AI Glossary / Statistics and probability / Intermediate
Lusser's law
Definition
Lusser's law states that the reliability of a series system equals the product of its component reliabilities when component survival events are statistically independent. A series system succeeds only if every required component succeeds. The law highlights how a chain of individually reliable parts can still have a substantially lower overall reliability. Its probabilities must refer to the same specified mission, duration, and operating conditions; multiplying unrelated performance percentages does not produce a valid system estimate.
What it means
“Series” describes the logic of success, not necessarily the physical arrangement of wires or equipment. If any required element's failure causes the mission to fail, all those elements must succeed together. For independent components with reliabilities R1 through Rn, system reliability is R1 times R2 times all the remaining terms through Rn.
Reliability is a probability of successful operation over a specified interval or mission. It is different from a component's accuracy on a benchmark and from steady-state availability in a repairable service. The product rule applies only when the component success events and system success definition actually match.
How it works
Define what counts as mission success and identify every required component. Draw the functional dependency structure, separating series requirements from genuine alternatives or redundancy. Estimate each component's survival probability for the relevant exposure and environment.
Check independence. Shared power, common software defects, environmental stress, and one component damaging another can create dependence. If independence is defensible, multiply component reliabilities. If it is not, model joint or conditional probabilities rather than substituting independent marginal probabilities.
Use the result for comparison and allocation, with uncertainty in component estimates retained. In a suitable independent series lifetime model, system hazard rates add, but this does not require constant hazards. Repairs, standby behavior, switching failures, and changing configurations need a model that includes those features rather than a simple first-failure calculation.
Three practical examples
- An engineering team estimates the probability that every essential module in an unattended instrument survives a specified mission.
- An operations team studies a workflow whose successful completion requires several independently reliable steps, checking whether that independence is credible.
- An AI service team maps required infrastructure dependencies before assessing whole-service reliability, rather than averaging the reliability figures of its components.
Think of it this way
Think of a relay in which every runner must complete their leg for the team to finish. Even if each runner is very likely to succeed, adding more required legs creates more opportunities for the relay to stop. The analogy assumes one runner's outcome does not change another's; shared conditions can break that assumption.
Worked example
Consider an original teaching system with three essential, independent components. Their probabilities of surviving the same mission are 0.98, 0.97, and 0.99. The system reliability is 0.98 times 0.97 times 0.99 = 0.941094, or approximately 94.11%.
The system's failure probability is 1 - 0.941094 = 0.058906, about 5.89%. Averaging the component reliabilities would give 98%, which is not the correct probability that all three succeed.
Suppose an improvement raises the first component's reliability to 0.995 while leaving the others unchanged. The new product is 0.995 times 0.97 times 0.99 = 0.9554985, approximately 95.55%. The gain is about 1.44 percentage points, not the full 1.5-point improvement in that component alone.
For a different illustration, ten independent required components each at 99% reliability produce 0.99 to the power 10, approximately 90.44%. Fifty such components produce approximately 60.50%. These are conditional design calculations under explicitly stated assumptions, not forecasts for arbitrary real systems. Adding a component that provides genuine redundancy can change the success logic and need not reduce reliability this way.
Illustrative case study
Fictional case: Elm Automation prepares an unattended inspection station with a sensor, controller, and communications module. The team initially reports each module's good reliability figure separately. A systems engineer models the all-required mission and calculates the combined probability.
Inspection reveals that all three modules share one power supply and enclosure. The team adds those dependencies and evaluates common environmental failures instead of assuming every failure event is independent. It also distinguishes mission completion without failure from the proportion of time the repairable station is available. The result guides testing and design priorities without presenting a simplified product as a complete risk assessment.
Common misconceptions and mistakes
The law is not an averaging rule and does not say the system equals its weakest component's reliability. The independent series product is no greater than any individual factor, but improving a component must still be evaluated in the full architecture.
Do not multiply marginal probabilities when failures share causes. A sequential workflow can use conditional success probabilities through the general chain rule, but those are different inputs from independent component reliabilities. Adding redundant elements, repair capability, or fault tolerance changes the model. A reliability estimate also does not quantify the severity of failure consequences.
Business use
Lusser's law encourages whole-system thinking in procurement, automation, and service design. It helps teams identify how many elements are truly essential and whether a proposed component improvement meaningfully changes mission success.
For AI applications, avoid treating retrieval accuracy, model accuracy, and tool uptime as automatically independent success probabilities. Errors can be correlated, and one stage may compensate for another or change the next stage's input distribution. Use end-to-end tests to validate architectural calculations, and report the mission definition, dependencies, and estimation uncertainty with the resulting reliability figure.
People also ask
Must components be physically in a line?
No; every one must be required for the defined system success.
Can the components have different reliabilities?
Yes, as in the worked example.
Does adding hardware always reduce reliability?
No. A required series component differs from a redundant alternative, and common-cause effects also matter.
Check your understanding
1. What is the original three-component reliability?
Answer: 0.941094, approximately 94.11%. 2.
Is the average component reliability the system reliability?
Answer: No; all-required independent success uses a product. 3.
Why investigate a shared power supply?
Answer: It can create a common cause that violates the assumed independence.
Related terms
- Reliability
- Survival function
- Redundancy
- Availability
- Conditional probability
- Series system
- common-cause failure
- fault tree
Learn next
Compare series and parallel reliability diagrams, then study common-cause failures and repairable-system availability. Practice defining the mission and distinguishing independent marginal probabilities from conditional stage-success probabilities.
More in Statistics and probability
- Longitudinal or serial data - Longitudinal or serial data contain repeated measurements of a response on the same subject or unit over time.
- Longitudinal study - A longitudinal study follows the same subjects or units over time, collecting observations that reveal change, continuity, and the ordering of events.
- Lot quality assurance sampling - Lot quality assurance sampling, or LQAS, is a sampling approach used to classify a lot, service area, or other defined group against quality or performance...
- Magnitude of effect - Magnitude of effect describes how large a difference or association is, rather than simply whether it is statistically detectable.
- Mail survey - A mail survey collects information through questionnaires sent to selected respondents by postal mail, usually for them to complete and return themselves.
- Main effect - A main effect is the effect of one factor averaged across the levels of other factors in a factorial analysis.
Sources
NIST/SEMATECH, “Series model.” https://www.itl.nist.gov/div898/handbook/apr/section1/apr182.htm — primary reliability guidance on assumptions, products, and hazard rates. Harry W. Jones, “Reliability and Failure in NASA Missions” (2015). https://ntrs.nasa.gov/api/citations/20160001189/downloads/20160001189.pdf — NASA author's account identifying the series-product formula as Lusser's law. The worked calculations and fictional case are original.
Previous: Lucky imaging Next: LZ4 (compression algorithm)