Concept

Confirmation Bias

Intro

Confirmation bias is the habit of looking for evidence that supports what you already believe and ignoring evidence that does not. It works in four ways: you go searching in places likely to confirm you, you read mixed evidence in your favor, you remember the hits and forget the misses, and you hold opposing evidence to a much higher standard than your own.

Peter Wason proved it in a famous 1960 experiment. He gave people the number sequence 2, 4, 6 and asked them to discover the rule. The actual rule was simply "any three numbers in ascending order." Most people guessed something like "even numbers going up by two" and then tested only triples that fit their guess (4, 6, 8 or 10, 20, 30). Very few tried triples like 1, 2, 100 that would have shown their first guess was wrong. The bias is the default setting of the human mind, not a glitch.

In religious and apologetic conversations, the charge gets thrown both ways. Atheists accuse Christians of remembering answered prayers and forgetting unanswered ones. Christians accuse atheists of only reading Dawkins and Hitchens while ignoring Plantinga and Hart. Both charges can land. Both can also be unfair.

The cure is not denial but discipline. Read the best arguments on the other side, not the worst. Ask what evidence would change your mind, and answer honestly. When you cite a source, apply the same credibility check to opposing sources of equal credentials. Christianity itself names this discipline: "renew your mind" (Romans 12:2); "be quick to hear, slow to speak" (James 1:19); "do not lean on your own understanding" (Proverbs 3:5).

Quick deployment note: when someone charges you with confirmation bias, do not deny it exists in you. Acknowledge the universal pull, then ask for specifics. "Which best-version argument do you think I haven't engaged?" Convert a blanket accusation into an actual conversation.

In full

The systematic cognitive tendency to seek, interpret, remember, and weight evidence in ways that confirm one's prior beliefs while ignoring, discounting, or forgetting evidence that disconfirms them. The first cognitive-bias entry in the Fallacies folder, distinct from formal logical fallacies because it operates at the level of psychological processing (how minds actually handle evidence) rather than at the level of inferential structure (whether the argument's logical form is valid). Both are reasoning errors that distort discourse, and apologetic engagement on contested questions (Resurrection, Problem of Evil, Moral Argument, Cosmological Argument) makes confirmation bias unavoidable as a topic, both because opponents will charge it AND because it genuinely affects everyone, including Christian apologists.

The bias was systematically demonstrated by Peter Wason in his 1960 "2-4-6 task" experiment at University College London: Wason gave subjects the number triple 2-4-6 and asked them to discover the underlying rule (which was simply "any ascending sequence"); subjects were allowed to test triples and would be told whether each fits the rule. Subjects characteristically proposed triples consistent with their initial hypothesis ("3-6-12" testing "doubling") rather than triples that would falsify it ("3-7-15" testing "doubling"). Most failed to discover the actual rule because they failed to seek disconfirming evidence, establishing confirmation bias as the default cognitive mode rather than a deviation requiring explanation.

Confirmation bias has at least four distinguishable mechanisms, all of which appear in apologetic discourse:

Selective search, preferentially looking for evidence that supports a held belief (Christian reading only Christian apologetic books; atheist reading only atheist critique books).
Selective interpretation, reading ambiguous evidence in support of held belief (a "miraculous" recovery interpreted as divine intervention OR as natural recovery, depending on prior belief).
Selective memory, recalling supporting evidence more readily than disconfirming evidence (remembering the answered prayers, forgetting the unanswered ones, or the inverse).
Asymmetric scrutiny, demanding higher epistemic standards from disconfirming evidence than from confirming evidence (Christian accepting Lourdes-Bureau-ratified miracle on basis that satisfies the skeptic's standard for the inverse claim; atheist demanding peer-reviewed double-blind RCT for any miracle claim while accepting hostile testimony for atrocity attributions).

The bias is major-severity rather than critical because the rational corrective (deliberate exposure to opposing best-arguments + asymmetric-scrutiny audit + falsifiability commitment) is well-developed in epistemology + philosophy of science + Christian apologetic methodology. The bias's existence does not undermine the possibility of genuine knowledge or genuine apologetic argument; it specifies the discipline required to reach knowledge through human cognitive architecture.

Canonical structure

Confirmation bias is not a fallacy in the strict syllogistic sense, it does not have premises and a conclusion. It is a cognitive-processing bias that affects which evidence reaches awareness + how that evidence is weighed + which conclusions feel epistemically warranted.

The general structure:

Stage 1: Subject S holds prior belief B (consciously or unconsciously)
Stage 2: S encounters mixed evidence E (some supporting B + some opposing B)
Stage 3: S systematically attends to / interprets favorably / remembers / scrutinizes-less-rigorously the supporting evidence; systematically inattends to / interprets unfavorably / forgets / scrutinizes-more-rigorously the opposing evidence
Stage 4: S concludes B is well-supported (when in fact the underlying-evidence balance was mixed)

The error: at no stage does S commit a formal logical fallacy. The premises (insofar as they exist) accurately reflect the evidence S consciously processed. The fallacy is upstream of the inference, at the level of which evidence reached the inferential process.

This makes confirmation bias harder to detect than formal fallacies (no syntactic marker) AND harder to correct than formal fallacies (the corrective is methodological discipline rather than pattern-recognition).

Diagnostic markers

Confirmation bias is operating when:

Asymmetric standards. Subject demands extraordinary evidence for opposing claims while accepting modest evidence for supporting claims. Detection test: ask the subject what evidence would change their mind on the held position; if no answer or answer that's effectively unobtainable, asymmetric standards are likely operating.
Citation asymmetry. Subject cites only sources within their tradition / political camp / methodological school for the held position; refuses to engage best opposing arguments. Detection test: can subject articulate the steel-manned best version of opposing position? If no, confirmation bias is likely operating in their evidence-search stage.
Memory selectivity in narratives. Subject recalls supporting anecdotes vividly + readily but draws blank or vague responses on disconfirming anecdotes. Detection test: ask for examples of when held-belief made a counterintuitive prediction that turned out false; struggling to recall any is a marker.
"Already-decided" rhetorical posture. Subject's argument feels foregone-conclusion-shaped, every fact arranges itself into the same pattern, no fact ever falls outside the pattern. Detection test: would any conceivable evidence be granted as disconfirming? If no, the position is unfalsifiable and confirmation bias is in operation (compare Special Pleading).
Source-of-evidence-evaluated-by-conclusion. Subject evaluates a source's reliability based on whether its conclusions match the held position. Detection test: when source S₁ supports held position, S₁ is "reliable, peer-reviewed, well-credentialed"; when source S₂ (with comparable credentials) opposes held position, S₂ is "biased, motivated, fringe." Substantively-symmetric source-evaluation patterns indicate confirmation bias.

Atheist deployment against Christian apologetics

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

"Christians only count answered prayers + forget unanswered ones." The general claim: religious belief survives because of selective memory of supportive cases (answered prayer) + selective forgetting of disconfirming cases (unanswered prayer / cancer death of devout believer). Sometimes formalized as the "Texas sharpshooter" variant: drawing the bullseye around the bullet holes after firing. Examples in literature: Sam Harris The End of Faith + Christopher Hitchens God is Not Great + Richard Dawkins The God Delusion. Substantive variant: Bart Ehrman God's Problem memoir of moving from evangelical to agnostic largely through unanswered-prayer + theodicy weight not previously attended to.
"Believers interpret ambiguous experience as religious." The general claim: religious experience is generic mystical/transcendent experience interpreted through prior theological framework (Christians interpret as Jesus; Hindus interpret as Krishna; Muslims interpret as Muhammad). The bias supposedly explains religious diversity itself: same psychological substrate + different prior beliefs → different "experiences." Examples: William James Varieties of Religious Experience (descriptively) + Daniel Dennett Breaking the Spell (skeptically).
"Apologists only read pro-Christian sources." The general claim: Christian apologists are systematically under-exposed to best atheist arguments (Russell, Hume, Mackie, Sobel, Smith, Oppy, Sinnott-Armstrong); Christian apologetic literature is incestuous; therefore Christian conclusions reflect echo-chamber selection rather than genuine evidence-weighing.
"Scripture-interpretation is confirmation-bias laundered." The general claim: when scripture appears to disconfirm Christian doctrine, it's reinterpreted (allegorized; recontextualized; harmonized; declared cultural-not-binding); when it appears to confirm, it's taken at face value. Hard cases (Calvinism vs Arminianism on the same texts; six-day vs old-earth creationism on Genesis 1; pacifism vs just-war on Sermon on the Mount) supposedly reveal that scripture can be made to mean whatever the interpreter wants.
"Christianity is unfalsifiable in practice." Following from the above: Christianity supposedly has post-hoc explanation for any conceivable disconfirming evidence (suffering = soul-making; unanswered prayer = greater good; divine hiddenness = preserving free response; historical failures = humans not Christianity; etc.), making the position effectively immune to disconfirmation.

Christian deployment against atheist arguments

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

"New Atheists only read New Atheist sources." The mirror charge: atheist apologists are systematically under-exposed to best Christian arguments (Aquinas, Anselm, Pascal, Newman, Plantinga, Swinburne, Craig, Hart, McGrath); atheist literature often engages strawmanned versions of theology; therefore atheist conclusions reflect echo-chamber selection. Empirical specificity: Hart's The Experience of God documents that Dawkins / Hitchens / Harris / Stenger systematically attack a strawmanned god-as-largest-being-among-beings rather than the classical-theist God-as-being-itself, confirmation bias at the level of which theological position even gets engaged.
"Atheists count atrocities by religious people but not atrocities by atheist regimes." The mirror selective-memory charge: 20th-century atheist regimes (USSR, Maoist China, Pol Pot's Cambodia, Albania) collectively account for 100+ million deaths; this counts in the historical ledger if one is doing comparative statistical reckoning of belief-system harm. The frequent atheist response ("those weren't really atheist motivations / Stalin had a seminary background / Mao's ideology was secular religion") parallels the No True Scotsman Fallacy move + selective inattention to disconfirming evidence.
"Selective-skepticism about miracles vs selective-credulity about naturalist explanations." Atheists who refuse to engage Marie Bailly (Lourdes 1902) + Carrel-Nobel-laureate-witness documentation while accepting much weaker natural-historical evidence in non-religious contexts exhibit asymmetric scrutiny. Same operative mechanism as Christians who refuse to engage best skeptic arguments while accepting weaker pro-Christian apologetic claims.
"Origin-of-life confidence runs ahead of evidence." Atheist apologetic frequent-claim that "abiogenesis is just chemistry, give it time" reflects evidential confidence that exceeds the actual biochemistry-research consensus (which acknowledges abiogenesis remains an open question with no empirically-demonstrated complete pathway). Confirmation bias on the held metaphysical-naturalist commitment selects for over-confident phrasing in popularization.
"Multiverse confidence as confirmation-bias-for-fine-tuning." When fine-tuning evidence makes design-inference rationally probable, the multiverse is invoked with high confidence despite multiverse being unobservable + lacking direct empirical support. The asymmetric-confidence pattern (multiverse: high-confidence-because-it-saves-naturalism vs design: high-skepticism-because-it-doesn't) is confirmation bias on the metaphysical-naturalist prior.

The "coincidence" and "confirmation bias" defeater cluster

When miracles or answered prayer are dismissed as "just coincidence" combined with "confirmation bias," four additional rebuttal moves are available beyond the general rebuttals below:

"Coincidence" is a verbal stop-sign, not an explanation. The word concedes the pattern is real and worth flagging, then refuses to account for it. Improbability does not vanish under relabeling. Biblical miracles are often providentially timed natural events, the Jordan stopping at Adam (Josh 3:15-16), the east wind parting the sea (Exod 14:21), quail arriving (Num 11:31), Elijah's drought breaking on cue (1 Kgs 18), the coin in the fish (Matt 17:27). The biblical category of miracle frequently is "coincidence" in form, God working through natural-means timing. Calling it "just a coincidence" describes how the miraculous looks from the inside while claiming to have explained it away.
Selective epistemics. The skeptic accepts "improbable convergence demands explanation" everywhere else, fine-tuning → multiverse, abiogenesis → deep time, consciousness → emergence, but treats answered prayer or fulfilled-prophecy clusters as "coincidence" requiring no cause. The asymmetry: explanation is demanded everywhere except where the cause might be personal. This is Special Pleading.
Confirmation bias is self-undermining. "Religious people see meaningful patterns where there are none" is itself a noticed pattern about a category of people. If pattern-detection is systematically unreliable enough to dissolve answered prayer, it dissolves the skeptic's own generalization about religious pattern-detection. The acid eats its own beaker. This parallels the Argument from Reason / Plantinga's EAAN, naturalism undercutting its own epistemic reliability.
Prospective prophecy defeats retrospective cherry-picking. Daniel 9's 70-weeks window closing in the early 1st century AD, Psalm 22's crucifixion details (~1000 BC, before crucifixion existed as a method, invented by Persians ~500 BC, adopted by Rome ~200 BC), the Cyrus prophecy in Isaiah 44-45 naming the king ~150 years out. Pre-registered predictions cannot be dismissed as retrospective cherry-picking, there is only one shot, recorded in advance. Document dating can be disputed, but not via "you only counted the hits."
The canon actively curates negative data. The Psalter's largest genre is lament. Job, Lamentations, Ecclesiastes, Habakkuk, Jeremiah are canonical, wall-to-wall unanswered prayer, divine silence, and protest. Mother Teresa published her decades of spiritual darkness; C.S. Lewis published A Grief Observed. A movement that preserves its negative data in its foundational documents is the structural opposite of confirmation bias.

Three load-bearing rebuttals

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

Universal applicability cuts both ways. Confirmation bias is a feature of human cognitive architecture; it operates in everyone, on every contested question, regardless of position. If the charge is meant to discredit Christian belief specifically, it must show that confirmation bias operates asymmetrically more in Christians than in atheists, and the empirical literature does not support this asymmetric claim. Studies of motivated reasoning find the bias broadly distributed across political + religious + ideological commitments. The generic charge thus cuts both ways: if it discredits Christians for selective citation, it equally discredits atheists who only read New Atheist sources. Either both positions are epistemically suspect (in which case we need methodological correctives, not blanket dismissal of one side) or neither is.
Christianity has substantive built-in falsifiers + corrective mechanisms. Christianity is not unfalsifiable in principle:

Resurrection-falsifier: "If Christ is not raised, your faith is futile" (1 Corinthians 15:14, 17). Paul explicitly grants the empirical-historical falsification condition. If first-century Jerusalem produced a body, Christianity is false.
Old-Testament-Messianic-prophecy falsifier: if the prophesied Messiah did not come from David's line, was not born in Bethlehem, did not suffer the prophesied death, did not undergo the prophesied resurrection, these were historically-falsifiable conditions that did NOT obtain when checked.
Practical-fruits falsifier: "By their fruits you will know them" (Matthew 7:16). Christianity's truth is partially testable by the fruits of Christ-conformed lives (charitable institutions, sustained communities of love-of-enemy, transformed lives across cultures). The 2,000-year continuity and global expansion of these fruits is evidence; their absence in any sustained way would have been disconfirmation.
Methodological corrective: the Christian apologetic tradition systematically engages best skeptic arguments (Augustine engaging Manichaeans + Pelagians; Aquinas engaging Averroist Aristotelianism; Pascal engaging Cartesian skepticism; modern apologetics engaging Hume + Mackie + Russell + Dawkins). The literature shape is not echo-chamber; the best apologetic literature explicitly steel-mans atheist arguments before responding.

Asymmetric-evidential weight is sometimes legitimate. Not all asymmetric scrutiny is confirmation bias. Bayesian rationality predicts that prior probability rationally affects how much weight new evidence gets, extraordinary claims (against well-established prior) require correspondingly more evidence. The atheist who demands more evidence for miracle claims than for ordinary natural explanations may simply be doing Bayesian updating from a prior naturalist commitment. The Christian who weights the historical resurrection evidence highly may be doing Bayesian updating from a prior theist commitment. The proper response to such asymmetry is NOT "you have confirmation bias" (which presupposes the asymmetric weighting is illegitimate); it's substantive engagement on which prior is rationally warranted (the Cumulative Case for Christian Theism move). Confirmation bias proper is asymmetric weighting of the evidence at hand (not of priors); it requires showing that the same standard would yield different conclusions if symmetrically applied, a substantive empirical claim.

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

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

Bayesian rational updating from differential priors. When a Christian gives more weight to a historical resurrection account than an atheist does, this need not reflect confirmation bias, it can reflect rational Bayesian updating from a different prior probability assessment. Charging "confirmation bias" without engaging the substantive prior-probability question is the misapplication; it conflates rational asymmetric weighting with biased asymmetric weighting.
Methodological-naturalism in scientific contexts. When scientists exclude supernatural causes from working hypotheses in laboratory chemistry, this is not confirmation bias, it's methodological-naturalism as a working stance for empirical science (as distinct from metaphysical-naturalism as a worldview claim). The two are distinct (per Plantinga + others); the methodological exclusion is part of how science works given its empirical-test-replication mandate. Christians who do science use methodological-naturalism without thereby committing confirmation bias against design-inferences in metaphysical contexts.
Steel-manned engagement that still concludes for one side. If a Christian apologist (e.g., Plantinga, Craig, McGrath, Hart) has demonstrably engaged best-version atheist arguments + still concludes for Christianity, the conclusion does not reflect confirmation bias. The bias requires that disconfirming evidence be selectively-ignored / discounted-at-search-stage; engagement-then-rejection is a different operation.
Christian theological confidence in core doctrines after extensive engagement. Athanasius's Trinitarian theology after extensive engagement with Arianism; Augustine's grace theology after extensive engagement with Pelagianism; Calvin's soteriology after extensive engagement with semi-Pelagian options; modern responses to liberal-Protestant + Bultmannian + Jesus-Seminar treatments, these reflect doctrinal positions held with high confidence after substantive engagement with disconfirming proposals, not confirmation-bias-driven pre-engagement positions.
Eyewitness testimony in legal + historical contexts. When jurors or historians give weight to eyewitness testimony for some-but-not-all witnesses, the differential weighting is not confirmation bias, it can reflect substantive judgments about witness credibility, motive, opportunity-to-observe, corroboration. The Resurrection-witness analysis (the disciples' transformation from fearful into bold preachers + their willingness to die for the testimony + the inclusion of women as primary witnesses against contemporary social conventions + multiple-attestation across independent sources) is differential weighting of testimony based on substantive credibility-criteria, not confirmation bias.
Conviction grown from converging-line cumulative case. When Christian apologetics presents a cumulative case (cosmological + teleological + moral + historical + religious-experiential lines converging), the resulting conviction is not confirmation bias, it's Bayesian updating across multiple independent evidential lines that each carry weight. The conviction reflects the convergence of independent lines, not selective attention within any single line. (See Cumulative Case for Christian Theism.)
Pastoral wisdom drawn from extensive case-experience. When a pastor with 30 years of pastoral counseling experience reports patterns observed across hundreds of cases, this is not confirmation bias, it's domain-expertise pattern-recognition from extended observational data. The patterns may be challengeable on independent grounds, but charging confirmation bias as opposed to substantive engagement is the misapplication.
Conversion testimonies that include best-effort-at-engagement-then-conviction. C.S. Lewis's documented progression from atheism through theism to Christianity; Antony Flew's progression from atheism to deism (see Antony Flew (Deism Conversion 2004)); Lee Strobel's progression from atheism to Christianity through investigative engagement (see Lee Strobel (Conversion 1981)), these reflect changed prior commitments under substantive evidential pressure, the opposite of confirmation bias's pattern. Confirmation bias predicts conviction-stability under disconfirming evidence; conversion against prior commitments is its inverse.

When confirmation bias IS operating

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

Christian who has never read best atheist arguments (Mackie, Sobel, Oppy, Hume, Russell) and rejects atheism on grounds primarily drawn from in-tradition literature.
Christian who interprets every personal-life event as God's hand (good things = blessing, bad things = testing / correction / mystery) without ever asking what would count as the inverse.
Atheist who has never read best Christian arguments (Aquinas, Anselm, Plantinga, Swinburne, Hart, McGrath, Craig) and rejects Christianity on grounds primarily drawn from New Atheist popularizations.
Atheist who interprets every religious atrocity as essentially-religious + every religious good as accidentally-religious (or vice versa for non-religious comparators).
Apologist (either side) who cannot articulate any plausible scenario of disconfirming evidence for the held position, unfalsifiability as a marker of confirmation bias having taken over the inferential process.
Selective-citation patterns where in-tradition sources are credentialed-respected and out-tradition sources of comparable credentials are dismissed-as-biased.
Scripture-interpretation where contested-passage handling consistently favors the held position regardless of grammatical-historical exegetical merits.
Source-of-evidence-evaluated-by-conclusion patterns: same source treated as reliable when conclusions match held position + unreliable when conclusions oppose it.

The corrective in each case is methodological discipline rather than denial: deliberate exposure to best opposing arguments + asymmetric-scrutiny audit + falsifiability commitment + steel-manning practice + cross-tradition source diversification.

Christian theology of cognitive bias

Christianity has a substantive theological framework for understanding why confirmation bias exists + what its corrective looks like:

Noetic effects of the fall (Romans 1:18-23 + Romans 8:7 + 1 Corinthians 2:14 + Ephesians 4:17-18). Human cognitive faculties are not neutral instruments; they bear distortion from sin's effect on the whole person including the mind. The Reformed tradition develops this as the sensus divinitatis operating but partially-suppressed; Plantinga's Reformed Epistemology develops it as warranted-Christian-belief grounded in proper-functioning faculties operating in the cognitive environment they were designed for.
Epistemic humility commanded, "trust in the LORD with all your heart and lean not on your own understanding" (Proverbs 3:5); "Do not be wise in your own eyes" (Proverbs 3:7); "the way of a fool is right in his own eyes" (Proverbs 12:15). Wisdom literature commands the explicit acknowledgment of cognitive limitation including bias.
Renewal of the mind (Romans 12:2). The Christian sanctification trajectory includes the renewing of the mind, which is, structurally, the corrective for confirmation bias and adjacent cognitive distortions. Sanctification of the intellect involves deliberate exposure to truth + rejection of self-deception + willingness to be corrected.
Truth-loving commanded, "speaking the truth in love" (Ephesians 4:15); "let every person be quick to hear, slow to speak, slow to anger" (James 1:19). The slow-to-speak imperative has an epistemological reading: take time to genuinely engage opposing views before responding.
Holy-Spirit illumination (1 Corinthians 2:10-14). Christian epistemology includes the Spirit's role in illuminating spiritual truth, which is NOT a confirmation-bias mechanism (it does not selectively support pre-held beliefs; it operates via the Spirit's ministry to disclose truth that natural cognition alone cannot reach). Distinguishing Spirit-illumination from confirmation bias requires methodological discipline + community-discernment + Scripture as objective-norm.

The Christian apologist who takes confirmation bias seriously thereby aligns with biblical wisdom-tradition + Reformation epistemology + classical theological anthropology, the bias is acknowledged + named + addressed methodologically, not denied or rationalized.

Apologetic deployment notes

When Christian engages confirmation-bias charge:

Acknowledge the bias is real + universal. Don't deny or minimize. Christians have it; atheists have it; you have it; everyone has it. The charge is not asymmetric-attack on Christianity specifically.
Demand symmetric application. If the charge is meant to discredit Christianity specifically, ask the opponent to specify what asymmetric pattern obtains in Christians-vs-atheists empirically; the literature does not support asymmetric distribution.
Cite Christianity's substantive falsifiers. 1 Corinthians 15:14 (resurrection), Old-Testament Messianic prophecies (historical-fulfillment conditions), Matthew 7:16 (practical-fruits test). Christianity is empirically- + historically- + practically-falsifiable in principle; opponents who claim it's unfalsifiable typically have not engaged these falsifiers.
Cite best-engagement Christian apologetic literature. Plantinga (Warranted Christian Belief), Craig (Reasonable Faith), Hart (The Experience of God), McGrath (Mere Apologetics), Lewis (Mere Christianity), Pascal (Pensées), Aquinas (Summa contra Gentiles), all explicitly engage best opposing arguments rather than echo-chamber-affirming. Demonstrate via specific citation.
Cite conversion-testimonies-against-prior-commitment. Antony Flew, Lee Strobel, C.S. Lewis, Augustine, Paul on the Damascus Road, conversions against held atheist / hostile / persecuting prior commitments are the structural inverse of confirmation bias's prediction.
Offer joint methodological discipline. "If we both acknowledge confirmation bias, what discipline can we adopt to control for it in this conversation? I'll commit to engaging your best-version arguments; will you do the same?" This converts the charge into joint epistemic commitment rather than position-discrediting.
For interlocutor's specific charge of bias, request specific evidence. "Which evidence do you think I'm selectively-ignoring? Which best-version argument do you think I haven't engaged? Let's work through it." Concrete-specific request displaces blanket-charge with addressable engagement.

When Christian charges atheist with confirmation bias:

Be specific not blanket. "You've cited Dawkins + Harris + Hitchens; have you read Plantinga + Hart + McGrath?", specific citation-asymmetry beats generic confirmation-bias accusation.
Acknowledge the symmetry. "The same bias operates in Christians too; the question is whether either of us is willing to do the methodological discipline." Symmetric framing keeps the conversation honest and avoids the same accusation being thrown back.

Scholarly resources

Peter Wason "On the failure to eliminate hypotheses in a conceptual task" Quarterly Journal of Experimental Psychology 12 (1960): 129-140, the original 2-4-6 task experiment establishing confirmation bias as a robust cognitive phenomenon.
Raymond Nickerson "Confirmation Bias: A Ubiquitous Phenomenon in Many Guises" Review of General Psychology 2 (1998): 175-220, comprehensive review of the empirical literature establishing the bias's distribution across domains.
Daniel Kahneman Thinking, Fast and Slow (Farrar, Straus and Giroux, 2011), System 1 / System 2 framework for understanding why confirmation bias is the cognitive default + what disciplinary correctives look like.
Hugo Mercier + Dan Sperber The Enigma of Reason (Harvard, 2017), argumentative theory of reason proposing that human reasoning evolved for argumentative-deployment-in-social-context, which predicts confirmation bias as a feature rather than a bug.
Alvin Plantinga Warrant and Proper Function (Oxford, 1993) + Warranted Christian Belief (Oxford, 2000), Reformed-epistemology framework for understanding cognitive faculties' design-purpose + the noetic effects of sin including bias.
David Bentley Hart The Experience of God (Yale, 2013), sustained demonstration that New Atheist literature engages strawmanned theology rather than classical-theist God-as-being-itself, an example of confirmation-bias-at-the-level-of-which-position-gets-engaged.
Diederik Aerts + Sandro Sozzo + Tomas Veloz literature on quantum-cognition models of confirmation-bias-resistant belief-revision (technical philosophy-of-mind).