Concept

Survivorship Bias

Intro

Survivorship bias is the mistake of looking at only the things that "made it" and ignoring all the things that didn't. The ones that didn't make it are invisible, so it is easy to draw wrong conclusions from the ones still in front of you.

The classic example is from World War II. The military wanted to add armor to bombers and noticed that returning planes had bullet holes mostly in the wings and tail. The first instinct was to armor those spots. The statistician Abraham Wald pointed out the obvious-once-you-see-it problem: the planes hit in the engine never came back. So the armor needed to go where the surviving planes had no holes. The data was telling you the opposite of what it looked like.

In apologetic conversation this shows up in both directions. A common atheist deployment: Christians count the answered prayers and forget the unanswered ones, so prayer "looks" effective when really we are just hearing from the survivors. A common Christian counter: atheists cite scientific theories that worked out and quietly forget the long list of naturalist predictions that failed, so naturalism "looks" successful by the same bias.

The fix is the same in either direction: ask "out of how many?" before drawing a conclusion. If nobody can give you the denominator, the conclusion is shakier than it looks.

In full

The systematic cognitive error of drawing inferences from a sample composed only of those who "survived" some selection process while ignoring the (often much larger) sample that did not survive and is therefore invisible. The second cognitive-bias entry in the Fallacies folder, sister to Confirmation Bias, distinct from confirmation bias because the distortion is not about how the perceiver weights evidence at hand but about which evidence ever reaches the perceiver in the first place. The non-survivors leave no trace; the surviving sample looks like the whole sample because it is the only sample available to inspect.

The canonical illustration is Abraham Wald's bomber-armor analysis (Statistical Research Group, Columbia, c. 1943). The U.S. military observed returning WWII bombers and tabulated where the bullet holes clustered (wings + fuselage + tail), then proposed armoring those areas. Wald inverted the analysis: the returning bombers showed where damage is survivable; the armor should go where the returning planes had no holes (the engines + cockpit), because the bombers hit there were the ones that did not return + therefore did not contribute their bullet-hole pattern to the dataset. Wald's insight is the load-bearing template for diagnosing survivorship bias in any domain: always ask what the missing-because-not-observed sample looks like + what selection process produced the gap.

Survivorship bias has at least four distinguishable manifestations that recur in apologetic discourse:

Outcome-survivorship, counting only successful outcomes (answered prayers, miraculous recoveries, surviving religions) without counting unsuccessful outcomes (unanswered prayers, deaths, extinct movements).
Source-survivorship, historical sources reaching us are by definition those preserved; non-preserved testimony cannot be checked for support / contradiction.
Witness-survivorship, only conversion-testimonies of those who stayed are typically published; deconversion testimonies are systematically less catalogued in tradition-internal literature.
Movement-survivorship, the religions / philosophies / institutions still extant are over-represented in cumulative-case arguments; the failed ones from the same evolutionary cohort are absent.

The bias is major-severity because (a) it is structurally invisible to the unaided observer (the missing data leave no positive trace), (b) its corrective requires explicit denominator-construction discipline (often impossible after the fact), and (c) it operates on both sides of the apologetic conversation, atheist deployments against miracles + Christian deployments in cumulative-case apologetics are both vulnerable to the diagnosis.

Canonical structure

Survivorship bias is not a fallacy in the strict syllogistic sense, like Confirmation Bias, it is an evidence-availability bias rather than an inferential-form error. Its general structure:

Stage 1: A population P contains members M₁, M₂,... Mₙ
Stage 2: A selection process σ (often invisible to the observer) admits only subset S ⊂ P into the observable dataset
Stage 3: Observer infers from S as if S = P (or as if S is a representative random sample from P)
Stage 4: Inference systematically over-represents whatever traits drove σ-survival

The classical formulation (Wald): if the question is "where do bullets damage planes?" + the dataset is "returning planes," the dataset's silence about the engine-region is evidence about engine-vulnerability, not evidence of engine-safety.

The error: at no stage is a formal logical fallacy committed. The premises are accurate descriptions of S; the conclusion follows logically from premises about S. The fallacy is upstream of inference, at the level of dataset construction.

This makes survivorship bias harder to detect than formal fallacies (no inferential marker) AND systematically under-corrected even after detection (full denominator data is often unavailable + must be estimated or imputed).

Diagnostic markers

Survivorship bias is operating when:

No denominator stated. Argument cites N positive instances ("N people were healed at Lourdes"; "N answered prayers documented") without specifying the population from which N was drawn ("of M total petitioners"; "of M reported prayer requests"). Detection test: ask "out of how many?", inability to answer is the marker.
Selection process not analyzed. Argument treats the available sample as random when in fact a non-random selection process generated it. Detection test: ask "what selection process produced this dataset, and what does that selection process exclude?", vague or absent answers indicate the bias is operating.
Asymmetric publication availability. Successes are systematically published; failures are systematically not. Detection test: where are the "I prayed and nothing happened" memoirs? Where are the "I went to Lourdes and was not healed" testimonies? If they are systematically less available than success-cases, the published-evidence dataset is survivorship-biased.
Anachronistic projection from extant institutions. Argument from "Christianity has lasted 2,000 years" without comparison to extinct contemporaneous movements (Mithraism, Manichaeism, Marcionism, Gnosticism, Bogomilism, Cathari, Shaker movement) that also claimed transcendent grounding but did not survive. Detection test: who are the comparators that might have made similar claims but didn't survive?
Conversion-narrative without deconversion-comparator. Citing N conversion-testimonies as evidence for X without acknowledging the parallel set of deconversion-testimonies from people who once held X with comparable conviction. Detection test: are deconversion-testimonies treated with the same weight?

Atheist deployment against Christian apologetics

The survivorship-bias charge against Christianity has multiple deployment patterns:

"Christians count answered prayers + ignore unanswered ones." The general claim: religious belief survives because of selection-bias on prayer-outcomes, the recovered cancer patient prays-and-is-healed → "miracle"; the dead cancer patient prayed-and-died → "God's mysterious will" (or, more commonly, simply not counted). Examples: Sam Harris Letter to a Christian Nation (2006), the chapter on petitionary prayer + statistical-failure literature (the STEP Prayer Study (Benson 2006) meta-analysis showing no effect when survivorship-controlled). Strong form: "if prayer worked at the rate Christians implicitly claim, amputees would be healed; they aren't, ever, in any documented case (the 'why won't God heal amputees?' meme)." The amputee-restriction is the survivorship-bias diagnostic at high resolution: the population of failed prayers is so total in this category that no survivorship-survival case can be cited.
"Lourdes / saint-cause cures are cherry-picked from millions of pilgrims." The general claim: Lourdes has received approximately 200 million pilgrims since 1858; the Lourdes Medical Bureau has approved approximately 70 cases (the recent Vannini case being one of the most rigorously vetted: see Marie Bailly (Lourdes 1902) + the 2026-05-09 Vannini canonization-miracle entry). Therefore: 70 / 200,000,000 = 0.000035% rate, which is lower than the spontaneous-remission rate for many cancers. The "miraculous" cures are statistical survivors from a vast sample of failures. Substantive variant: Hume Of Miracles (1748) anticipates this in its argument that the standing weight of regularity-of-nature evidence dwarfs the testimonial-evidence for any single miracle.
"Religious survival doesn't validate the religion." The general claim: many religions and philosophical movements have survived for centuries (Hinduism, Buddhism, Shinto, Daoism, Zoroastrianism). Christianity's survival is unremarkable in that population. The companion charge: many movements did not survive (Mithraism, Manichaeism, Catharism, Shaker celibacy-movement); their failure-to-survive is uninformative about Christianity's truth-claims. Survivorship bias makes the surviving-religions look more meaningful than they are.
"Conversion testimonies are published; deconversion testimonies are suppressed." The general claim: Christian publishing houses (Zondervan, Baker, Crossway, IVP, Eerdmans, Cokesbury) systematically publish conversion-narratives; comparable major presses do not exist for deconversion-narratives, so the available-published-narrative dataset is survivorship-biased toward conversion. Examples: Lee Strobel's conversion-from-atheism narrative is a Christian best-seller; Bart Ehrman's parallel deconversion-from-evangelicalism is published by a secular press (HarperOne) and reaches a different audience. The asymmetric publishing infrastructure shapes what counts as evidence.
"Apologetic literature cites the Christianity-survives-empires move without counting the empires it joined-with-and-then-fell-with." The Roman Empire survived and Christianity grew within it; Byzantine Empire fell and lost its Christian core; the Holy Roman Empire fell and many of its principalities secularized; Spanish + Portuguese Catholic empires fell and the colonized regions are now mixed. Survivorship-bias-on-which-imperial-Christian-trajectory-is-cited shapes the apologetic narrative.

Christian deployment against atheist arguments

The charge runs symmetrically, survivorship bias also operates in atheist apologetic discourse:

"Naturalistic prediction-success is over-reported; prediction-failure is under-reported." The mirror charge: science textbooks emphasize successful naturalistic predictions (general relativity confirmed; DNA structure resolved; periodic table organized) while quietly retiring failed predictions (the perpetual stable-state universe, panspermia + spontaneous-generation hypothesis, junk-DNA dominance, Lamarckian inheritance, Marxist historical-determinism, behaviorist explanations of language). The published-history of science is survivorship-biased. Compare: Christianity's resurrection prediction has held since Easter; many naturalist predictions have been retracted within decades.
"Origin-of-life confidence relies on Earth-as-survivor selection bias." The mirror charge: when atheists argue "abiogenesis happened once on Earth, so it's possible naturally," they are inferring from the only known case where it happened. Earth is the data-point that survived (because we are here observing it); failed-abiogenesis-on-other-worlds would leave us without observers and thus no observation. The single survivor-case provides no genuine probability information about how likely abiogenesis is. James Tour, Hubert Yockey, and Robert Shapiro have made structurally-similar critiques of the standard popularization.
"Anthropic principle is survivorship-bias formalized." When fine-tuning evidence is countered by "we observe a fine-tuned universe because we couldn't observe an un-fine-tuned one," this is the explicit application of survivorship bias to the cosmology question. Christians can respond: yes, the anthropic-principle move is a survivorship-bias appeal, but applied symmetrically, it equally legitimates religious-experience-as-survivor-of-faith-tradition reasoning. The principle cannot be deployed against fine-tuning without simultaneously legitimating parallel theistic moves.
"Atheist regimes' survival is selectively cited." The mirror selective-survival charge: 20th-century atheist regimes (USSR survived 70 years, PRC still extant, North Korea still extant, Cuba still extant) are sometimes cited as proving atheism's organizing-power. But Albanian-style hyperatheism collapsed; East-German communism collapsed; Khmer Rouge collapsed within years; Afghan Marxist regime collapsed; the Soviet bloc collapsed entirely. Survivorship-bias in which atheist regime gets cited shapes the historical apologetic in either direction.
"Atheist conversion-narratives are over-cited; reversion-narratives are under-acknowledged." Mirror to charge #4 above: Antony Flew's conversion-from-atheism-to-deism (see Antony Flew (Deism Conversion 2004)) is a heavy-hitter Christian apologetic citation; less-cited are atheists who briefly considered theism and rejected it on substantive engagement (Richard Carrier's atheist post-Christianity; A.J. Ayer's deathbed-experience-but-no-conversion; Bertrand Russell's lifelong atheism after extensive Christian-tradition engagement). Symmetric availability of both directions is required for honest apologetic engagement.

Three load-bearing rebuttals

When the survivorship-bias charge is deployed against Christian apologetics, three rebuttals do most of the work:

The Lourdes Bureau structurally controls for survivorship in its own methodology. The Bureau's vetting protocol (since 1883, modernized in 1947 + 2008) is designed to address survivorship bias: (a) it requires medical-team documentation of the initial diagnosis + prognosis (so the denominator of comparable cases-without-cure is implicit in medical baseline); (b) it requires pre-recovery documentation that distinguishes spontaneous remission frequency from genuinely-anomalous recovery; (c) it requires long-term follow-up (typically 10-30 years) to rule out delayed-relapse + spontaneous-remission masquerading as cure; (d) it requires international medical committee (CMIL, Comité Médical International de Lourdes) review of the medical record, not just the testimonial; (e) the rejection rate is >99% of submitted cases (Bureau receives roughly 7,000 reported recoveries per year from pilgrim petitioners; approves ~0-1 as scientifically inexplicable). The structural rejection-rate-asymmetry is the Bureau's explicit survivorship-bias control. The Marie Bailly case (1902, witnessed by Nobel-laureate Alexis Carrel under explicit expectation that she would die in transit) and the Vannini case (1948 cure of Vincenza Vannini, ratified 2026 for Carlo Acutis canonization after 78 years of follow-up) are the kind of cases that survive a denominator-controlled vetting. The objector cannot dismiss these via generic survivorship-bias appeal; they have to engage the specific case. (See Lourdes Medical Bureau + the Tier-1 miracle entries.)
Christianity's central truth-claim is non-survivorship structured. Christianity's foundational empirical claim is the resurrection of Jesus of Nazareth, a single, datable, falsifiable historical event that, on Christian theology, must have happened or Christianity is false (1 Corinthians 15:14). This is the inverse of a survivorship-bias structure: the claim does not depend on aggregating over a sample of cases (where survivorship could selectively-include); it depends on one specific event whose occurrence + witness-attestation + non-disconfirmation (no body produced in 1st-century Jerusalem) are independently checkable. The minimal-facts methodology (Habermas, Licona, see Resurrection) explicitly draws on data accepted by the majority of NT historians including hostile/skeptical scholars, selecting on consensus across opposed positions rather than on survivor-friendly Christian sources. The methodology is structured against survivorship bias.
The "answered prayer" survivorship-bias charge confuses two distinct theological claims. Christianity does NOT predict that prayer will achieve any-petitioner's-stated-temporal-outcome with high success rate; that strawman is the only target the survivorship-bias charge engages. The actual Christian theological claim is: (a) God hears all prayers (1 John 5:14-15), (b) God answers according to His will + greater-good consideration (Mark 14:36, "not what I will but what You will"; Romans 8:28), (c) God uses suffering instrumentally for sanctification + greater purposes (James 1:2-4, Romans 5:3-5), (d) the cross itself is the paradigm where the Son's "let this cup pass" prayer was not answered in the petitioner's stated outcome (Mark 14:36, "yet not what I will, but what You will" + Hebrews 5:7-9). Christianity's theology of prayer is built around the reality that not all prayed-for outcomes are received. Hebrews 11:13 + 11:39-40 explicitly names faithful witnesses who did not receive the promised outcome in their lifetime. The survivorship-bias charge presupposes a theology Christianity rejects, the prosperity-gospel / health-and-wealth distortion, and conflates that with classical Christianity. Engaging the actual theology dissolves the charge.

False-fallacy examples (cases where named-but-misapplied)

Cases where "survivorship bias" is charged but does not apply:

Lourdes Medical Bureau ratified cases. The Bureau's >99% rejection rate + multi-decade follow-up + international medical-committee review is structurally a survivorship-bias control. Charging "but it's still survivorship bias" without engaging the methodology is a misapplication of the diagnosis. The objector who wants to maintain the charge needs to specify which step in the Bureau's protocol fails to address the bias.
Habermas-Licona minimal-facts resurrection methodology. Selecting on historians' consensus including hostile + skeptical scholars is structurally a survivorship-bias control: the dataset is constructed from a population not friendly to the resurrection claim, then used to argue for the claim. Charging "you only cite scholars who agree with you" is precisely wrong, the methodology cites scholars who disagree with the conclusion but accept the underlying historical facts.
Old Testament Messianic prophecy fulfillment. Specific predictive prophecies (Bethlehem birth: Mic 5:2 → Mt 2:1; Davidic descent: 2 Sam 7:12-16 → Mt 1:1; suffering-servant death: Isa 53 → Mark 15; resurrection: Ps 16:10 → Acts 2:25-32; ~50+ specifics surveyed in Messianic Prophecies literature) constitute non-survivorship-controlled predictions: they were written centuries before the candidate fulfillment + are independently datable (Dead Sea Scrolls evidence on Isaiah; LXX evidence on multiple texts) + were either fulfilled in the Christ-event or they weren't. Survivorship bias does not explain the fulfillment-pattern because the predictions themselves did not select for any survival-process; they were locked-in pre-fulfillment.
Conversion testimony from a person whose prior commitments were strongly atheist + who maintained engagement with both sides extensively before converting. When C.S. Lewis documents his progression from atheism through theism to Christianity (in Surprised by Joy); when Antony Flew (Deism Conversion 2004) documents his shift via specific engagement with cosmological + DNA / abiogenesis evidence (in There Is a God, 2007); when Lee Strobel documents his investigative-journalist methodology (in Lee Strobel (Conversion 1981)), these conversions are not survivorship-bias-selected. The convert was already in the opposed sample; their cross-camp testimony reflects substantive engagement, not selection effect.
The structural unique-claim that Christianity makes against ANE comparators. Survivorship bias would expect surviving religions to look broadly similar (same selection pressures). But Christianity's structural shape (suffering Messiah; Trinity; Incarnation; salvation by grace through faith; the church as Christ's body) is anomalous against the ANE / Greco-Roman religious landscape it emerged from + the Second-Temple Jewish landscape it grew out of (high Christology emerging within decades; cross-as-victory inverting Roman political theology; cf. Bauckham + Hurtado on early high Christology). The survivorship-bias prediction (Christianity-looks-like-other-survivors) is disconfirmed by the structural-anomaly evidence.
Cumulative-case apologetic explicitly drawing on multiple independent evidential lines. The Cumulative Case for Christian Theism structure is built on convergence of independent lines (cosmological + teleological + moral + religious-experiential + historical-resurrection + transformative-fruits). Each independent line carries its own evidential weight; the convergence is not survivorship-biased because the lines are not selected from a pool of failed-other-evidence-types. Charging survivorship bias against the cumulative case requires showing that comparable cumulative cases for opposing positions exist + are equally well-attested + are systematically suppressed, a strong claim the objector typically does not defend.
Failed-prayer documentation in Christian pastoral + theological literature. Christian tradition extensively catalogues unanswered prayer + suffering of the faithful: the entire genre of biblical lament-psalms (Pss 6, 13, 22, 38, 42-43, 51, 73, 88, 102, 130, 142, 143; Lamentations as a complete book); the Job narrative; Jeremiah's confessions; Paul's thorn (2 Cor 12:7-10); the long Christian tradition of the dark night of the soul (John of the Cross; Mother Teresa's Come Be My Light documenting decades of felt-divine-absence). The literature on unanswered prayer + felt-divine-absence is not less-published than the answered-prayer literature; it is one of the most sustained themes of Christian devotional + theological writing across two millennia. The asymmetric-publishing charge fails empirically.
Documented falsifiers Christianity exposes itself to. Christianity's exposure to the resurrection-falsifier (1 Cor 15:14), Messianic-prophecy-falsifier, fruits-falsifier (Matt 7:16) is not a survivorship-biased dataset because exposure to falsifiers is the opposite of selection-on-survival. Christianity can be (and historically has been) tested against these conditions; it survives them by passing the tests, not by selection-bias.

When survivorship bias IS operating

Cases where the diagnosis genuinely applies (which Christians should acknowledge + correct):

Generic miracle-anecdote citation without denominator. When a Christian apologetic cites "I knew someone who was healed when prayed for" without acknowledging the much larger sample of comparable cases that did not recover, the citation is survivorship-biased. The corrective is denominator-construction (the medical-team baseline; the Bureau-style protocol).
Argumentum-ad-Christendom-survival without comparator analysis. When a Christian argues from the persistence of Christian institutions (universities, hospitals, charities, denominations) without comparing to extinct contemporaneous movements, the argument is survivorship-biased. The corrective is to acknowledge the comparator population + identify what makes Christianity's survival pattern qualitatively + quantitatively distinct.
Conversion-testimony citation without deconversion-testimony engagement. When a Christian apologist cites N Christian conversions without engaging the parallel set of well-documented deconversions (Bart Ehrman, John Loftus, Dan Barker, Marlene Winell), the citation is survivorship-biased. The corrective is symmetric engagement: honest reading of the deconversion literature + addressing the substantive issues raised.
Citing successful prayer-for-healing without acknowledging the much larger sample of devout-but-unhealed. The pastoral problem of the cancer-stricken faithful Christian whose prayer was apparently not answered is a real case the apologist should engage rather than evade. Survivorship-biased prayer-apologetic damages credibility.
The "prosperity gospel" health-and-wealth distortion is a near-pure survivorship-bias artifact. Selectively cataloguing wealthy + healthy + successful Christians while inattending to faithful Christians who are poor + sick + suffering. Christianity's own theology of suffering (Romans 8:17-18; Phil 1:29; 1 Pet 4:12-19) explicitly rejects the prosperity-gospel framing.
Selective citation of "Christianity built Western civilization" without engaging the negative episodes. Christianity historically built universities + hospitals + abolition movements + literacy + scientific institutions; Christians historically also produced the Crusades + Inquisitions + colonial atrocities + slavery defenses + religious wars. Honest historical apologetic engages both sides; survivorship-biased apologetic cites only the positive contributions. (See Christians Behaving Badly for the complete engagement.)
Apologetic that cites only successfully-fulfilled Old Testament Messianic prophecies without acknowledging the population of unfulfilled prophecies + delayed-fulfillment expectations. Survivorship-biased prophecy apologetics. The corrective: distinguish first-coming-fulfilled from second-coming-pending, address apparent-failure cases (Tyre; the timing of Daniel's seventy weeks; Ezekiel's dating of post-exilic restoration), engage the harder cases honestly rather than citing only the easier ones.

The corrective in each case is denominator-construction discipline + symmetric engagement of the failure-cases + substantive theological framing of suffering + non-fulfillment as part of the Christian truth-claim, not denial.

Christian theology of survivorship

Christianity has a substantive theological framework that pre-empts the survivorship-bias charge as a worldview-level critique:

Hebrews 11, the faithful who did NOT receive the promises in their lifetime. "All these died in faith, without receiving the promises, but having seen them and welcomed them from a distance" (Heb 11:13); "And all these, having gained approval through their faith, did not receive what was promised, because God had provided something better for us" (Heb 11:39-40). Hebrews 11 is structurally an anti-survivorship-bias chapter: it commemorates faithful witnesses across the dimension of unrealized promise, the very kind of data survivorship bias would suppress. Christianity's theology of faith includes this category as canonically central, not as embarrassment.
The Beatitudes name suffering as blessed. Matthew 5:3-12 systematically pronounces blessing on the poor in spirit + those who mourn + the meek + those who hunger and thirst for righteousness + the persecuted. The Beatitudes are an explicit valuation framework that does NOT track survival-success; they track conformity to a different good. Survivorship-bias diagnostic does not apply when the position being evaluated explicitly rejects survival-success as the measure.
The lament-psalm tradition preserves prayers that did NOT receive their requested resolution. Psalms 6, 13, 22, 38, 42-43, 51, 73, 88, 102, 130, 142, 143 are lament-psalms; Psalm 88 ends without resolution ("darkness is my closest friend", v. 18 NIV); the imprecatory psalms (see Imprecatory Psalms Objection) preserve prayers for retribution that were largely not realized in the petitioner's lifetime. The Psalter is structurally inclusive of the non-survivor data.
The cross itself is the anti-survivorship paradigm. God's chosen Messiah-Son did NOT escape death; he was tortured, executed, buried. The resurrection inverts the survivorship-bias trajectory: God works through the failure-of-survival, not by selecting survivors over non-survivors. Christianity's central event is not "faithful Jesus survived persecution"; it is "faithful Jesus was killed by persecution + raised on the third day." The selection logic is foreign to Christianity's actual theological claim.
Romans 8:36, sheep for the slaughter. "For Your sake we are being put to death all day long; we were considered as sheep to be slaughtered" (Rom 8:36 quoting Ps 44:22). Paul explicitly names suffering-unto-death as the apostolic-Christian existential condition; not as anomaly but as expectation.
2 Cor 4:7-12, treasure in jars of clay. Paul's apostolic theology explicitly thematizes vulnerability, persecution, perplexity, being-struck-down, and frames these as the means by which Christ's life is manifested, not the obstacle to it. The framework values non-survivor data.
Mother Teresa's Come Be My Light + the dark-night-of-the-soul tradition. John of the Cross's Dark Night of the Soul; Mother Teresa's documented decades of felt-divine-absence; Pascal's Pensées on the hiddenness of God; Job's unanswered theodicy questions. Christian tradition has systematically catalogued the felt-failure-of-divine-presence, not suppressed it. The publishing record of these works (Mother Teresa's letters were withheld by her at her own request + published posthumously after extensive theological vetting; the dark-night literature is in the Christian classics canon) is the empirical refutation of the asymmetric-publishing charge.

The Christian apologist who takes survivorship bias seriously thereby aligns with biblical lament-tradition + Reformation theology of the cross (Luther's theologia crucis) + classical Christian asceticism, the bias is acknowledged + named + addressed methodologically, and the underlying theological framework is one that already values + preserves the non-survivor data.

Apologetic deployment notes

When Christian engages survivorship-bias charge:

Acknowledge the bias is real + name where it applies. Don't deny the diagnosis applies to generic miracle-anecdote citation, prosperity-gospel framing, or cherry-picked church history. Owning the diagnosis where it applies builds credibility for distinguishing where it doesn't.
Cite the Lourdes Medical Bureau methodology. When the charge is deployed against miracle claims generically, route to the Bureau as the institutional case where survivorship-bias is structurally controlled. The 99% rejection rate + multi-decade follow-up + CMIL international medical-committee review is the kind of denominator-construction the diagnosis demands.
Cite the minimal-facts resurrection methodology. When the charge is deployed against the resurrection apologetic, Habermas-Licona's selection-on-hostile-and-skeptical-consensus is the structural corrective. The methodology selects evidence against the apologetic conclusion; survivorship bias cannot account for the resulting case.
Cite Hebrews 11 + the lament-psalm tradition. When the charge is deployed against Christianity's prayer-theology, route to the explicit canonical inclusion of the unrealized-promise + felt-divine-absence categories. Christianity's theology is the anti-survivorship framework; the charge presupposes a strawman (prosperity gospel) Christianity rejects.
Demand symmetric application. Survivorship bias also operates in atheist origin-of-life apologetic (Earth-as-survivor-case), in selective citation of naturalist prediction-success, in the anthropic-principle deployment. If the charge is to be principled, it must be applied symmetrically. If the opponent will not, the principled-application failure is itself diagnostic.
Specify the alleged missing-data + how the opponent knows it's missing. Generic charges of survivorship bias are weaker than specific charges with proposed-missing-data. Asking "what specifically would you expect to see in the unobserved sample, and what's your basis for the expectation?" sharpens the conversation.
For the "amputee" deployment specifically: engage the substantive theology. The amputee-restriction is rhetorically powerful but theologically uninstructive: Christianity's prayer-theology does not predict 100%-success-rate for any given category of petition. The amputee challenge is the strawman-prosperity-gospel theology applied selectively to a category where the strawman is most visible. Engage the actual theology (intercessory prayer + God's discretion + the suffering-Messiah paradigm + Hebrews 11's unrealized-promise category) directly.

When Christian charges atheist with survivorship bias:

Be specific not blanket. "Citing the multiverse to absorb fine-tuning evidence requires a denominator of universes that you cannot observe + cannot estimate empirically; that's structural survivorship bias" is sharper than "you have survivorship bias."
Acknowledge the symmetry where it applies. "Christians do this too in answered-prayer apologetic; the question is whether either side is willing to construct denominators honestly", symmetric framing keeps the conversation honest + preempts the Tu Quoque deflection.

Scholarly resources

Abraham Wald "A Method of Estimating Plane Vulnerability Based on Damage of Survivors" Statistical Research Group memoirs (1943); reprinted in Marc Mangel + Francisco Samaniego "Abraham Wald's Work on Aircraft Survivability" Journal of the American Statistical Association 79 (1984): 259-267, the founding case that defined survivorship-bias diagnosis in operations research.
Daniel Kahneman Thinking, Fast and Slow (Farrar, Straus and Giroux, 2011), System 1 / System 2 framework; chapter on availability + selection biases includes survivorship-bias as a primary illustration.
Leonard Mlodinow The Drunkard's Walk: How Randomness Rules Our Lives (Pantheon, 2008), popular treatment of survivorship-bias + selection effects in financial markets, sports, science.
Nassim Nicholas Taleb Fooled by Randomness (Random House, 2001) + The Black Swan (Random House, 2007), sustained treatments of survivorship bias in financial-market analysis + the silent-evidence problem.
Paul Copan + William Lane Craig Contending with Christianity's Critics (B&H Academic, 2009), engages atheist survivorship-bias-style critiques of miracle apologetics.
Craig Keener Miracles: The Credibility of the New Testament Accounts 2 vols. (Baker Academic, 2011), comprehensive engagement with miracle-claims literature; explicit denominator-construction discipline; the most sustained Christian response to Hume-style survivorship-bias arguments.
Gary Habermas The Risen Jesus and Future Hope (Rowman & Littlefield, 2003), minimal-facts methodology for resurrection apologetic; structurally addresses survivorship-bias by selecting on hostile + skeptical scholarly consensus.
Michael Licona The Resurrection of Jesus: A New Historiographical Approach (IVP Academic, 2010), methodology of historical Jesus research with explicit attention to selection effects + criteria.
Régis Burnet Lourdes: Histoire et théologie d'un site (CNRS, 2018), French Catholic-academic treatment of Lourdes including the Bureau methodology.
Bernard François + collaborators, Lourdes Medical Bureau (Bureau Médical de Lourdes) annual reports + the CMIL (Comité Médical International de Lourdes) review protocols, available via sanctuary publications.
Antony Flew There Is a God: How the World's Most Notorious Atheist Changed His Mind (HarperOne, 2007), conversion-against-prior-commitment case that structurally inverts survivorship-bias prediction.
Lee Strobel The Case for Christ (Zondervan, 1998), investigative-journalist methodology for engaging the resurrection apologetic; explicit attention to evidential standards + denominator construction.