1 Definition and basic properties
The survival function is a central quantity in probability and statistics for describing time until an event occurs. It is most often used with nonnegative random variables such as lifetimes, failure times, or durations. By expressing the chance that the event has not yet occurred by time t, it gives a direct way to summarize persistence over time.
1.1 Formal definition
If T is a random variable representing the time to an event, the survival function is defined by
S(t) = P(T > t).
Here, S(t) gives the probability that the event time exceeds t. In many settings, T may represent death, component failure, recovery, relapse, or another endpoint of interest.
1.2 Relationship to the cumulative distribution function
The survival function is the complement of the cumulative distribution function F(t). Since F(t) = P(T \le t), it follows that
S(t) = 1 - F(t)
for a right-continuous convention. This complementary form makes survival functions especially useful for studying events that have not yet happened.
1.3 Range and boundary values
For typical time-to-event variables with support on nonnegative values, the survival function takes values between 0 and 1. At time zero, S(0) is often 1 if the event cannot occur before time zero. As time increases, S(t) generally decreases toward 0 when the event is certain to occur eventually.
1.4 Monotonicity and continuity
Survival functions are nonincreasing in t, because longer waiting times cannot raise the probability that the event has still not occurred. Their continuity depends on the underlying distribution. Continuous event times often produce continuous survival functions, while discrete times may create stepwise drops at event times.
2 Interpretation in probability and statistics
The survival function is interpreted differently depending on the application, but in each case it represents persistence beyond a threshold time.
2.1 Time-to-event interpretation
In time-to-event analysis, S(t) is the probability that an observed subject has not experienced the event by time t. This interpretation is common in medical studies, where the event may be death, recurrence, or treatment failure, and the focus is on how long the subject remains event-free.
2.2 Reliability interpretation
In reliability engineering, the survival function is often called the reliability function. It measures the probability that a device, machine, or component continues operating beyond time t without failure. This makes it a natural tool for describing product durability and system performance.
2.3 Complementary probability view
From a general probability viewpoint, the survival function provides the complement of event occurrence by a given time. It is therefore a simple way to express “not yet happened” probabilities, especially when comparing multiple groups or studying different waiting-time patterns.
3 Mathematical relationships
The survival function is closely connected to several other functions used in statistical theory and modeling.
3.1 Survival function and density function
For a continuous random variable with density f(t), the survival function can be written as
S(t) = \int_t^\infty f(u)\,du.
Differentiating gives f(t) = -dS(t)/dt when the derivative exists. Thus, the density describes how quickly survival probability is lost at each time point.
3.2 Survival function and hazard function
The hazard function describes the instantaneous event rate among those still event-free. It is related to the survival function by
h(t) = f(t) / S(t)
for continuous distributions where S(t) > 0. The hazard and survival functions together give a fuller picture: one captures instantaneous risk, while the other captures cumulative persistence.
3.3 Cumulative hazard function
The cumulative hazard function H(t) is defined as the accumulated hazard up to time t. It is linked to survival by
S(t) = exp(-H(t))
in the continuous case under suitable conditions. This relationship is widely used because it transforms multiplicative survival behavior into an additive hazard scale.
3.4 Quantile function
The quantile function can be obtained from the survival function by inversion. If p is a probability level, the corresponding time threshold is the point at which survival drops to p. This is useful for median survival and other percentile-based summaries.
4 Types of survival functions
Survival functions vary according to whether the underlying time variable is continuous, discrete, or a mixture of both.
4.1 Continuous survival functions
For continuous event times, the survival function changes smoothly or continuously over time. This is common in physical failure processes and many biological settings, where exact event times can be measured on a continuous scale.
4.2 Discrete survival functions
For discrete event times, survival functions decrease in jumps at specific time points. This occurs when events can only happen at integer times or other fixed intervals, such as in count-based waiting-time models or grouped observation schedules.
4.3 Mixed distributions
Some random variables combine continuous and discrete components. Their survival functions may have both smooth declines and sudden drops. Such mixtures appear when some events occur at exact times while others are spread over an interval.
5 Estimation from data
In practice, the survival function is usually estimated from observed data rather than known exactly.
5.1 Empirical survival function
The empirical survival function is a direct sample-based estimate. It is formed by counting the proportion of observed times that exceed each time point. This method is simple, but it does not handle censoring as effectively as specialized survival estimators.
5.2 Kaplan-Meier estimator
The Kaplan-Meier estimator is the standard nonparametric estimate of the survival function in the presence of censoring. It is a step function that changes value at observed event times and remains constant between them. Because it uses only the information available up to each event time, it is widely applied in medical and reliability studies.
5.3 Actuarial life table methods
Life table methods estimate survival over grouped time intervals rather than exact event times. They are useful when data are recorded in intervals or when only summary counts are available. These methods provide approximate survival estimates and are often used in large-scale demographic or actuarial analyses.
5.4 Parametric estimation
Parametric estimation assumes that survival times follow a specified distribution, such as exponential or Weibull. Model parameters are fitted from data, and the resulting survival function is then computed from the chosen distribution. Parametric approaches can be efficient when the assumed form is reasonable.
6 Statistical models using survival functions
Many statistical models are built directly around the survival function or its associated hazard structure.
6.1 Exponential model
The exponential model assumes a constant hazard over time. Its survival function has the form S(t) = exp(-\lambda t), where \lambda is the rate parameter. This model is mathematically simple and often serves as a baseline case for lifetime analysis.
6.2 Weibull model
The Weibull model generalizes the exponential case by allowing the hazard to increase or decrease with time. Its flexibility makes it useful for reliability data and biological survival processes. Depending on parameter values, it can represent early failures, aging effects, or constant risk.
6.3 Log-normal model
In the log-normal model, the logarithm of survival time follows a normal distribution. The resulting survival function often describes processes with skewed waiting times and nonmonotone hazard patterns. It is frequently used when event times cluster around a characteristic scale.
6.4 Cox proportional hazards model
The Cox proportional hazards model links covariates to survival through the hazard function while leaving the baseline survival unspecified. It is one of the most widely used semiparametric methods in survival analysis. The model is valued for its flexibility and for comparing relative risk across groups.
7 Inference and analysis
Statistical analysis of survival data must account for incomplete observation and other features unique to time-to-event settings.
7.1 Censoring
Censoring occurs when the event time is only partially known. Common forms include right censoring, where the event has not occurred by the end of observation. Proper handling of censoring is essential for unbiased estimation of survival functions.
7.2 Truncation
Truncation arises when some event times are unobserved because of the way data are sampled. Unlike censoring, truncated observations are not merely incomplete; they may be entirely absent from the dataset. This can affect both estimation and interpretation of survival curves.
7.3 Confidence intervals
Confidence intervals quantify uncertainty in estimated survival functions. They are often constructed around Kaplan-Meier estimates or model-based survival curves. These intervals help assess the precision of survival comparisons and support statistical reporting.
7.4 Hypothesis testing
Hypothesis tests in survival analysis examine whether survival patterns differ between groups or whether a model parameter is consistent with a specified value. Common procedures include tests comparing survival curves and tests of covariate effects in regression settings. Such methods support formal inference about time-to-event behavior.
8 Applications
Survival functions appear across many disciplines wherever durations or failure times matter.
8.1 Medical survival analysis
In medicine, survival functions are used to study patient outcomes after diagnosis or treatment. They help summarize time until death, relapse, or recovery and are often displayed as survival curves. These summaries are standard in clinical research and epidemiology.
8.2 Reliability engineering
In reliability engineering, survival functions describe the lifespan of components and systems. They are used to estimate failure probabilities, compare product designs, and plan maintenance schedules. This makes them important in manufacturing, electronics, and industrial testing.
8.3 Actuarial science
Actuarial science uses survival functions to model mortality and other duration-based risks. They support the pricing of life-contingent financial products and the study of longevity patterns. Life tables and related survival estimates are foundational tools in this field.
8.4 Queueing and waiting-time models
Survival functions also appear in queueing theory and waiting-time analysis. They describe how long a customer, packet, or process remains in a system before service or completion. This is useful in operations research, telecommunications, and service design.
9 Extensions and related concepts
Survival analysis has produced several extensions that broaden the use of survival functions in complex settings.
9.1 Competing risks
Competing risks occur when more than one type of event can terminate observation, and the occurrence of one event prevents the others from happening first. In such settings, separate event-specific functions are often studied alongside overall survival. This framework is important when different failure causes must be distinguished.
9.2 Restricted mean survival time
Restricted mean survival time is the expected survival time up to a fixed horizon. It summarizes the area under the survival curve over a chosen interval. This measure is useful when long-term extrapolation is uncertain or when an average time-based summary is preferred.
9.3 Net survival
Net survival seeks to isolate survival from a specific cause by removing other sources of mortality or failure. It is especially relevant in population studies and disease-specific analyses. The concept helps distinguish between overall survival and cause-related survival experience.
9.4 Multistate models
Multistate models generalize survival analysis by allowing individuals to move through several states over time. Rather than focusing on a single endpoint, they track transitions among intermediate and terminal conditions. Survival functions remain useful within this broader framework as summaries of time spent before particular transitions.