Survivorship bias and attrition bias both mislead by excluding relevant cases, but they describe different mechanisms. Survivorship bias comes from conditioning on cases that passed a selection or visibility filter; attrition bias comes from informative loss from an initially enrolled or observed sample during follow-up. The distinction matters in research, test-score claims, and retention dashboards.
The short answer
| Bias | Core problem | Key question |
|---|---|---|
| Survivorship bias | The evidence includes only cases that survived a filter or remained visible. | Who was filtered out before this conclusion was drawn? |
| Attrition bias | People or cases initially observed are lost during follow-up in a way that distorts the result. | Who left after observation began, and could their absence change the estimate? |
Attrition can leave a researcher with a survivor-only group, but the terms point to different facts. Survivorship bias emphasizes the filter that determines who is visible. Attrition bias emphasizes loss from a defined starting sample over time.
For both, the repair begins with the denominator: identify the full group the claim is about, not just the cases still available to inspect.
Survivorship bias: drawing lessons from visible survivors
Survivorship bias occurs when a conclusion rests on cases that survived a selection, screening, or visibility filter while failed, discontinued, or otherwise absent cases are ignored.
The filter does not need to be literal survival. It may be a company remaining in business, a student finishing a program, a product still being used, or a public story that receives attention because it ended well.
Research example: graduates of a program
A university surveys only students who completed a graduate program and finds that they report strong career outcomes. It concludes that the program reliably produces those outcomes.
The concern is survivorship bias because the survey conditions on program completion. Students who enrolled but did not complete the program are absent, and their outcomes could be relevant to a claim about the program’s overall results.
Rejected applicants and people who attended other programs may also matter if the university wants to claim the program caused the difference. Those are separate comparison-group concerns, not the central survivorship problem in this example.
Exam-score example: advice from high scorers
A test-prep site interviews students with excellent scores. Many say they completed a full practice test every weekend, so a reader concludes that this schedule reliably produces excellent results.
That conclusion may reflect survivorship bias. The reader sees successful students who used the schedule, but not students who used a similar schedule and earned lower scores, burned out, or stopped preparing.
The advice may still be useful. What it cannot establish by itself is that the schedule caused the score or that it works for most students. This is a common version of the broader reasoning error covered in Survivorship Bias Explained: Examples, Mistakes, and a Quick Quiz.
Product example: analyzing only retained users
A subscription product team studies its most active current users and finds that they frequently use a feature during their first week. The team decides that promoting the feature will improve retention for all new users.
The observed users survived the retention filter: they are still active. Former users may have tried the feature and disliked it, or may never have reached it because onboarding failed. Current users are not automatically representative of everyone who signed up.
A better starting point is the original signup cohort, followed by a comparison of retained and churned users. The familiar historical illustration is Survivorship Bias in WWII: The Airplane With the Bullet Holes, Explained.
Attrition bias: informative loss during follow-up
Attrition bias arises when members of an initially enrolled or observed sample are lost during follow-up, and that loss changes what the remaining data can fairly show.
Dropout alone is not enough. Unequal loss rates between groups are a warning sign, but they are neither necessary nor sufficient to prove attrition bias. Loss can be equally common in two groups yet still be informative if the people leaving differ in outcome-relevant ways. Conversely, different loss rates may not materially bias a result if the losses are unrelated to the outcome being estimated.
The central question is not simply, “How many left?” It is, “Could the missing cases have changed the comparison or estimate?”
Research example: a stress-management course
A study follows employees assigned to a stress-management course and employees in a comparison condition. During follow-up, course participants with the highest initial stress are especially likely to stop responding to surveys.
At the final survey, the remaining course participants report lower stress than the comparison group. That result may be distorted because the observed course group no longer represents everyone who began the intervention. The loss is informative: it is related to a characteristic that matters to the outcome.
A careful report should show how many participants began each group, how many remained at each follow-up, and what is known about the missing observations.
Exam-score example: a reasoning workshop
A school follows students enrolled in a ten-week reasoning workshop and measures practice-test improvement at the end. Students who struggle most are more likely to stop attending and miss the final assessment.
If the school reports average improvement only among students who completed the final test, the estimate may be too favorable. The problem is not simply that some students left; it is that loss from the initially enrolled sample is connected to likely performance.
This is attrition bias because the relevant distortion develops during follow-up. It is not merely a success-filtered collection of testimonials.
Product-retention example: an onboarding experiment
A product team assigns new users to one of two onboarding versions and asks them to complete a satisfaction survey two months later. Users who cancel early rarely complete the survey.
Suppose Version A receives higher satisfaction ratings among respondents. The result might show better onboarding, but it might also reflect attrition bias if dissatisfied users were disproportionately missing from Version A’s follow-up responses. The two respondent groups may no longer be comparable to their original assigned groups.
The useful analysis keeps the original cohorts visible and reports follow-up completion alongside satisfaction results.
How to tell survivorship bias from attrition bias
Use this sequence when reading a study, argument, score claim, or dashboard.
- Define the population the conclusion claims to describe.
- Identify the observed cases used as evidence.
- Ask whether visibility depended on surviving a selection filter.
- Ask whether cases were initially observed but later lost during follow-up.
- Determine whether missing cases could plausibly change the conclusion.
Call it survivorship bias when the argument treats visible survivors as representative while disregarding filtered-out cases. Call it attrition bias when informative loss from the original sample during follow-up distorts the result.
Real cases can contain both. A study might recruit only graduates from a program, then lose graduates unevenly during a later follow-up. Name both mechanisms when each does separate explanatory work.
This kind of diagnosis is useful for logical-reasoning questions: separate the stated evidence from the group a conclusion claims to cover. For a broader approach to identifying argument gaps, see LSAT Logical Reasoning Question Types: How to Recognize and Solve Each One.
Quick identification practice
1. Profitable startups
A podcast interviews founders whose companies reached profitability and concludes that avoiding outside funding is the best strategy because most featured founders bootstrapped.
Answer: Survivorship bias. The podcast selected profitable companies, then generalizes from those visible survivors. It omits bootstrapped companies that failed or did not reach profitability.
2. Course-completion results
An online course measures students at enrollment and again after eight weeks. Students making little progress are much more likely to stop logging in and miss the final assessment. The provider reports gains only for students who completed the assessment.
Answer: Attrition bias. Students began in the observed sample but were lost during follow-up, and their loss is related to progress, the outcome being measured.
3. Equipment reliability
A hospital asks staff to rate the reliability of devices currently in use. Devices removed after repeated problems are not included.
Answer: Survivorship bias. The evidence is limited to devices that remained in use, excluding failed devices that would affect the reliability estimate.
Three items is enough to see the pattern, not enough to make it automatic. The survivorship bias quiz gives you twelve more scenarios, marked the moment you answer, with the filter named in each explanation.
FAQ
Is survivorship bias the same as attrition bias?
No. Survivorship bias concerns conditioning on visible cases that passed a filter. Attrition bias concerns informative loss from an initially observed sample during follow-up. Attrition may create a group of remaining “survivors,” but the causal mechanism is different.
Is unequal dropout always attrition bias?
No. Unequal dropout rates should prompt scrutiny, but they do not alone establish bias. The missing cases must plausibly differ in a way that changes the estimate or comparison.
Can equal dropout rates still create attrition bias?
Yes. If people with worse outcomes leave at similar rates in both groups, a remaining-sample estimate may still be distorted. Similar percentages do not guarantee similar missing information.
How can I spot survivorship bias in test-prep advice?
Ask what happened to people who used the same method but did not succeed, stopped studying, or never appeared in the success-story sample. Advice from high scorers can generate hypotheses, but it is not automatically evidence of a universal method.
Conclusion
Survivorship bias is about the cases made visible by a selection filter. Attrition bias is about informative loss from a defined sample during follow-up. In either case, better reasoning starts by accounting for who is missing, why they are missing, and whether the available cases can support the claimed conclusion.