Survivorship bias

Survivorship bias is the statistical error of drawing conclusions from data that includes only entities that survived some selection process, while overlooking entities that did not. It is a specific form of selection bias caused by non-random missing data: the entities still available for observation differ systematically from the original population in ways that bias estimates of the underlying parameters. The classic methodological treatment is Abraham Wald's 1943 memoranda for the Statistical Research Group at Columbia University on estimating aircraft vulnerability from data on returning bombers — survivors of combat — using probabilistic methods to recover the underlying vulnerability structure that the survivor-only data obscured.

The financial-economics literature provides the most rigorously demonstrated quantitative effects. Brown, Goetzmann, Ibbotson and Ross (1992 Review of Financial Studies) showed that mutual fund performance studies including only currently-existing funds systematically overstate average performance because failed funds (which exit the dataset) had below-average returns. Elton, Gruber and Blake (1996) and subsequent work have documented that survivorship bias in US equity fund data typically inflates estimated average returns by 0.5 to 1.5 percentage points per year — a substantively large effect for long-horizon investment analysis. The financial survivorship literature is methodologically rigorous and the effect is well-characterized empirically.

The basic phenomenon is one of the more robustly supported methodological concepts in research methodology. It has clear formal definition (left-truncation of the sample by a non-random survival process), documented quantitative effects in well-studied domains (finance), and broad generalization across selection-bias contexts (business success literature, academic publication, clinical trials, historical interpretation). The popular treatments — including the iconic illustrated bomber diagram that circulates widely online — sometimes overstate the universality or embellish the historical narrative, but the underlying methodological insight is real and the empirical demonstrations in finance are rigorous.

Survivorship bias matters at three levels with different evidence bases.

For empirical research and data analysis. Survivorship bias is one of the most important and most commonly missed selection effects in applied data analysis. Any dataset that includes only entities that survived some process — companies still in business, mutual funds still operating, students who completed a program, patients who continued treatment, customers who renewed subscriptions — is potentially affected. The bias is particularly dangerous because it operates silently: the analyst sees the data they have without seeing the data they are missing, and standard inference procedures assume the missing data are missing-at-random when in survivorship cases they are decidedly not. Recognition of survivorship bias has reshaped methodological practice in finance, clinical trials, organizational research, and many other applied fields.

For business strategy and success advice. The most common popular application of survivorship bias is to business-success literature: profiles of successful companies and founders that imply their specific practices caused their success, while ignoring the much larger number of failed companies and founders that used similar practices. The implication: studying surviving successes without studying matched failures produces overly optimistic conclusions about which practices “cause” success. The honest framing: a practice that 30 surviving successful companies share may also be shared by 300 failed companies that no one is writing about. Causal inference about success requires matched comparison, not survivor-only analysis. The point is widely cited in popular business literature (Mlodinow 2008 The Drunkard's Walk is the most accessible popular treatment) and applies whenever success-only data is used to draw causal conclusions.

For personal decision-making and self-evaluation. Survivorship bias has practical implications for how to interpret information about life paths, career outcomes, and personal decisions. The visible success stories in any domain are a non-random sample of the people who tried; the failures are largely invisible. This matters for career planning (the visible startup founders are not a representative sample of people who started startups), academic decisions (the visible PhD-holders are not representative of people who started PhD programs), creative pursuits (the visible successful authors are not representative of people who wrote novels), and many other domains. Recognizing the bias does not mean dismissing visible successes; it means contextualizing them against the much larger but invisible population of similar attempts.

Survivorship bias as a methodological concept has multiple parallel origins in different applied fields, with the wartime statistical work of Abraham Wald providing the most-cited foundational example.

Wald 1943 SRG memoranda. Abraham Wald (1902-1950), Hungarian-American statistician at Columbia University, joined the Statistical Research Group (SRG) during World War II. The SRG, directed by W. Allen Wallis, was a team of mathematicians and statisticians working on military applications. Wald wrote eight memoranda on aircraft vulnerability between 1942 and 1944, with the central problem being how to estimate the vulnerability of different parts of an airplane from data on planes that had returned from combat — survivors. The military wanted to add armor to the aircraft sections that showed the most damage on returning planes; Wald's insight was that this approach was backwards.

The reasoning: returning planes constituted a non-random sample of planes hit by enemy fire. Sections of the aircraft that showed many bullet holes on returning planes were sections where damage was survivable. Sections that showed few bullet holes on returning planes were sections where damage was likely fatal — the planes hit in those areas did not return to be photographed. The armor should be placed on the sections with the fewest holes on returning aircraft, not the most. Wald developed formal probabilistic methods for estimating the vulnerability of different parts of the plane given the survivor-only data, using assumptions about the distribution of hits and the conditional probability of plane destruction given a hit on a specific section.

The memoranda were classified during the war. They were declassified and reissued in 1980 by the Document Center of the Center for Naval Analyses (CNA), 2000 North Beauregard St., Alexandria, Virginia, after W. Allen Wallis recovered the original papers and provided them to Phil DePoy at CNA. The canonical academic re-analysis is Mangel and Samaniego (1984) “Abraham Wald's Work on Aircraft Survivability” in Journal of the American Statistical Association, which provides a careful contemporary exposition of Wald's methods and assumptions.

An important historical caveat. The famous illustrated diagram showing red dots on a bomber silhouette, widely circulated online and in popular survivorship-bias treatments, is a modern visualization created by Wikimedia contributor McGeddon in 2016 based on a Lockheed PV-1 Ventura drawing. It is not a reproduction of any figure from Wald's actual 1943 memoranda — Wald's eight memoranda are technical and unillustrated. The popular retelling sometimes implies the bomber diagram is a historical artifact from the 1940s SRG work; it is not. The methodological substance is real and well-documented; the iconic visual is a pedagogical reconstruction. This matters for citation: the historical record should not be embellished beyond what the original memoranda actually contain.

The financial-economics literature. The most rigorous quantitative demonstration of survivorship bias comes from finance. Brown, Goetzmann, Ibbotson and Ross (1992 Review of Financial Studies) showed that mutual fund performance studies that include only currently-existing funds systematically overstate average performance because failed funds (which have exited the dataset) had below-average returns. The paper argued that the apparent persistence of past mutual fund performance documented in earlier work was partially an artifact of survivorship bias: funds with high volatility relative to their alpha would tend to be censored from the data at higher rates when their returns were low, biasing the surviving-fund cross-section toward apparent performance persistence.

Elton, Gruber and Blake (1996) in Review of Financial Studies provided independent estimates of survivorship-bias magnitude in mutual fund data, finding the bias inflated estimated returns by approximately 0.9 percentage points per year on average over their sample period. Brown, Goetzmann and Ross (1995) in Journal of Finance developed the formal framework for analyzing the survival distribution itself. Carhart (1997) in Journal of Finance — one of the most-cited mutual fund performance papers — specifically corrects for survivorship bias and is the canonical reference for properly-corrected estimates. The cumulative literature has established that survivorship bias in US equity fund data inflates estimated average returns by 0.5 to 1.5 percentage points per year, with the magnitude varying by sample period and fund type.

The popular synthesis. Leonard Mlodinow's 2008 book The Drunkard's Walk: How Randomness Rules Our Lives (Pantheon Books) provided the most accessible popular treatment of survivorship bias along with related selection effects. The book uses the Wald story (with appropriate methodological substance), the financial-economics evidence, and additional examples from business success literature to communicate the basic concept to non-technical audiences. Jordan Ellenberg's 2014 How Not to Be Wrong: The Power of Mathematical Thinking also features the Wald story prominently. These popular treatments are responsible for much of the contemporary awareness of survivorship bias in general audiences.

The contemporary state. Survivorship bias is one of the more robustly established methodological concepts in research methodology. The Wald historical example provides the canonical illustration; the financial-economics literature provides rigorous quantitative demonstration; the popular literature has communicated the basic insight broadly. Contemporary methodological practice in finance, clinical trials, organizational research, and most empirical fields explicitly accounts for survivorship and related selection effects. The remaining methodological work focuses on specific corrections in specific data contexts (Heckman selection models, inverse probability weighting, formal causal-inference frameworks for handling selection bias).

Survivorship bias has a precise statistical structure. Understanding the mechanism matters because the formal structure distinguishes survivorship bias from related selection biases and clarifies when correction methods are applicable.

The basic structure

Survivorship bias is a form of selection bias caused by non-random sample selection based on the outcome being studied or correlates of it. The structure: a population P of entities is observed at time t, but only entities that have survived some selection process from time t-k to t are observed. If survival is correlated with the variable of interest, the observed sample distribution differs systematically from the true population distribution at time t-k. Standard inference procedures that treat the observed sample as representative of the population produce biased estimates.

Formally: let X be the variable of interest (e.g., investment return, business performance, treatment outcome). Let S = 1 if an entity survives to be observed and S = 0 otherwise. The observed mean of X in surviving entities is E[X | S=1]. The population mean is E[X]. Survivorship bias is the difference E[X | S=1] - E[X], which is non-zero whenever X is correlated with S. In mutual fund data, the correlation is positive (high-return funds are more likely to survive), so survivor-only estimates overstate the population mean.

Distinguishing survivorship from related selection biases

Several closely-related selection biases involve different specific structures:

• Survivorship bias proper: non-random selection by survival; selection is correlated with the outcome variable being studied; survivors are observed in cross-section at a single time point. Classic example: mutual fund cross-sectional return analysis at year-end.
• Left-truncation: entities that fall below some threshold before the observation window are excluded from the sample. Classic example: salary data that omits people who quit before being surveyed.
• Right-censoring: outcome variable is observed up to some time point but not afterward. Classic example: clinical trial follow-up that ends at 5 years, so outcomes occurring after 5 years are unobserved.
• Attrition bias: longitudinal samples where participants drop out at rates correlated with the outcome. Classic example: longitudinal study of depression where more depressed participants are more likely to stop responding to follow-up surveys.
• Publication bias: studies with statistically significant or interesting results are more likely to be published than those without. Classic example: meta-analyses systematically overestimating true effect sizes because null results never reach publication.
• Sampling bias more broadly: the sample is drawn non-randomly from the population. Survivorship is one specific form of sampling bias.

The distinctions matter because correction methods differ. Survivorship bias proper can sometimes be corrected with inverse-probability weighting if survival probabilities can be modeled. Attrition bias requires longitudinal data on dropout patterns. Right-censoring is handled by survival-analysis methods (Kaplan-Meier estimation, Cox proportional hazards). Publication bias requires meta-analytic correction methods (trim-and-fill, selection models, RoBMA-PSMA).

Correction methods

Several methodological approaches address survivorship bias when survival data is available:

• Include the dead. The most robust correction is to include entities that did not survive in the analysis. In finance: include funds that closed during the sample period, not just funds that survived to the analysis date. Brown, Goetzmann, Ibbotson and Ross (1992) and Carhart (1997) data sources explicitly track dead funds; modern fund databases (CRSP Mutual Fund Database, Morningstar) provide survivorship-bias-free data for academic research.
• Heckman selection correction. Two-step procedure (Heckman 1979 Nobel-prize-winning work) that models the selection process explicitly and includes a correction term in the outcome equation. Standard tool in econometric work where selection is a concern.
• Inverse probability weighting. Reweight observed entities by the inverse of their probability of being observed, to recover the population distribution. Requires modeling the selection process; can be unstable when selection probabilities are extreme.
• Sensitivity analysis. When formal correction is not feasible, analyze how conclusions would change under plausible assumptions about the missing entities. Useful when the magnitude of the bias is the question of interest.

The Wald 1943 work was an early example of formal correction: Wald developed probabilistic methods to estimate the vulnerability of plane sections that destroyed planes, using only data on surviving planes that had been hit in survivable sections. The substantive insight was the recognition that the data needed correction; the formal correction was a substantial methodological contribution.

Survivorship bias is “measured” in the sense of estimating its magnitude when it is present. Several approaches are standard.

Direct comparison of survivor-only and survivorship-corrected datasets. The most rigorous approach when data permit. In finance: compare estimated mean returns using only currently-existing funds vs. using a survivorship-bias-free database that includes funds that closed during the sample period. Brown, Goetzmann, Ibbotson and Ross (1992) demonstrated this approach for mutual funds; Elton, Gruber and Blake (1996) provided independent estimates with similar magnitude (~0.9 percentage points/year upward bias). The general framework: bias magnitude = mean of survivors - mean of all original entities. When both samples are available, the bias is directly measurable.

Bounds analysis under assumptions. When direct comparison is not possible (the dead are unobserved), bound the bias under assumptions about how the dead would have performed. Worst-case bounds (assume all dead would have had the worst possible performance) and best-case bounds (assume all dead would have had average performance) bracket the true population mean. Useful for sensitivity analysis but typically produces wide bounds.

Cross-context replication. Compare estimates in domains where both survivor-only and survivorship-corrected data exist, then use the relationship to inform corrections in similar domains where corrected data is unavailable. Less rigorous than direct measurement; useful when no within-domain correction is possible.

Heckman-style selection modeling. When the selection process can be modeled (e.g., probability of fund closure depends on past performance, fund size, fees), the selection model provides estimates of the bias under the model assumptions. Standard in econometric work; depends heavily on correct specification of the selection process.

What the LBL tools capture. The Cognitive Bias Susceptibility tool in the Behavior Lab measures susceptibility to several reasoning biases including base-rate neglect, conjunction fallacy, framing, and anchoring; survivorship bias is conceptually adjacent to these. The Crossroads Lab tools (Career Pivot Decision Matrix, Should I Quit My Job?) implement decision-support frameworks where survivorship considerations matter: evaluating career options based only on visible success stories is a form of survivorship-biased reasoning. The LBL tools do not specifically test for survivorship-bias susceptibility, but the broader pattern of base-rate reasoning and matched-comparison thinking that the tools support is the underlying reasoning skill relevant to survivorship-bias recognition. Awareness of the bias is more useful than a specific test score for most practical applications.

Survivorship bias sits among several related selection biases that are frequently confused.

• vs. selection bias (broader category). Selection bias is the general category of biases arising from non-random sample selection. Survivorship bias is one specific form — selection by survival or persistence. Other forms include self-selection (volunteers differ from the general population), gatekeeper selection (admitted vs. rejected applicants differ), and assignment-related selection in non-randomized studies. The category “selection bias” is broader than survivorship bias and the two terms should not be used interchangeably.
• vs. cognitive bias. Cognitive biases are systematic departures from normative inference in human reasoning. Survivorship bias is a statistical bias in data analysis — the data are biased, not the reasoner's cognition. However, survivorship bias often interacts with cognitive biases: humans are more likely to notice and remember visible successes than invisible failures (availability heuristic), and they often fail to consider the missing data that would correct the bias. The cognitive failure to recognize survivorship structure when present is a cognitive bias; the statistical bias in the data is survivorship bias proper. The terms describe different but related phenomena.
• vs. publication bias. Publication bias is the tendency for studies with statistically significant or interesting results to be more likely to be published. It is a specific form of selection bias in the published research literature. Survivorship bias is broader and can apply to any data where survival is non-random; publication bias is a specific case where “survival” means being published. The two concepts are closely related and both produce overestimation of true effect sizes in meta-analyses.
• vs. attrition bias. Attrition bias is the bias in longitudinal samples where participants drop out at rates correlated with the outcome. Survivorship bias is the bias in cross-sectional samples where current observation requires having survived from some earlier time. The structural difference: survivorship bias operates over a single observation time looking backward; attrition bias operates over multiple observation times with dropout between them. The mechanisms and correction methods differ.
• vs. left-truncation and right-censoring. Left-truncation: entities below some threshold are excluded from the sample. Right-censoring: outcome variable is observed up to some time point but not afterward. Both are technical forms of incomplete data that are related to but distinct from survivorship bias. Survival analysis (Kaplan-Meier estimation, Cox proportional hazards) provides the formal apparatus for handling right-censoring.
• vs. availability heuristic. The availability heuristic is the cognitive shortcut of judging probability or frequency by ease of mental retrieval. Visible success stories are more available than invisible failures, contributing to cognitive failure to recognize survivorship bias when present. The availability heuristic is the cognitive mechanism that makes survivorship bias hard to notice; survivorship bias is the structural property of the data that the cognitive failure produces biased conclusions about.
• vs. halo effect. The halo effect is the cognitive bias of letting overall impressions of a person or entity influence judgments of specific traits. Survivorship bias is a data-structural property, not a cognitive bias. The two can interact (overall success leads to halo effects in evaluation, while the success itself reflects survivorship); but they are distinct concepts.
• vs. Dunning-Kruger effect. Dunning-Kruger is a specific cognitive bias about the relationship between competence and self-assessment. Survivorship bias is a data-structural bias. The two are unrelated except that both are well-known “biases” in popular discussions.
• vs. sunk cost fallacy. Sunk-cost reasoning is the tendency to let past unrecoverable costs influence current decisions. Survivorship bias is about data structure, not decision-making. The two are conceptually distinct and both contribute to decision quality through different mechanisms.

Example 1 — The startup success book

A reader picks up a popular business book that profiles 50 successful technology startups and identifies the practices their founders share. The book argues that these practices — aggressive hiring, particular cultural values, specific management approaches — explain the startups' success and recommends adopting them. The reader finds the case studies compelling and the analytical framework persuasive.

This is the canonical popular application of survivorship bias. The book studies a non-random sample of 50 startups — the visible successful ones — without studying the matched comparison of similar startups that did not succeed. If 50 successful startups share certain practices, the question is: how many failed startups also shared those same practices? If 200 failed startups had similar approaches but did not succeed, the practices clearly do not cause success in any straightforward sense. Without the comparison data, causal inference from the surviving sample alone is not valid. The honest framing: practices that 50 successful startups share are correlated with their success in the data observed; whether they cause success requires comparing to the full population of attempts, not just the survivors. Mlodinow (2008) and many subsequent popular treatments use this exact example. The recommendation for readers: be skeptical of success-only case studies, and ask explicitly about the comparison failures the analyst did not study.

Example 2 — The mutual fund advertisement

An investor reviews an advertisement for a mutual fund company that highlights the strong 10-year track records of three specific funds. The funds returned 14%, 12%, and 11% annually over the decade — substantially above the S&P 500 benchmark. The investor is impressed and considers investing.

This is the financial-economics example documented rigorously in Brown, Goetzmann, Ibbotson and Ross (1992) and Elton, Gruber and Blake (1996). The fund company offered, ten years ago, perhaps 50 funds. Some of those 50 funds performed badly and were closed by the fund company during the decade — their poor returns are no longer visible in marketing materials. The three funds being advertised are the survivors of an internal selection process at the fund family. The 14%-12%-11% returns are real, but they are the returns of survivors selected from a larger initial population whose typical return was much lower. Without the data on the closed funds, the advertised returns substantially overstate what an investor would have earned by selecting funds at the start of the decade. The corrected average return across the original 50 funds (including the ones that closed) would typically be 1-2 percentage points lower than the advertised survivors. The honest framing: the advertised performance is not fraudulent but it is non-representative; the methodologically appropriate comparison is to the entire initial fund cohort, not the surviving subset. This is why academic mutual fund research now uses survivorship-bias-free databases (CRSP Mutual Fund Database) that include closed funds, and why retail investors should be skeptical of fund family marketing claims based on selected fund performance.

Survivorship bias is one of the better-supported methodological concepts in research methodology. The substantive caveats are also well-documented.

• The famous Wald bomber diagram is a modern reconstruction. The iconic illustrated diagram showing red dots on a B-17 or similar bomber silhouette, widely circulated online and in popular survivorship-bias treatments, was created by Wikimedia contributor McGeddon in 2016 based on a Lockheed PV-1 Ventura drawing. It is not a reproduction of any figure from Wald's actual 1943 SRG memoranda — Wald's eight memoranda are technical and unillustrated documents that focus on probabilistic estimation methods, not on producing illustrated guidance for engineers. The methodological substance of Wald's work is real and well-documented in Mangel and Samaniego (1984 JASA); the iconic visual is a pedagogical reconstruction. Citations should not present the modern diagram as a historical artifact from the 1940s.
• Not all selection bias is survivorship bias. Popular treatments sometimes use “survivorship bias” as a general term for any selection effect, but the concept has a specific structural meaning: non-random sample selection by survival or persistence. Other selection biases (self-selection into samples, gatekeeper selection, assignment-related selection) have different structures and require different corrections. Conceptual precision matters for choosing appropriate analytical methods.
• Effect-size magnitude varies substantially by context. The 0.5-1.5 percentage points/year survivorship bias in US mutual fund data is well-characterized; survivorship effects in other domains (business success literature, academic publication, historical interpretation) have less rigorous quantitative documentation. The basic phenomenon is robust; specific magnitude claims should be calibrated to the specific domain rather than extrapolated from finance.
• Correction requires data on the missing entities. The most robust survivorship-bias correction (include the dead) requires data on entities that did not survive. When such data is unavailable, correction options are limited and depend on strong assumptions. Sensitivity analysis can bound the bias under assumptions; formal correction (Heckman selection, inverse probability weighting) requires modeling the selection process explicitly. These methods have their own assumptions and limitations.
• The popular success-book example is qualitative, not quantitative. The argument that business-success literature suffers from survivorship bias is logically sound — case studies of surviving successes without matched comparison of failures cannot support causal claims. But the magnitude of the resulting bias is harder to characterize than the financial-economics case. Specific quantitative claims about “X% of practices identified in business success books would also be found in failed companies” are difficult to substantiate empirically because the data on failed companies is harder to access.
• Survivorship is a relative property, not absolute. An entity is a survivor of one process and may not be a survivor of another. A company surviving 10 years has survived business failure but not necessarily competitive irrelevance, ethical violations, or acquisition. Conclusions about “survivors” depend on the specific selection process being analyzed.
• Awareness does not automatically produce correction. Recognizing survivorship bias in an analysis does not automatically tell the analyst how to correct it. The correction depends on the specific selection process, available data, and modeling assumptions. Many analyses acknowledge survivorship concerns without providing rigorous corrections; the acknowledgment is honest but does not eliminate the bias.
• Some popular applications overstate the universality. Survivorship bias is genuinely pervasive in many domains, but not all observations of successful entities suffer from it. When the relevant population is well-defined and matched comparison data exists, conclusions can be drawn from successful entities without survivorship bias concerns. The popular framing that “all success advice is survivorship-biased” is an overstatement; some success analysis uses appropriate comparison and is methodologically sound.
• The cognitive failure (not noticing the bias) is psychologically real but the bias itself is statistical. Some treatments conflate the cognitive bias (failing to notice survivorship structure in data) with the statistical bias (the data is biased). These are related but distinct. Cognitive training can improve recognition of survivorship structure; eliminating the statistical bias requires data correction, not just cognitive awareness.

The Cognitive Bias Susceptibility tool in the Behavior Lab measures susceptibility to several reasoning biases including base-rate neglect, conjunction fallacy, framing, and anchoring; survivorship-bias recognition is conceptually adjacent to these reasoning skills. The Crossroads Lab tools (Career Pivot Decision Matrix, Should I Quit My Job?) implement decision-support frameworks where survivorship considerations matter: evaluating career options based only on visible success stories is a form of survivorship-biased reasoning that the tools help structure against. Together these tools support the broader pattern of base-rate reasoning and matched-comparison thinking that survivorship-bias recognition requires.

Survivorship bias is the statistical error of drawing conclusions from data that includes only entities that survived some selection process, while overlooking entities that did not. It is a specific form of selection bias caused by non-random missing data. The classic methodological treatment is Abraham Wald's 1943 memoranda for the Statistical Research Group at Columbia University on estimating aircraft vulnerability from data on returning bombers, declassified and reissued in 1980; the canonical academic exposition is Mangel and Samaniego (1984 JASA). The famous illustrated bomber diagram widely circulated online is a modern visualization created in 2016, not from Wald's actual unillustrated memoranda — the methodological substance is real; the iconic visual is a pedagogical reconstruction. The most rigorous quantitative demonstration is Brown, Goetzmann, Ibbotson and Ross (1992 Review of Financial Studies) on mutual fund performance, with subsequent work by Elton, Gruber and Blake (1996) and Carhart (1997) establishing that survivorship bias in US equity fund data typically inflates estimated returns by 0.5 to 1.5 percentage points per year. The phenomenon generalizes broadly: business success literature, academic publication, clinical trials, historical interpretation. Popular treatments (Mlodinow 2008 The Drunkard's Walk) have communicated the concept broadly. Survivorship bias is one of the better-supported methodological concepts in research methodology; conceptual precision matters because not all selection bias is survivorship bias, correction methods depend on the specific selection structure, and some popular applications overstate universality. The basic phenomenon is robust; specific applications require calibration to the specific domain and data context.

This entry is intended as a citable scholarly reference. Choose the format that matches your context. The retrieval date should reflect when you accessed the page, which may differ from the entry's last-reviewed date shown above.

LifeByLogic. (2026). Survivorship Bias: Wald 1943, Brown-Goetzmann. https://lifebylogic.com/glossary/survivorship-bias/

LifeByLogic. "Survivorship Bias: Wald 1943, Brown-Goetzmann." LifeByLogic, 14 May 2026, https://lifebylogic.com/glossary/survivorship-bias/.

LifeByLogic. 2026. "Survivorship Bias: Wald 1943, Brown-Goetzmann." May 14. https://lifebylogic.com/glossary/survivorship-bias/.

@misc{lblsurvivorshipbias2026,
  author = {{LifeByLogic}},
  title = {Survivorship Bias: Wald 1943, Brown-Goetzmann},
  year = {2026},
  month = {may},
  publisher = {LifeByLogic},
  url = {https://lifebylogic.com/glossary/survivorship-bias/},
  note = {Accessed: 2026-05-14}
}