claim) or rests on confused probability reasoning. A standard live-cite for theists deploying the Bayesian resurrection argument.

Robin Collins, Messiah College philosopher; the leading Bayesian fine-tuning analyst. Argues that the cosmological constants are fine-tuned for life in a sense that yields a massive likelihood ratio favoring theism over naturalism-plus-chance, and that even the multiverse hypothesis (the leading naturalist alternative) faces fine-tuning problems one level up. Key essays: "The Teleological Argument" in Craig & Moreland, eds., Blackwell Companion to Natural Theology (2009).

Stephen Unwin, physicist; The Probability of God (2003), a popular Bayesian estimate that explicitly walks through Swinburne-style evidence-streams to a numerical posterior. Of pedagogical value more than scholarly weight; useful for showing skeptics that "Bayesian theism" can produce specific numbers and isn't hand-waving.

Elliott Sober (b. 1948), Wisconsin philosopher of science; frequentist critic of Bayesian apologetics. Argues that prior-dependence makes Bayesian theistic arguments rhetorically unsatisfying, what counts as the prior for theism is itself a contested question. The serious version of the "subjectivity of priors" objection. Worth steel-manning before deploying.

Alvin Plantinga, not a Bayesian by primary methodology (defends Reformed Epistemology), but has engaged the framework: Warranted Christian Belief (2000) accommodates Bayesian considerations while arguing that warrant for theism need not flow exclusively through evidential probability-update.

William Lane Craig, primarily deductive in his Kalam and resurrection arguments; uses Bayesian analysis selectively (especially in The Son Rises, 1981, and in his exchanges with Bart Ehrman). Treats Bayes as a tool within a larger natural-theology toolbox.

Apologetic applications

(a) Swinburne's cumulative case for God's existence

The architecture: take each traditional theistic argument as supplying evidence E_i with positive likelihood ratio favoring theism over naturalism. Multiply likelihood ratios across independent or near-independent evidence-streams. Apply to a non-zero prior. Compute posterior.

Swinburne's evidence-streams in The Existence of God (2004):

The existence of a contingent complex universe (cosmological).
The orderly law-governed character of the universe (teleological-of-laws).
Fine-tuning of constants (teleological-of-constants).
The existence of conscious beings (consciousness).
Moral consciousness.
Providence in history.
Reports of religious experience.
The evidence for specific miracles (especially the resurrection).

Each, Swinburne argues, raises P(theism) by a substantial likelihood ratio. The cumulative posterior is "more probable than not." Critics quibble individual likelihood assignments; theists respond that even on conservative numbers, the cumulative direction is unambiguously theistic. See Cumulative Case for Christian Theism.

(b) McGrew Bayesian-resurrection argument

Tim & Lydia McGrew's 2009 chapter computes a Bayes factor for the resurrection-hypothesis (R) vs the disjunction of naturalistic alternatives (~R), evaluated against the historical-minimal-facts evidence (E), empty tomb, post-mortem appearances to multiple groups including hostiles, disciples' transformation, Paul's conversion, James's conversion.

The structural claim: each minimal fact is highly improbable on ~R (a Bayes factor of, e.g., 100:1 against naturalism for the appearances alone), the facts are largely independent (no single naturalistic story dissolves all of them), and the multiplied Bayes factor swamps any reasonable prior against resurrection. The argument:

Posterior(R | E) / Posterior(~R | E) = [P(E|R) / P(E|~R)] × [Prior(R) / Prior(~R)]

If the multiplied likelihood ratio exceeds 10^14 (the McGrews' estimate on conservative-to-moderate numbers), then even a prior of 10^-10 against R yields a posterior overwhelmingly favoring R. See Argument from the Resurrection, Minimal Facts Argument, Argument from Miracles.

(c) Bayesian fine-tuning

Robin Collins's formulation: P(life-permitting constants | theism) >> P(life-permitting constants | naturalism + chance). The likelihood ratio is astronomical (estimated by Penrose at 10^123 for the initial low-entropy condition alone). The Bayes factor swamps any reasonable prior against theism. The leading naturalist response is the multiverse, but the multiverse itself either requires fine-tuned generator-physics (pushing the problem up one level) or is empirically untestable. See Fine-Tuning Argument, Anthropic Principle.

(d) Defeating Hume on miracles

David Hume (An Enquiry concerning Human Understanding §10, "Of Miracles", 1748): no testimony can establish a miracle, because the prior probability of a miracle is so low that any quantity of testimony must be more probably false than the miracle is true.

The Bayesian rebuttal (Earman, Hume's Abject Failure, 2000; the McGrews; Swinburne):

Hume miscalibrates the prior. The prior against a miracle is not "the frequency of miracles in nature" (which begs the question), it is P(miracle | God exists), which is not low if God exists and has reasons to act.
Hume ignores the likelihood ratio. Even with a low prior, evidence with a massive likelihood ratio can yield a high posterior. Hume's argument structurally rules out all extraordinary discoveries (continental drift, quantum mechanics), a reductio.
Multiple independent testimonies multiply. If three independent honest witnesses each have 90% reliability, the combined probability of agreement under falsehood is only 0.1%, Bayesian aggregation lets testimony overcome even very low priors.
Hume's "uniform experience" against miracles is question-begging. Whether experience is uniformly against miracles is itself part of what's at issue.

The rebuttal does not show that any particular miracle occurred, it shows that the Humean structural objection fails. The first-order evidence for specific miracles (the resurrection in particular) must then be assessed on its own. See Argument from Miracles.

Common objections

"Priors are subjective, Bayesianism is just rationalized prejudice."

Rebuttal: Bayesian theism is robust under wide prior variation. Swinburne, the McGrews, and Collins typically demonstrate that the posterior favors theism for any reasonable prior in a wide range (e.g., 0.001 to 0.5). The argument doesn't depend on picking a favorable prior; it depends on the evidence multiplying the prior strongly enough. The Dutch-book argument also constrains rational priors (priors that violate the probability axioms are exploitable).

The genuine version of this objection, Sober's frequentist critique, is that Bayesian priors for one-off hypotheses (theism, the resurrection) cannot be calibrated against frequencies, so the "subjectivity" cuts deeper than for ordinary scientific hypotheses. The theist's reply: this is a problem for any hypothesis about a unique event (the origin of the universe, the origin of life), not a special problem for theism.

Reference-class problem

To compute P(miracle | naturalism), what is the relevant reference class? "Past events"? "Past claimed miracles"? "Past historically-attested miracles"? Different reference classes yield different priors.

Rebuttal: The reference-class problem is real but does not undermine the qualitative direction of the cumulative case. Conservative reference-class choices still yield posteriors favoring theism given the magnitudes of likelihood ratios involved. The reference-class concern reduces certainty, not direction.

Updating paralysis

"With infinitely many hypotheses, infinitely many evidences, and infinitely complex dependencies, no real agent can actually compute Bayesian posteriors."

Rebuttal: The Bayesian framework is normative, not operational. We don't compute exact posteriors; we use Bayesian reasoning to evaluate the direction of evidential support and the qualitative magnitude of likelihood ratios. Engineers don't compute exact Maxwell-equations solutions either; they use the framework to reason about field behavior. The objection is a counsel of perfection, not a defeater.

"God of the gaps"

"You're invoking God to fill explanatory gaps where naturalism hasn't yet succeeded."

Rebuttal: Bayesian inference to the best explanation isn't "of the gaps"; it's of the data. The fine-tuning, cosmological-beginning, moral-fact, and resurrection-historical data are positive evidence with computable likelihood ratios, not gaps. The shrinking-gaps charge applies when theism only explains residuals; the Bayesian theist argues theism explains the whole pattern more cleanly than naturalism, not just the leftovers.

Why this matters for theism

The core methodological insight: multiple independent lines of evidence multiply, not add. A skeptic who concedes that each of the cosmological, teleological, moral, consciousness, and historical arguments is "interesting but not decisive" has not thereby defeated theism. If each argument yields a likelihood ratio of just 3:1 favoring theism (a very modest claim), and there are five quasi-independent such arguments, the multiplied Bayes factor is 243:1. Starting from a prior of even 1:100 against theism (a strong anti-theist prior), the posterior odds become 243:100, theism more probable than not.

This is why the apologetic strategy of treating arguments as a portfolio is mathematically grounded, not rhetorically convenient. Single-argument defeats don't dissolve a cumulative case. The cumulative case is mathematically robust against single-argument concessions in a way that deductive arguments are not. See Cumulative Case for Christian Theism for the full theological deployment.

A second insight: the bar for "extraordinary evidence" (Sagan's slogan: "extraordinary claims require extraordinary evidence") is rigorously a high likelihood ratio, not an impossibility-of-establishment. Sagan's slogan is true in spirit but is routinely deployed as if no actual evidence could ever clear the bar. Bayesian analysis shows that ordinary evidence multiplied across enough independent streams can clear arbitrarily high bars. The resurrection case is the showcase.

Cumulative Case for Christian Theism, the methodological master-hub.
Bayesian Argument for Theism, the explicit Swinburne-style cumulative-Bayesian deductive argument.
Minimal Facts Argument, the historical-evidential case that the McGrews Bayes-format.
Argument from the Resurrection, the deployed resurrection case.
Argument from Miracles, the framework-level miracle argument.
Fine-Tuning Argument, Bayesian teleology.
Argument from Mathematical Truth, mathematical realism as Bayesian-style evidence.
Anthropic Principle, the selection-effect counter-move that fails to dissolve fine-tuning.
Apologetic Method Comparison, Bayesian-evidentialism vs presuppositionalism vs classical-evidentialism.