NCSS Statistical Software NCSS.

com

Chapter 550

Distribution (Weibull) Fitting

Introduction
This procedure estimates the parameters of the exponential, extreme value, logistic, log-logistic, lognormal,
normal, and Weibull probability distributions by maximum likelihood. It can fit complete, right censored, left
censored, interval censored (readout), and grouped data values. It also computes the nonparametric
Kaplan-Meier and Nelson-Aalen estimates of survival and associated hazard rates. It outputs various
statistics and graphs that are useful in reliability and survival analysis. When the choice of the probability
distribution is in doubt, the procedure helps select an appropriate probability distribution from those
available.
Features of this procedure include:
1. Probability plotting, hazard plotting, and reliability plotting for the common life distributions. The
data may be any combination of complete, right censored, left censored, and interval censored data.
2. Maximum likelihood and probability plot estimates of distribution parameters, percentiles, reliability
(survival) functions, hazard rates, and hazard functions.
3. Confidence intervals for distribution parameters and percentiles.
4. Nonparametric estimates of survival using the Kaplan-Meier procedure.

Overview of Survival and Reliability Analysis

This procedure may be used to conduct either survival analysis or reliability analysis. When a study concerns
a biological event associated with the study of animals (including humans), it is usually called survival
analysis. When a study concerns machines in an industrial setting, it is usually called reliability analysis.
Survival analysis emphasizes a nonparametric estimation approach (Kaplan-Meier estimation), while
reliability analysis emphasizes a parametric approach (Weibull or lognormal estimation). In the rest of this
chapter, when we refer to survival analysis, you can freely substitute ‘reliability’ for ‘survival.’ The two terms
refer to the same type of analysis.
We will give a brief introduction to the subject in this section. For a complete account of survival analysis, we
suggest the book by Klein and Moeschberger (1997).
Survival analysis is the study of the distribution of lifetimes. That is, it is the study of the elapsed time
between an initiating event (birth, start of treatment, diagnosis, or start of operation) and a terminal event
(death, relapse, cure, or machine failure). The data values are a mixture of complete (terminal event
occurred) and censored (terminal event has not occurred) observations. From the data values, the survival
analyst makes statements about the survival distribution of the failure times. This distribution allows
questions about such quantities as survivability, expected lifetime, and mean time to failure to be answered.

550-1
© NCSS, LLC. All Rights Reserved.
NCSS Statistical Software NCSS.com

Distribution (Weibull) Fitting

Let T be the elapsed time until the occurrence of a specified event. The event may be death, occurrence of a
disease, disappearance of a disease, appearance of a tumor, etc. The probability distribution of T may be
specified using one of the following basic functions. Once one of these functions has been specified, the
others may be derived using the mathematical relationships presented.
1. Probability density function, f(t). This is the probability that an event occurs at time t.
2. Cumulative distribution function, F(t). This is the probability that an individual survives until time t.

𝑡𝑡

𝐹𝐹(𝑡𝑡) = � 𝑓𝑓(𝑥𝑥)𝑑𝑑𝑑𝑑
0

3. Survival function, S(t) or Reliability function, R(t). This is the probability that an individual survives
beyond time t. This is usually the first quantity that is studied. It may be estimated using the
nonparametric Kaplan-Meier curve or one of the parametric distribution functions.

𝑅𝑅(𝑡𝑡) = 𝑆𝑆(𝑡𝑡) = � 𝑓𝑓(𝑥𝑥)𝑑𝑑𝑑𝑑 = 1 − 𝐹𝐹(𝑡𝑡)

𝑡𝑡

𝑡𝑡

𝑆𝑆(𝑡𝑡) = exp �− � ℎ(𝑥𝑥)𝑑𝑑𝑑𝑑 � = exp[−𝐻𝐻(𝑡𝑡)]

0

4. Hazard rate, h(t). This is the probability that an individual at time t experiences the event in the next
instant. It is a fundamental quantity in survival analysis. It is also known as the conditional failure
rate in reliability, the force of mortality in demography, the intensity function in stochastic
processes, the age-specific failure rate in epidemiology, and the inverse of Mill’s ratio in economics.
The empirical hazard rate may be used to identify the appropriate probability distribution of a
particular mechanism, since each distribution has a different hazard rate function. Some
distributions have a hazard rate that decreases with time, others have a hazard rate that increases
with time, some are constant, and some exhibit all three behaviors at different points in time.

𝑓𝑓(𝑡𝑡)
ℎ(𝑡𝑡) =
𝑆𝑆(𝑡𝑡)

5. Cumulative hazard function, H(t). This is integral of h(t) from 0 to t.

𝑡𝑡

𝐻𝐻(𝑡𝑡) = � ℎ(𝑥𝑥)𝑑𝑑𝑑𝑑 = − ln[𝑆𝑆(𝑡𝑡)]

0

550-2
© NCSS, LLC. All Rights Reserved.
NCSS Statistical Software NCSS.com

Distribution (Weibull) Fitting

Nonparametric Estimators of Survival

There are two competing nonparametric estimators of the survival distribution, S(t), available in this
procedure. The first is the common Kaplan-Meier Product limit estimator. The second is the Nelson-Aalen
estimator of the cumulative hazard function, H(t).

Kaplan-Meier Product-Limit Estimator

The most common nonparametric estimator of the survival function is called the Kaplan-Meier product limit
estimator. This estimator is defined as follows in the range of time values for which there are data.

1
if 𝑡𝑡 < 𝑡𝑡1
𝑑𝑑
𝑆𝑆̂(𝑡𝑡) = �� �1 − 𝑖𝑖 � if 𝑡𝑡 ≤ 𝑡𝑡
1
𝑌𝑌𝑖𝑖
𝑡𝑡1 ≤𝑡𝑡

In the above equation, 𝑑𝑑𝑖𝑖 represents the number of deaths at time 𝑡𝑡𝑖𝑖 and 𝑌𝑌𝑖𝑖 represents the number of
individuals who are at risk at time 𝑡𝑡𝑖𝑖 .
The variance of S(t) is estimated by Greenwood’s formula

𝑑𝑑𝑖𝑖
𝑉𝑉� �𝑆𝑆̂(𝑡𝑡)� = 𝑆𝑆̂(𝑡𝑡)2 �
𝑌𝑌𝑖𝑖 (𝑌𝑌𝑖𝑖 − 𝑑𝑑𝑖𝑖 )
𝑡𝑡𝑖𝑖 <𝑡𝑡

The product limit estimator may be used to estimate the cumulative hazard function H(t) using the
relationship

� (𝑡𝑡) = − log�𝑆𝑆̂(𝑡𝑡)�
𝐻𝐻

Linear (Greenwood) Confidence Limits

This estimator may be used to create confidence limits for S(t) using the formula

𝑆𝑆̂(𝑡𝑡) ± 𝑧𝑧1−𝛼𝛼⁄2 𝜎𝜎𝑆𝑆 (𝑡𝑡)𝑆𝑆̂(𝑡𝑡)

where

𝑉𝑉� �𝑆𝑆̂(𝑡𝑡)�
𝜎𝜎𝑆𝑆2 (𝑡𝑡) =
𝑆𝑆̂ 2 (𝑡𝑡)

and z is the appropriate value from the standard normal distribution. We call this the Linear
(Greenwood) confidence interval.

550-3
© NCSS, LLC. All Rights Reserved.
NCSS Statistical Software NCSS.com

Distribution (Weibull) Fitting

Log Hazard Confidence Limits

Better confidence limits may be calculated using the logarithmic transformation of the hazard functions.
These limits are

𝑆𝑆̂(𝑡𝑡)1⁄𝜃𝜃 , 𝑆𝑆̂(𝑡𝑡)𝜃𝜃

where

𝑧𝑧1−𝛼𝛼⁄2 𝜎𝜎𝑆𝑆 (𝑡𝑡)

𝜃𝜃 = exp � �
log�𝑆𝑆̂(𝑡𝑡)�

ArcSine-Square Root Hazard Confidence Limits

Another set of confidence limits using an improving transformation is given by the (intimidating) formula

1⁄2
𝑆𝑆̂(𝑡𝑡)
2
sin �max �0, arcsin (𝑆𝑆 ̂ (𝑡𝑡)1⁄2 − 0.5𝑧𝑧1−𝛼𝛼⁄2 𝜎𝜎𝑆𝑆 (𝑡𝑡) � � �� ≤ 𝑆𝑆(𝑡𝑡)
1 − 𝑆𝑆̂(𝑡𝑡)
1⁄2
𝜋𝜋 𝑆𝑆̂(𝑡𝑡)
≤ sin �min � , arcsin(𝑆𝑆̂(𝑡𝑡) + 0.5𝑧𝑧1−𝛼𝛼⁄2 𝜎𝜎𝑆𝑆 (𝑡𝑡) �
2 1 ⁄2
� ��
2 1 − 𝑆𝑆̂(𝑡𝑡)

Nelson-Aalen Hazard Confidence Limits

An alternative estimator of H(t), which has better small sample size properties is the Nelson-Aalen estimator
given by
0 if 𝑡𝑡 < 𝑡𝑡1
� (𝑡𝑡) = �
𝐻𝐻 𝑑𝑑𝑖𝑖
� if 𝑡𝑡 ≤ 𝑡𝑡1
𝑌𝑌𝑖𝑖
𝑡𝑡1 ≤𝑡𝑡

The variance of this estimate is given by the formula

(𝑌𝑌𝑖𝑖 − 𝑑𝑑𝑖𝑖 )𝑑𝑑𝑖𝑖

𝜎𝜎𝐻𝐻2 (𝑡𝑡) = �
(𝑌𝑌𝑖𝑖 − 1)𝑌𝑌𝑖𝑖2
𝑡𝑡𝑖𝑖 ≤𝑡𝑡

The 100(1-alpha)% confidence limits for H(t) are calculated using

� (𝑡𝑡)exp �±𝑧𝑧1−𝛼𝛼⁄2 𝜎𝜎𝐻𝐻 (𝑡𝑡) ∕ 𝐻𝐻

𝐻𝐻 � (𝑡𝑡)�

This hazard function may be used to generate the Nelson-Aalen estimator of S(t) using the formula

𝑆𝑆̃(𝑡𝑡) = 𝑒𝑒 𝐻𝐻�(𝑡𝑡)

Using these formulas, a fourth set of confidence limits for S(t) may be calculated as

� (𝑡𝑡) ± 𝑧𝑧1−𝛼𝛼⁄2 𝜎𝜎𝐻𝐻 (𝑡𝑡)�

exp�𝐻𝐻

550-4
© NCSS, LLC. All Rights Reserved.
NCSS Statistical Software NCSS.com

Distribution (Weibull) Fitting

Parametric Survival Distributions

This section presents the parametric probability distributions that may be analyzed with this procedure.

Normal Distribution
The normal distribution is one of the most commonly used in statistics. However, it is used infrequently as a
lifetime distribution because it allows negative values while lifetimes are always positive. It has been found
that the logarithms of failure times may be fit by the normal distribution. Hence the lognormal has become
a popular distribution in reliability work, while the normal has been put on the sideline.
The normal distribution is indexed by a location (M) and a scale (S) parameter. A threshold parameter is
meaningless in this case, since it is an adjustment to the location parameter. Using these symbols, the
normal density function may be written as

1 1 𝑡𝑡−𝑀𝑀 2
− � �
𝑓𝑓(𝑡𝑡|𝑀𝑀, 𝑆𝑆) = 𝑒𝑒 2 𝑆𝑆 , − ∞ < 𝑀𝑀 < ∞, 𝑆𝑆 > 0, −∞ < 𝑡𝑡 < ∞
𝑆𝑆√2𝜋𝜋

Location Parameter - M
The location parameter of the normal distribution is often called the mean.

Scale Parameter - S
The scale parameter of the normal distribution is usually called the standard deviation.

Lognormal Distribution
The normal distribution is one of the most commonly used in statistics. Although the normal distribution
itself does not often work well with time-to-failure data, it has been found that the logarithm of failure time
often does. Hence the lognormal has become a popular distribution in reliability work.
The lognormal distribution is indexed by a shape (S) , a scale (M), and a threshold (D) parameter. Using these
symbols, the three parameter lognormal density function may be written as

2
1 1 ln(𝑡𝑡−𝐷𝐷)−𝑀𝑀
− � �
𝑓𝑓(𝑡𝑡|𝑀𝑀, 𝑆𝑆, 𝐷𝐷) = 𝑒𝑒 2 𝑆𝑆 , −∞ < 𝑀𝑀 < ∞, 𝑆𝑆 > 0, −∞ < 𝐷𝐷 < ∞, 𝑡𝑡 > 𝐷𝐷
(𝑡𝑡 − 𝐷𝐷)𝑆𝑆√2𝜋𝜋

It is often more convenient to work with logarithms to the base 10 (denoted by log) rather than logarithms
to the base e (denoted by ln). The lognormal10 density function is written as

2
1 1 log(𝑡𝑡−𝐷𝐷)−𝑀𝑀
− � �
𝑓𝑓(𝑡𝑡|𝑀𝑀, 𝑆𝑆, 𝐷𝐷) = 𝑒𝑒 2 𝑆𝑆 , −∞ < 𝑀𝑀 < ∞, 𝑆𝑆 > 0, −∞ < 𝐷𝐷 < ∞, 𝑡𝑡 > 𝐷𝐷
ln(10)(𝑡𝑡 − 𝐷𝐷)𝑆𝑆√2𝜋𝜋

550-5
© NCSS, LLC. All Rights Reserved.
NCSS Statistical Software NCSS.com

Distribution (Weibull) Fitting

Shape Parameter - S
The shape parameter of the lognormal distribution of t is the standard deviation in the normal distribution
of ln(t-D) or log(t-D). That is, the scale parameter of the normal distribution is the shape parameter of the
lognormal distribution.

Scale Parameter - M
The scale parameter of the lognormal distribution t is the mean in the normal distribution of ln(t-D) or log(t-
D). That is, the location parameter of the normal distribution is the scale parameter of the lognormal
distribution.

Threshold Parameter - D
The threshold parameter is the minimum value of the random variable t. When D is set to zero, we obtain
the two parameter lognormal distribution. When a positive value is given to D, we are inferring that no
failures can occur between zero and D.

Reliability Function
The reliability (or survivorship) function, R(t), gives the probability of surviving beyond time t. For the
Lognormal distribution, the reliability function is

ln(𝑡𝑡 − 𝐷𝐷) − 𝑀𝑀
𝑅𝑅(𝑡𝑡) = 1 − Φ � �
𝑆𝑆

where Φ(𝑧𝑧) is the standard normal distribution function.

The conditional reliability function, R(t,T), may also be of interest. This is the reliability of an item given that it
has not failed by time T. The formula for the conditional reliability is

𝑅𝑅(𝑇𝑇 + 𝑡𝑡)
𝑅𝑅(𝑡𝑡) =
𝑅𝑅(𝑇𝑇)

Hazard Function
The hazard function represents the instantaneous failure rate. For this distribution, the hazard function is

𝑓𝑓(𝑡𝑡)
ℎ(𝑡𝑡) =
𝑅𝑅(𝑡𝑡)

550-6
© NCSS, LLC. All Rights Reserved.
NCSS Statistical Software NCSS.com

Distribution (Weibull) Fitting

Weibull Distribution
The Weibull distribution is named for Professor Waloddi Weibull whose papers led to the wide use of the
distribution. He demonstrated that the Weibull distribution fit many different datasets and gave good
results, even for small samples. The Weibull distribution has found wide use in industrial fields where it is
used to model time to failure data.
The three parameter Weibull distribution is indexed by a shape (B), a scale (C), and a threshold (D)
parameter. Using these symbols, the three parameter Weibull density function may be written as

𝐵𝐵 𝑡𝑡 − 𝐷𝐷 (𝐵𝐵−1) −�𝑡𝑡−𝐷𝐷�𝐵𝐵
𝑓𝑓(𝑡𝑡|𝐵𝐵, 𝐶𝐶, 𝐷𝐷) = � � 𝑒𝑒 𝐶𝐶 , 𝐵𝐵 > 0, 𝐶𝐶 > 0, −∞ < 𝐷𝐷 < ∞, 𝑡𝑡 > 𝐷𝐷.
C 𝐶𝐶

The symbol t represents the random variable (usually elapsed time). The threshold parameter D represents
the minimum value of t that can occur. Setting the threshold to zero results in the common, two parameter
Weibull distribution.

Shape Parameter - B
The shape (or power) parameter controls the overall shape of the density function. Typically, this value
ranges between 0.5 and 8.0. The estimated standard errors and confidence limits displayed by the program
are only valid when B > 2.0.
One of the reasons for the popularity of the Weibull distribution is that it includes other useful distributions
as special cases or close approximations. For example, if
B=1 The Weibull distribution is identical to the exponential distribution.
B=2 The Weibull distribution is identical to the Rayleigh distribution.
B = 2.5 The Weibull distribution approximates the lognormal distribution.
B = 3.6 The Weibull distribution approximates the normal distribution.

Scale Parameter - C
The scale parameter only changes the scale of the density function along the time axis. Hence, a change in
this parameter has the same effect on the distribution as a change in the scale of time—for example, from
days to months or from hours to days. However, it does not change the actual shape of the distribution.
C is known as the characteristic life. No matter what the shape, 63.2% of the population fails by t = C+D.
Some authors use 1/C instead of C as the scale parameter. Although this is arbitrary, we prefer dividing by
the scale parameter since that is how you usually scale a set of numbers. For example, remember how you
create a z-score when dealing with the normal data or create a percentage by dividing by the maximum.

Threshold Parameter - D
The threshold parameter is the minimum value of the random variable t. Often, this parameter is referred
to as the location parameter. We use ‘threshold’ rather than ‘location’ to stress that this parameter sets the
minimum time. We reserve ‘location’ to represent the center of the distribution. This is a fine point and we
are not upset when people refer to this as the location parameter.

550-7
© NCSS, LLC. All Rights Reserved.
NCSS Statistical Software NCSS.com

Distribution (Weibull) Fitting

When D is set to zero, we obtain the two parameter Weibull distribution. It is possible, but unusual, for D to
have a negative value. When using a search algorithm to find the estimated value of D, a nonzero value will
almost certainly be found. However, you should decide physically if a zero probability of failure in the
interval between 0 and D is truly warranted.
A downward or upward sloping tail on the Weibull probability plot or values of B > 6.0 are indications that a
nonzero threshold parameter will produce a better fit to your data.
Negative values of D represent an amount of time that has been subtracted from the actual times. On the
other hand, positive values of D represent a period of time between the starting point and when any failures
can occur. For example, positive values of D may represent the amount of time between the production of
an item and when it is placed in service.

Relationship to the Extreme Value Distribution

The extreme value distribution is directly related to the Weibull distribution. If x = ln(t) and t follows the
Weibull distribution, x follows the extreme value distribution.

Reliability Function
The reliability (or survivorship) function, R(t), gives the probability of surviving beyond time t. For the Weibull
distribution, the reliability function is
𝑡𝑡−𝐷𝐷 𝐵𝐵
−� �
𝑅𝑅(𝑡𝑡) = 𝑒𝑒 𝐶𝐶

The reliability function is one minus the cumulative distribution function. That is,

𝑅𝑅(𝑡𝑡) = 1 − 𝐹𝐹(𝑡𝑡)

The conditional reliability function, R(t,T), may also be of interest. This is the reliability of an item given that it
has not failed by time T. The formula for the conditional reliability is

𝑅𝑅(𝑇𝑇 + 𝑡𝑡)
𝑅𝑅(𝑡𝑡) =
𝑅𝑅(𝑇𝑇)

Hazard Function
The hazard function represents the instantaneous failure rate. For this distribution, the hazard function is

𝑓𝑓(𝑡𝑡) 𝐵𝐵 𝑡𝑡 − 𝐷𝐷 𝐵𝐵−1
ℎ(𝑡𝑡) = = � �
𝑅𝑅(𝑡𝑡) 𝐶𝐶 𝐶𝐶

Depending on the values of the distribution’s parameters, the Weibull’s hazard function can be decreasing
(when B<1), constant (when B=1 at 1/C), or increasing (when B>1) over time.

550-8
© NCSS, LLC. All Rights Reserved.
NCSS Statistical Software NCSS.com

Distribution (Weibull) Fitting

Extreme Value Distribution

The extreme value distribution is occasionally used to model lifetime data. It is included here because of its
close relationship to the Weibull distribution. It turns out that if t is distributed as a Weibull random variable,
then ln(t) is distributed as the extreme value distribution.
The density of the extreme value distribution may be written as

1 𝑡𝑡 − 𝑀𝑀 𝑡𝑡 − 𝑀𝑀
𝑓𝑓(𝑡𝑡|𝑀𝑀, 𝑆𝑆) = exp � � exp �−exp � �� , 𝑆𝑆 > 0
𝑆𝑆 𝑆𝑆 𝑆𝑆

Exponential Distribution
The exponential distribution was one of the first distributions used to model lifetime data. It has now been
superceded by the Weibull distribution, but is still used occasionally. The exponential distribution may be
found from the Weibull distribution by setting B = 1.
The exponential distribution is a model for the life of products with a constant failure rate. The two
parameter exponential distribution is indexed by both a scale and a threshold parameter. The density of the
exponential distribution may be written as

1 𝑡𝑡 − 𝐷𝐷
𝑓𝑓(𝑡𝑡|𝑆𝑆, 𝐷𝐷) = exp �− �, 𝑆𝑆 > 0
𝑆𝑆 𝑆𝑆

Scale Parameter - S
The scale parameter changes the scale of the density function along the time axis. Hence, a change in this
parameter has the same effect on the distribution as a change in the scale of time—for example, from days
to months or from hours to days. However, it does not change the actual shape of the distribution.
Some authors use 1/S instead of S as the scale parameter. Although this is arbitrary, we prefer dividing by
the scale parameter since that is how a set of numbers is usually scaled. For example, remember how z-
scores are created when dealing with the normal distribution.

Threshold Parameter - D
The threshold parameter is the minimum value of the random variable t. Often, this parameter is referred
to as the location parameter. We use ‘threshold’ rather than ‘location’ to stress that this parameter sets the
minimum time. We reserve ‘location’ to represent the center of the distribution. This is a fine point and we
are not upset when people refer to this as the location parameter.
When D is set to zero, we obtain the two parameter exponential distribution. It is possible, but unusual, for
D to have a negative value. When using a search algorithm to find the estimated value of D, a nonzero value
will almost certainly be found. However, you should decide physically if a zero probability of failure in the
interval between 0 and D is truly warranted.
A downward or upward sloping tail on the exponential probability plot is an indication that a nonzero
threshold parameter will produce a better fit to your data.

550-9
© NCSS, LLC. All Rights Reserved.
NCSS Statistical Software NCSS.com

Distribution (Weibull) Fitting

Negative values of D represent an amount of time that has been subtracted from the actual times. On the
other hand, positive values of D represent a period of time between the starting point and when any failures
can occur. For example, positive values of D may represent the amount of time between the production of
an item and when it is placed in service.

Reliability Function
The reliability (or survivorship) function, R(t), gives the probability of surviving beyond time t. For the
exponential distribution, the reliability function is

𝑡𝑡−𝐷𝐷
−� �
𝑅𝑅(𝑡𝑡) = 𝑒𝑒 𝑆𝑆

Hazard Function
The hazard function represents the instantaneous failure rate. For this distribution, the hazard function is
constant. The rate is given by the function

𝑓𝑓(𝑡𝑡) 1
ℎ(𝑡𝑡) = =
𝑅𝑅(𝑡𝑡) 𝑆𝑆

Logistic Distribution
The density of the logistic distribution may be written as

𝑡𝑡 − 𝑀𝑀
exp � 𝑆𝑆 �
𝑓𝑓(𝑡𝑡|𝑀𝑀, 𝑆𝑆) = , 𝑆𝑆 > 0
𝑡𝑡 − 𝑀𝑀 2
𝑆𝑆 �1 + exp � 𝑆𝑆 ��

Log-logistic Distribution
The density of the log-logistic distribution may be written as

ln(𝑡𝑡 − 𝐷𝐷) − 𝑀𝑀
exp � 𝑆𝑆
�
𝑓𝑓(𝑡𝑡|𝑀𝑀, 𝑆𝑆, 𝐷𝐷) = 2, 𝑆𝑆 > 0
ln(𝑡𝑡 − 𝐷𝐷) − 𝑀𝑀
(𝑡𝑡 − 𝐷𝐷)𝑆𝑆 �1 + exp � ��
𝑆𝑆

The log-logistic distribution is used occasionally to model lifetime data, but it is so similar to the lognormal
and the Weibull distributions that it adds little and is thus often dropped from consideration.

550-10
© NCSS, LLC. All Rights Reserved.
NCSS Statistical Software NCSS.com

Distribution (Weibull) Fitting

Parameter Estimation
The parameters of the reliability distributions may be estimated by maximum likelihood or by applying least
squares regression to the probability plot. The probability plot method is popular because it uses a nice
graphic (the probability plot) which allows a visual analysis of the goodness of fit of the distribution to the
data. Maximum likelihood estimation is usually favored by statisticians because it has been shown to be
optimum in most situations and because it provides estimates of standard errors and confidence limits.
However, there are situations in which maximum likelihood does not do as well as the regression approach.
For example, maximum likelihood does not do a good job of estimating the threshold parameter. When you
want to include the threshold parameter in your model, we suggest that you use the regression approach to
estimate it and then treat the threshold value as a known quantity in the maximum likelihood estimation.

Maximum Likelihood
Maximum likelihood estimation consists of finding the values of the distribution parameters that maximize
the log-likelihood of the data values. Loosely speaking, these are the values of the parameters which
maximize the probability that the current set of data values occur.
The general form of the log-likelihood function is given by

𝐿𝐿�𝑃𝑃� = � ln �𝑓𝑓�𝑃𝑃, 𝑡𝑡𝑘𝑘 �� + � ln �𝑆𝑆�𝑃𝑃, 𝑡𝑡𝑘𝑘 �� + � ln �𝐹𝐹�𝑃𝑃, 𝑡𝑡𝑘𝑘 �� + � ln �𝑓𝑓�𝑃𝑃, 𝑡𝑡𝑢𝑢𝑢𝑢 � − 𝑓𝑓�𝑃𝑃, 𝑡𝑡𝑙𝑙𝑙𝑙 ��
𝐹𝐹 𝑅𝑅 𝐿𝐿 𝐼𝐼

where F stands for the set of failed items, R stands for the set of right censored items, L stands for the set of
left censored items, and I stands for the set of interval censored items. In the case of interval censored
observations, 𝑡𝑡𝑙𝑙𝑙𝑙 represents the first time of the interval and 𝑡𝑡𝑢𝑢𝑢𝑢 represents the last time of the interval.
Also, P represents one or two parameters as the case may be.
L(P) is maximized using two numerical procedures. First, a recently developed method called differential
evolution is used. This is a robust maximization procedure that only requires evaluations of the function, but
not its derivatives. The solution of the differential evolution phase is then used as starting values for the
Newton-Raphson algorithm. Newton-Raphson is used because it provides an estimate of the variance-
covariance matrix of the parameters, which is needed in computing confidence limits. The amount of
emphasis on the differential evolution phase as opposed to the Newton-Raphson phase may be controlled
using the maximum number of iterations allowed for each. Numerical differentiation is used to compute the
first and second order derivatives that are needed by the Newton-Raphson procedure.

550-11
© NCSS, LLC. All Rights Reserved.
NCSS Statistical Software NCSS.com

Distribution (Weibull) Fitting

Probability Plot – F(t) Calculation Method

This option specifies the method used to determine F(t), which influences the probability plot estimates of
the parameters and is used to calculate the vertical plotting positions of points in the probability plot (the
probability plot shows time (t) on the vertical axis and the distribution (normal, beta, Weibull, etc.) quantile
on the horizontal axis).
The five calculation options are

• Median (Approximate) (F(tⱼ) = [j - 0.3]/[n + 0.4])

The most popular method is to calculate the median rank for each sorted data value. This is the median
rank of the jth sorted time value out of n values. Since the median rank requires extensive calculations,
this approximation to the median rank is often used.

𝑗𝑗 − 0.3
𝐹𝐹�𝑡𝑡𝑗𝑗 � =
𝑛𝑛 + 0.4

• Median (Exact) (F(tⱼ) = 1/[1 + F(0.5, 2[n-j+1], 2j)×[n-j+1]/j])

The most popular method is to calculate the median rank for each sorted data value. This is the median
rank of the jth sorted time value out of n values. The exact value of the median rank is calculated using
the formula

1
𝐹𝐹�𝑡𝑡𝑗𝑗 � =
𝑛𝑛 − 𝑗𝑗 + 1
1+� � 𝐹𝐹0.5,2(𝑛𝑛−𝑗𝑗+1),2𝑗𝑗
𝑗𝑗

• Mean (F(tⱼ) = j/[n + 1])

The mean rank is sometimes recommended. In this case, the formula is

𝑗𝑗
𝐹𝐹�𝑡𝑡𝑗𝑗 � =
𝑛𝑛 + 1

• White's Formula (F(tⱼ) = [j - 3/8]/[n + 1/4])

A formula proposed by White is sometimes recommended. The formula is

𝑗𝑗 − 3⁄8
𝐹𝐹�𝑡𝑡𝑗𝑗 � =
𝑛𝑛 + 1⁄4

• F(tⱼ) = [j - 0.5]/n
The following formula is sometimes used

𝑗𝑗 − 0.5
𝐹𝐹�𝑡𝑡𝑗𝑗 � =
𝑛𝑛

550-12
© NCSS, LLC. All Rights Reserved.
NCSS Statistical Software NCSS.com

Distribution (Weibull) Fitting

Data Structure
Survival data is somewhat more difficult to enter because of the presence of various types of censoring.

Failed or Complete
A failed observation is one in which the time until the terminal event was measured exactly; for example,
the machine stopped working or the mouse died of the disease being studied.

Right Censored
A right censored observation provides a lower bound for the actual failure time. All that is known is that the
failure occurred (or will occur) at some point after the given time value. Right censored observations occur
when a study is terminated before all items have failed. They also occur when an item fails due to an event
other than the one of interest.

Left Censored
A left censored observation provides an upper bound for the actual failure time. All we know is that the
failure occurred at some point before the time value. Left censoring occurs when the items are not checked
for failure until sometime after the study has begun. When a failed item is found, we do not know exactly
when it failed, only that it was at some point before the left censor time.

Interval Censored or Readout

An interval censored observation is one in which we know that the failure occurred between two time
values, but we do not know exactly when. This type of data is often called readout data. It occurs in
situations where items are checked periodically for failures.

Sample Dataset
Most data sets require two (and often three) variables: the failure time variable, an optional censor variable
indicating the type of censoring, and an optional count variable which gives the number of items occurring
at that time. If the censor variable is omitted, all time values represent failed items. If the count variable is
omitted, all counts are assumed to be one.
The table below shows the results of a study to test the failure rate of a particular machine. This particular
experiment began with 30 items being tested. After the twelfth item failed at 152.7 hours, the experiment
was stopped. The remaining eighteen observations were right censored. That is, we know that they will fail
at some time in the future. These data are contained in the WEIBULL database.

550-13
© NCSS, LLC. All Rights Reserved.
NCSS Statistical Software NCSS.com

Distribution (Weibull) Fitting

Weibull Dataset

Time Censor Count

12.5 1 1
24.4 1 1
58.2 1 1
68.0 1 1
69.1 1 1
95.5 1 1
96.6 1 1
97.0 1 1
114.2 1 1
123.2 1 1
125.6 1 1
152.7 1 1
152.7 0 18

550-14
© NCSS, LLC. All Rights Reserved.
NCSS Statistical Software NCSS.com

Distribution (Weibull) Fitting

Example 1 – Fitting a Weibull Distribution

This section presents an example of how to fit the Weibull distribution. The data used were shown above
and are found in the Weibull dataset.

Setup
To run this example, complete the following steps:

1 Open the Weibull example dataset

• From the File menu of the NCSS Data window, select Open Example Data.
• Select Weibull and click OK.

2 Specify the Distribution (Weibull) Fitting procedure options

• Find and open the Distribution (Weibull) Fitting procedure using the menus or the Procedure
Navigator.
• The settings for this example are listed below and are stored in the Example 1 settings file. To load
these settings to the procedure window, click Open Example Settings File in the Help Center or File
menu.

Variables Tab
_________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________

Time Variable................................................... Time

Frequency Variable.......................................... Count
Censor Variable ............................................... Censor

Options Tab
_________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________

Random Seed .................................................. 4002922 (for reproducibility)

Derivatives ....................................................... 0.00006

Plots Tab
_________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________

Hazard Function Plot ....................................... Checked

Hazard Rate Plot ............................................. Checked

Survival/Reliability Plot Format (Click the Button)

___________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________

Confidence Limits (Distribution Fit Line) .......... Checked

Hazard Function Plot Format (Click the Button)

___________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________

Confidence Limits (Distribution Fit Line) .......... Checked

Hazard Rate Plot Format (Click the Button)

___________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________

Confidence Limits (Distribution Fit Line) .......... Checked

3 Run the procedure

• Click the Run button to perform the calculations and generate the output.

550-15
© NCSS, LLC. All Rights Reserved.
NCSS Statistical Software NCSS.com

Distribution (Weibull) Fitting

Data Summary

Data Summary
────────────────────────────────────────────────────────────────────────
Type of
Observation Rows Count Percent (%) Minimum Maximum
────────────────────────────────────────────────────────────────────────────────────────────────────
Failed 12 12 40% 12.5 152.7
Right Censored 1 18 60% 152.7 152.7
Left Censored 0 0
Interval Censored 0 0
Total 13 30 100% 12.5 152.7
────────────────────────────────────────────────────────────────────────

This report displays a summary of the amount of data that were analyzed. Scan this report to determine if
there were any obvious data errors by double checking the counts and the minimum and maximum.

Parameter Estimation

Weibull Parameter Estimation

────────────────────────────────────────────────────────────────────────
Maximum Likelihood (MLE)
───────────────────────────────────
95% Confidence
Probability Interval Limits
Plot Standard ────────────────
Parameter Estimate* Estimate Error Lower Upper
─────────────────────────────────────────────────────────────────────────────────────────────────────────
B (Shape) 1.26829 1.511543 0.4130418 0.8847539 2.58237
C (Scale) 279.7478 238.3481 57.24827 148.8551 381.6452
D (Threshold) 0 0
Log-Likelihood -80.05649
Mean 259.7101 214.9709
Median 209.5383 187.0276
Mode 82.19741 116.3898
Sigma 206.2112 144.9315
─────────────────────────────────────────────────────────────────────────────────────────────────────────

Estimation Details
──────────────────────────────────────────────────
Differential Evolution Iterations 39
Newton-Raphson Restart 1
Newton-Raphson Iterations 10
User-Entered Random Seed 4002922
──────────────────────────────────────────────────
────────────────────────────────────────────────────────────────────────
* Probability plot estimates were generated with F(t) calculated using the approximate median and using the model Time = A +
B(F).

This report displays parameter estimates along with standard errors and confidence limits for the maximum
likelihood estimates. In this example, we have set the threshold parameter to zero so we are fitting the two-
parameter Weibull distribution.

550-16
© NCSS, LLC. All Rights Reserved.
NCSS Statistical Software NCSS.com

Distribution (Weibull) Fitting

Probability Plot Estimate

This estimation procedure uses the data from the probability plot to estimate the parameters. The
estimation formula depends on which option was selected for the Prob Plot Model (in the Estimation tab
window).

Probability Plot Model: F = A + B(Time)

The cumulative distribution function F(t)

𝑡𝑡−𝐷𝐷 𝐵𝐵
−� �
𝐹𝐹(𝑡𝑡) = 1 − 𝑒𝑒 𝐶𝐶

may be rearranged as (assuming D is zero)

ln �−ln�1 − 𝐹𝐹(𝑡𝑡)�� = −𝐵𝐵[ln(𝐶𝐶)] + 𝐵𝐵[ln(𝑡𝑡)]

This is now in a linear form. If we let y = ln(-ln(1-F(t))) and x = ln(t), the above equation becomes

𝑦𝑦 = −𝐵𝐵[ln(𝐶𝐶)] + 𝐵𝐵𝐵𝐵

Using simple linear regression, we can estimate the intercept and slope. Using these estimates, we obtain
estimates of the Weibull parameters B and C as

𝐵𝐵 = slope

and

−intercept
𝐶𝐶 = exp � �
𝐵𝐵

We assumed that D was zero. If D is not zero, it is treated as a known value and subtracted from the time
values before the above regression is calculated.

Probability Plot Model: Time = A + B(F)

The cumulative distribution function F(t)

𝑡𝑡−𝐷𝐷 𝐵𝐵
−� �
𝐹𝐹(𝑡𝑡) = 1 − 𝑒𝑒 𝐶𝐶

may be rearranged as (assuming D is zero)

1
� � ln �−ln�1 − 𝐹𝐹(𝑡𝑡)�� + ln(𝐶𝐶) = ln(𝑡𝑡)
𝐵𝐵
This is now in a linear form. If we let x = ln(-ln(1-F(t))) and y = ln(t), the above equation becomes

1
𝑦𝑦 = 𝑥𝑥 + ln(𝐶𝐶)
𝐵𝐵

550-17
© NCSS, LLC. All Rights Reserved.
NCSS Statistical Software NCSS.com

Distribution (Weibull) Fitting

Using simple linear regression, we can estimate the intercept and slope. Using these estimates, we obtain
estimates of the Weibull parameters B and C as

1
𝐵𝐵 =
slope

and

𝐶𝐶 = exp(intercept)

Parameter estimates for the other distributions are found in a similar manner.

Maximum Likelihood Estimates of B, C, M, and S

These are the usual maximum likelihood estimates (MLE) of the parameters. The formulas for the standard
errors and confidence limits use the estimated variance covariance matrix, which is the inverse of the Fisher
information matrix, �𝑣𝑣𝑣𝑣𝑖𝑖,𝑗𝑗 �. The standard errors are given as the square roots of the diagonal elements 𝑣𝑣𝑣𝑣1,1
and 𝑣𝑣𝑣𝑣2,2.
In the case of the Weibull distribution, the confidence limits for B are

𝐵𝐵�
𝐵𝐵�𝑙𝑙𝑙𝑙𝑙𝑙𝑙𝑙𝑙𝑙,1−𝛼𝛼⁄2 =
𝑧𝑧1−𝛼𝛼⁄2 �𝑣𝑣𝑣𝑣1,1
exp � �
𝐵𝐵�

𝑧𝑧1−𝛼𝛼⁄2 �𝑣𝑣𝑣𝑣1,1
𝐵𝐵�𝑢𝑢𝑢𝑢𝑢𝑢𝑢𝑢𝑢𝑢,1−𝛼𝛼⁄2 = 𝐵𝐵�exp � �
𝐵𝐵�

In the case of the Weibull distribution, the confidence limits for C are

𝐶𝐶̂
𝐶𝐶̂𝑙𝑙𝑙𝑙𝑙𝑙𝑙𝑙𝑙𝑙,1−𝛼𝛼⁄2 =
𝑧𝑧1−𝛼𝛼⁄2 �𝑣𝑣𝑣𝑣2,2
exp � �
𝐶𝐶̂

𝑧𝑧1−𝛼𝛼⁄2 �𝑣𝑣𝑣𝑣2,2
𝐶𝐶̂𝑢𝑢𝑢𝑢𝑢𝑢𝑢𝑢𝑢𝑢,1−𝛼𝛼⁄2 = 𝐶𝐶̂ exp � �
𝐶𝐶̂

In the case of all other distributions, the confidence limits for M are

�𝑙𝑙𝑙𝑙𝑙𝑙𝑙𝑙𝑙𝑙,1−𝛼𝛼⁄2 = 𝑀𝑀
𝑀𝑀 � − 𝑧𝑧1−𝛼𝛼⁄2 �𝑣𝑣𝑣𝑣1,1

�𝑢𝑢𝑢𝑢𝑢𝑢𝑢𝑢𝑢𝑢,1−𝛼𝛼⁄2 = 𝑀𝑀
𝑀𝑀 � + 𝑧𝑧1−𝛼𝛼⁄2 �𝑣𝑣𝑣𝑣1,1

550-18
© NCSS, LLC. All Rights Reserved.
NCSS Statistical Software NCSS.com

Distribution (Weibull) Fitting

In the case of all other distributions, the confidence limits for S are

𝑆𝑆̂
𝑆𝑆̂𝑙𝑙𝑙𝑙𝑙𝑙𝑙𝑙𝑙𝑙,1−𝛼𝛼⁄2 =
𝑧𝑧1−𝛼𝛼⁄2 �𝑣𝑣𝑣𝑣2,2
exp � �
𝑆𝑆̂

𝑧𝑧1−𝛼𝛼⁄2 �𝑣𝑣𝑣𝑣2,2
𝑆𝑆̂𝑢𝑢𝑢𝑢𝑢𝑢𝑢𝑢𝑢𝑢,1−𝛼𝛼⁄2 = 𝑆𝑆̂exp � �
𝑆𝑆̂

Log-Likelihood
This is the value of the log-likelihood function calculated using the maximum likelihood parameter
estimates. This is the value of the function being maximized. It is often used as a goodness-of-fit statistic.
You can compare the log-likelihood values achieved by each distribution and select as the best fitting the
one with the maximum value.
Note that we have found that several popular statistical programs calculate this value without including all
of the terms. Hence, they present erroneous values. The error occurs because they omit the 1/t term in the
denominator of the lognormal and the Weibull log-likelihood functions. Also, they may fail to include a
correction for using the logarithm to the base 10 in the lognormal10. Hopefully, future editions of these
programs will calculate the likelihood correctly.

Mean
This is the mean time to failure (MTTF). It is the mean of the random variable (failure time) being studied
given that the fitted distribution provides a reasonable approximation to your data’s actual distribution. In
the case of the Weibull distribution, the formula for the mean is

1
𝑀𝑀𝑀𝑀𝑀𝑀𝑀𝑀 = 𝐷𝐷 + 𝐶𝐶Γ �1 + �
𝐵𝐵

where Γ(𝑥𝑥) is the gamma function.

Median
The median is the value of t where F(t)=0.5. In the case of the Weibull distribution, the formula for the
median is

𝑀𝑀𝑀𝑀𝑀𝑀𝑀𝑀𝑀𝑀𝑀𝑀 = 𝐷𝐷 + 𝐶𝐶(log 2)1⁄𝐵𝐵

Mode
The mode of the Weibull distribution is given by

1 1⁄𝐵𝐵
𝑀𝑀𝑀𝑀𝑀𝑀𝑀𝑀 = 𝐷𝐷 + 𝐶𝐶 �1 − �
𝐵𝐵

550-19
© NCSS, LLC. All Rights Reserved.
NCSS Statistical Software NCSS.com

Distribution (Weibull) Fitting

Sigma
This is the standard deviation of the failure time. The formula for the standard deviation (sigma) of a Weibull
random variable is

2 1
𝜎𝜎 = 𝐶𝐶 �Γ �1 + � − Γ 2 �1 + �
𝐵𝐵 𝐵𝐵

where Γ(𝑥𝑥) is the gamma function.

Differential Evolution Iterations

This is the number of iterations used in the differential evolution phase of the maximum likelihood
algorithm. If this value is equal to the maximum number of generations allowed, the algorithm did not
converge, so you should increase the maximum number of generations and re-run the procedure.

Newton Raphson Restarts

This is the number of times the Newton Raphson phase of the maximum likelihood algorithm was restarted
because the algorithm diverged. Make sure that the maximum number of restarts was not reached.

Newton Raphson Iterations

This is the number of iterations used in the Newton Raphson phase of the maximum likelihood algorithm. If
this value is equal to the maximum number of iterations allowed, the algorithm did not converge, so you
should increase the maximum number of iterations and re-run the procedure.

Variance-Covariance Matrix

Weibull Variance-Covariance Matrix

────────────────────────────────────────────────────────────────────────
Parameter
─────────────────
Parameter B (Shape) C (Scale)
─────────────────────────────────────────────────
B (Shape) 0.1706035 -14.33368
C (Scale) -14.33368 3277.364
────────────────────────────────────────────────────────────────────────

This table gives the inverse of the Fisher information matrix evaluated at the maximum likelihood estimates
which is an asymptotic estimate of the variance-covariance matrix of the two parameters. These values are
calculated using numerical second-order derivatives.
Note that because these are numerical derivatives based on a random start provided by differential
evolution, the values of the last two decimal places may vary from run to run. You can stabilize the values by
changing the value of Derivatives constant, but this will have little effect on the overall accuracy of your
results.

550-20
© NCSS, LLC. All Rights Reserved.
NCSS Statistical Software NCSS.com

Distribution (Weibull) Fitting

Kaplan-Meier Product-Limit Survival Distribution

Kaplan-Meier Product-Limit Survival Distribution

────────────────────────────────────────────────────────────────────────
Confidence Limits Method: Linear (Greenwood)
────────────────────────────────────────────────────────────────────────
Product-Limit Survival Hazard Function
───────────────────── ─────────────────────
95% Confidence 95% Confidence
Interval Limits Interval Limits
Failure ──────────── ──────────── Sample
Time Estimate Lower Upper Estimate Lower Upper Size
────────────────────────────────────────────────────────────────────────────────────────────────────────────
12.5 0.9667 0.9024 1.0000 0.0339 0.0000 0.1027 30
24.4 0.9333 0.8441 1.0000 0.0690 0.0000 0.1695 29
58.2 0.9000 0.7926 1.0000 0.1054 0.0000 0.2324 28
68.0 0.8667 0.7450 0.9883 0.1431 0.0118 0.2943 27
69.1 0.8333 0.7000 0.9667 0.1823 0.0339 0.3567 26
95.5 0.8000 0.6569 0.9431 0.2231 0.0585 0.4203 25
96.6 0.7667 0.6153 0.9180 0.2657 0.0855 0.4856 24
97.0 0.7333 0.5751 0.8916 0.3102 0.1148 0.5532 23
114.2 0.7000 0.5360 0.8640 0.3567 0.1462 0.6236 22
123.2 0.6667 0.4980 0.8354 0.4055 0.1799 0.6972 21
125.6 0.6333 0.4609 0.8058 0.4568 0.2160 0.7746 20
152.7 0.6000 0.4247 0.7753 0.5108 0.2545 0.8564 19
152.7+ 18
────────────────────────────────────────────────────────────────────────

This report displays the Kaplan-Meier product-limit survival distribution and hazard function along with
confidence limits. The formulas used were presented earlier. Note that these estimates do not use the
selected parametric distribution in any way. They are the nonparametric estimates and are completely
independent of the distribution that is being fit.
Note that censored observations are marked with a plus sign on their time value. The survival and hazard
functions are not calculated for censored observations. Also note that left censored and interval censored
observations are treated as failed observations for the calculations on this report.
Also note that the Sample Size is given for each time period. As time progresses, participants are removed
from the study, reducing the sample size. Hence, the survival results near the end of the study are based on
only a few participants and are therefore less precise. This shows up as a widening of the confidence limits.

550-21
© NCSS, LLC. All Rights Reserved.
NCSS Statistical Software NCSS.com

Distribution (Weibull) Fitting

Nonparametric Hazard Rate

Nonparametric Hazard Rate

────────────────────────────────────────────────────────────────────────
95% Confidence Interval
Limits for Hazard Rate
Failure Nonparametric Standard Error ─────────────────
Time Hazard Rate of Hazard Rate Lower Upper
──────────────────────────────────────────────────────────────────────────────────────────
8 0.0016 0.0014 0.0003 0.0090
16 0.0019 0.0014 0.0005 0.0078
24 0.0016 0.0012 0.0004 0.0066
32 0.0015 0.0010 0.0004 0.0052
40 0.0016 0.0009 0.0005 0.0047
48 0.0021 0.0011 0.0008 0.0059
56 0.0024 0.0014 0.0008 0.0075
64 0.0027 0.0015 0.0009 0.0082
72 0.0036 0.0016 0.0015 0.0086
80 0.0042 0.0017 0.0019 0.0095
88 0.0043 0.0018 0.0018 0.0099
96 0.0045 0.0019 0.0019 0.0105
104 0.0052 0.0022 0.0023 0.0117
112 0.0054 0.0022 0.0024 0.0121
120 0.0047 0.0021 0.0019 0.0113
128 0.0038 0.0020 0.0014 0.0105
136 0.0038 0.0021 0.0013 0.0111
144 0.0038 0.0036 0.0006 0.0246
152 0.0081 0.0081 0.0011 0.0575
────────────────────────────────────────────────────────────────────────

This report displays the nonparametric estimate of the hazard rate, h(t). Note that this is not the cumulative
hazard function H(t) shown in the last report. It is the derivative of H(t). Since it is h(t) that needs to be
studied in order to determine the characteristics of the failure process, this report and its associated plot
(which is shown below) become very import.
The formula for the Nelson-Aalen estimator of the cumulative hazard is

0 if 𝑡𝑡 < 𝑡𝑡1
� (𝑡𝑡) = �
𝐻𝐻 𝑑𝑑𝑖𝑖
� if 𝑡𝑡1 ≤ 𝑡𝑡
𝑌𝑌𝑖𝑖
𝑡𝑡1 ≤𝑡𝑡

The variance of this estimate is

(𝑌𝑌𝑖𝑖 − 𝑑𝑑𝑖𝑖 )𝑑𝑑𝑖𝑖
𝜎𝜎𝐻𝐻2 (𝑡𝑡) = �
(𝑌𝑌𝑖𝑖 − 1)𝑌𝑌𝑖𝑖2
𝑡𝑡𝑖𝑖 ≤𝑡𝑡

In the above equation, 𝑑𝑑𝑖𝑖 represents the number of deaths at time 𝑡𝑡𝑖𝑖 and 𝑌𝑌𝑖𝑖 represents the number of
individuals who are at risk at time 𝑡𝑡𝑖𝑖 .

550-22
© NCSS, LLC. All Rights Reserved.
NCSS Statistical Software NCSS.com

Distribution (Weibull) Fitting

The hazard rate is estimated using kernel smoothing of the Nelson-Aalen estimator as given in Klein and
Moeschberger (1997). The formulas for the estimated hazard rate and its variance are given by

1 𝑡𝑡 − 𝑡𝑡𝑘𝑘
ℎ�(𝑡𝑡) = � 𝐾𝐾 � � (𝑡𝑡𝑘𝑘 )
� ∆𝐻𝐻
𝑏𝑏 𝑏𝑏
𝐷𝐷

1 𝑡𝑡 − 𝑡𝑡𝑘𝑘 2
𝜎𝜎 2 �ℎ�(𝑡𝑡)� = � 𝐾𝐾 � � ∆𝑉𝑉� �𝐻𝐻
� (𝑡𝑡𝑘𝑘 )�
𝑏𝑏 2 𝑏𝑏
𝐷𝐷

where b is the bandwidth about t and

� (𝑡𝑡𝑘𝑘 ) = 𝐻𝐻
Δ𝐻𝐻 � (𝑡𝑡𝑘𝑘 ) − 𝐻𝐻
� (𝑡𝑡𝑘𝑘−1 )

� (𝑡𝑡𝑘𝑘 )� = 𝑉𝑉� �𝐻𝐻

Δ𝑉𝑉� �𝐻𝐻 � (𝑡𝑡𝑘𝑘 )� − 𝑉𝑉� �𝐻𝐻
� (𝑡𝑡𝑘𝑘−1 )�

Three choices are available for the kernel function K(x) in the above formulation. These are defined
differently for various values of t. Note that the 𝑡𝑡𝑘𝑘 ’s are for failed items only and that 𝑡𝑡𝐷𝐷 is the maximum
failure time. For the uniform kernel the formulas for the various values of t are

1
𝐾𝐾(𝑥𝑥) = for 𝑡𝑡 − 𝑏𝑏 ≤ 𝑡𝑡 ≤ 𝑡𝑡 + 𝑏𝑏
2

4(1 + 𝑞𝑞 3 ) 6(1 − 𝑞𝑞)

𝐾𝐾𝐿𝐿 (𝑥𝑥) = + 𝑥𝑥 for 𝑡𝑡 < 𝑏𝑏
(1 + 𝑞𝑞)4 (1 + 𝑞𝑞)3

4(1 + 𝑟𝑟 3 ) 6(1 − 𝑟𝑟)

𝐾𝐾𝑅𝑅 (𝑥𝑥) = − 𝑥𝑥 for 𝑡𝑡𝐷𝐷 − 𝑏𝑏 < 𝑡𝑡 < 𝑡𝑡𝐷𝐷
(1 + 𝑟𝑟)4 (1 + 𝑟𝑟)3

where

𝑡𝑡
𝑞𝑞 =
𝑏𝑏

𝑡𝑡𝐷𝐷 − 𝑡𝑡
𝑟𝑟 =
𝑏𝑏

For the Epanechnikov kernel the formulas for the various values of t are

3
𝐾𝐾(𝑥𝑥) = (1 − 𝑥𝑥 2 ) for 𝑡𝑡 − 𝑏𝑏 ≤ 𝑡𝑡 ≤ 𝑡𝑡 + 𝑏𝑏
4

𝐾𝐾𝐿𝐿 (𝑥𝑥) = 𝐾𝐾(𝑥𝑥)(𝐴𝐴 + 𝐵𝐵𝐵𝐵) for 𝑡𝑡 < 𝑏𝑏

4(1 + 𝑟𝑟 3 ) 6(1 − 𝑟𝑟)

𝐾𝐾𝑅𝑅 (𝑥𝑥) = − 𝑥𝑥 for 𝑡𝑡𝐷𝐷 − 𝑏𝑏 < 𝑡𝑡 < 𝑡𝑡𝐷𝐷
(1 + 𝑟𝑟)4 (1 + 𝑟𝑟)3

550-23
© NCSS, LLC. All Rights Reserved.
NCSS Statistical Software NCSS.com

Distribution (Weibull) Fitting

where

64(2 − 4𝑞𝑞 + 6𝑞𝑞 2 − 3𝑞𝑞3 )

𝐴𝐴 =
(1 + 𝑞𝑞)4 (19 − 18𝑞𝑞 + 3𝑞𝑞 2 )

240(1 − 𝑞𝑞)2
𝐵𝐵 =
(1 + 𝑞𝑞)4 (19 − 18𝑞𝑞 + 3𝑞𝑞 2 )

𝑡𝑡
𝑞𝑞 =
𝑏𝑏

𝑡𝑡𝐷𝐷 − 𝑡𝑡
𝑟𝑟 =
𝑏𝑏

For the biweight kernel the formulas for the various values of t are

15
𝐾𝐾(𝑥𝑥) = (1 − 𝑥𝑥 2 )2 for 𝑡𝑡 − 𝑏𝑏 ≤ 𝑡𝑡 ≤ 𝑡𝑡 + 𝑏𝑏
16

𝐾𝐾𝐿𝐿 (𝑥𝑥) = 𝐾𝐾(𝑥𝑥)(𝐴𝐴 + 𝐵𝐵𝐵𝐵) for 𝑡𝑡 < 𝑏𝑏

𝐾𝐾𝑅𝑅 (𝑥𝑥) = 𝐾𝐾(−𝑥𝑥)(𝐴𝐴 − 𝐵𝐵𝐵𝐵) for 𝑡𝑡𝐷𝐷 − 𝑏𝑏 < 𝑡𝑡 < 𝑡𝑡𝐷𝐷

where

64(8 − 24𝑞𝑞 + 48𝑞𝑞 2 − 45𝑞𝑞 3 + 15𝑞𝑞4 )

𝐴𝐴 =
(1 + 𝑞𝑞)5 (81 − 168𝑞𝑞 + 126𝑞𝑞 2 − 40𝑞𝑞 3 + 5𝑞𝑞 4 )

1120(1 − 𝑞𝑞)3
𝐵𝐵 =
(1 + 𝑞𝑞)5 (81 − 168𝑞𝑞 + 126𝑞𝑞2 − 40𝑞𝑞 3 + 5𝑞𝑞 4 )

𝑡𝑡
𝑞𝑞 =
𝑏𝑏

𝑡𝑡𝐷𝐷 − 𝑡𝑡
𝑟𝑟 =
𝑏𝑏

Confidence intervals for h(t) are given by

𝑧𝑧1−𝛼𝛼⁄2 𝜎𝜎�ℎ�(𝑡𝑡)�
ℎ�(𝑡𝑡)exp �± �
ℎ�(𝑡𝑡)

Care must be taken when using these kernel-smoothed estimators since they are actually estimating a
smoothed version of the hazard rate, not the hazard rate itself. Thus, they may be biased and are greatly
influenced by the choice of the bandwidth b. We have found that you must experiment with b to find an
appropriate value for each dataset.

550-24
© NCSS, LLC. All Rights Reserved.
NCSS Statistical Software NCSS.com

Distribution (Weibull) Fitting

Parametric Hazard Rate

Weibull Hazard Rate

────────────────────────────────────────────────────────────────────────
95% Confidence Interval
Weibull Limits for Hazard Rate
Failure Hazard ─────────────────
Time Rate Lower Upper
──────────────────────────────────────────────────────────
8 0.0011 0.0005 0.0023
16 0.0016 0.0008 0.0032
24 0.0020 0.0010 0.0040
32 0.0023 0.0011 0.0046
40 0.0025 0.0012 0.0052
48 0.0028 0.0014 0.0057
56 0.0030 0.0015 0.0062
64 0.0032 0.0016 0.0066
72 0.0034 0.0017 0.0070
80 0.0036 0.0018 0.0074
88 0.0038 0.0019 0.0078
96 0.0040 0.0020 0.0081
104 0.0041 0.0020 0.0085
112 0.0043 0.0021 0.0088
120 0.0045 0.0022 0.0091
128 0.0046 0.0023 0.0094
136 0.0048 0.0023 0.0097
144 0.0049 0.0024 0.0100
152 0.0050 0.0025 0.0103
160 0.0052 0.0025 0.0105
────────────────────────────────────────────────────────────────────────

This report displays the maximum likelihood estimates of the hazard rate, h(t), based on the selected
probability distribution and the definition of the hazard rate

𝑓𝑓(𝑡𝑡)
ℎ(𝑡𝑡) =
𝑅𝑅(𝑡𝑡)

Asymptotic confidence limits are computed using the formula from Nelson (1991) page 294.

𝑧𝑧1−𝛼𝛼⁄2 𝑠𝑠�ℎ�(𝑡𝑡)�
ℎ�(𝑡𝑡)exp �± �
ℎ�(𝑡𝑡)

where

2 2
𝜕𝜕ℎ� 𝜕𝜕ℎ� 𝜕𝜕ℎ� 𝜕𝜕ℎ�
𝑠𝑠 �ℎ�(𝑡𝑡)� = �
2
� 𝑣𝑣𝑣𝑣1,1 + � � 𝑣𝑣𝑣𝑣2,2 + 2 � �� � 𝑣𝑣𝑣𝑣1,2
𝜕𝜕𝑃𝑃1 𝜕𝜕𝑃𝑃2 𝜕𝜕𝑃𝑃1 𝜕𝜕𝑃𝑃2

The partial derivatives are evaluated using numerical differentiation.

Note that we have found that the above approximation behaves poorly for some distributions. However,
this is the only formula that we have been able to find, so this is what we provide. If you find that the
confidence limits have a strange appearance (especially, in that the width goes to zero), please ignore them.
They should appear as nice expanding lines about the estimated hazard rate.

550-25
© NCSS, LLC. All Rights Reserved.
NCSS Statistical Software NCSS.com

Distribution (Weibull) Fitting

Parametric Failure Distribution

Weibull Failure Distribution

────────────────────────────────────────────────────────────────────────
Maximum Likelihood (MLE)
──────────────────────
95% Confidence
Probability Interval Limits
Failure Plot Estimate Estimate ────────────
Time of Failure of Failure Lower Upper
──────────────────────────────────────────────────────────────────────────
8 0.0110 0.0059 0.0005 0.0622
16 0.0262 0.0167 0.0027 0.1011
24 0.0434 0.0306 0.0067 0.1345
32 0.0619 0.0469 0.0127 0.1649
40 0.0814 0.0651 0.0209 0.1935
48 0.1014 0.0849 0.0310 0.2209
56 0.1219 0.1060 0.0431 0.2477
64 0.1427 0.1281 0.0569 0.2741
72 0.1637 0.1510 0.0723 0.3005
80 0.1849 0.1747 0.0888 0.3272
88 0.2060 0.1989 0.1064 0.3543
96 0.2271 0.2235 0.1245 0.3819
104 0.2480 0.2483 0.1431 0.4102
112 0.2689 0.2734 0.1617 0.4391
120 0.2895 0.2984 0.1801 0.4688
128 0.3099 0.3234 0.1982 0.4991
136 0.3301 0.3483 0.2158 0.5298
144 0.3500 0.3730 0.2328 0.5607
152 0.3695 0.3975 0.2491 0.5917
160 0.3888 0.4216 0.2648 0.6225
────────────────────────────────────────────────────────────────────────

This report displays the estimated values of the cumulative failure distribution, F(t), at the time values that
were specified in the Times option of the Reports Tab. These failure values are the estimated probability
that failure occurs by the given time point. For example, the maximum likelihood estimate that a unit will fail
within 88 hours is 0.1989. The 95% confidence estimate of this probability is 0.1064 to 0.3543.
The asymptotic confidence limits are computed using the following formula:

𝐹𝐹�𝐿𝐿 (𝑡𝑡) = 𝐹𝐹� �𝑢𝑢� − 𝑧𝑧1−𝛼𝛼⁄2 �𝑉𝑉�(𝑢𝑢�)�

𝐹𝐹�𝑈𝑈 (𝑡𝑡) = 𝐹𝐹� �𝑢𝑢� + 𝑧𝑧1−𝛼𝛼⁄2 �𝑉𝑉� (𝑢𝑢�)�

where

𝑡𝑡 − 𝜇𝜇̂
𝑢𝑢� =
𝜎𝜎�

𝑉𝑉(𝜇𝜇̂ ) + 𝑢𝑢�2 𝑉𝑉(𝜎𝜎�) + 2𝑢𝑢�𝐶𝐶𝐶𝐶𝐶𝐶(𝜇𝜇̂ , 𝜎𝜎�)

𝑉𝑉� (𝑢𝑢�) =
𝜎𝜎� 2

Note that limits for the Weibull, lognormal, and log-logistic are found using the corresponding extreme
value, normal, and logistic probability functions using the substitution y=ln(t).

550-26
© NCSS, LLC. All Rights Reserved.
NCSS Statistical Software NCSS.com

Distribution (Weibull) Fitting

Parametric Reliability

Weibull Reliability
────────────────────────────────────────────────────────────────────────
Maximum Likelihood (MLE)
───────────────────────
95% Confidence
Probability Interval Limits
Failure Plot Estimate Estimate ────────────
Time of Survival of Survival Lower Upper
───────────────────────────────────────────────────────────────────────────
8 0.9890 0.9941 0.9378 0.9995
16 0.9738 0.9833 0.8989 0.9973
24 0.9566 0.9694 0.8655 0.9933
32 0.9381 0.9531 0.8351 0.9873
40 0.9186 0.9349 0.8065 0.9791
48 0.8986 0.9151 0.7791 0.9690
56 0.8781 0.8940 0.7523 0.9569
64 0.8573 0.8719 0.7259 0.9431
72 0.8363 0.8490 0.6995 0.9277
80 0.8151 0.8253 0.6728 0.9112
88 0.7940 0.8011 0.6457 0.8936
96 0.7729 0.7765 0.6181 0.8755
104 0.7520 0.7517 0.5898 0.8569
112 0.7311 0.7266 0.5609 0.8383
120 0.7105 0.7016 0.5312 0.8199
128 0.6901 0.6766 0.5009 0.8018
136 0.6699 0.6517 0.4702 0.7842
144 0.6500 0.6270 0.4393 0.7672
152 0.6305 0.6025 0.4083 0.7509
160 0.6112 0.5784 0.3775 0.7352
────────────────────────────────────────────────────────────────────────

This report displays the estimated reliability (survival) at the time values that were specified in the Times
option of the Reports Tab. Reliability may be thought of as the probability that failure occurs after the given
failure time. Thus, (using the ML estimates) the probability is 0.9531 that failure will not occur until after 32
hours. The 95% confidence for this estimated probability is 0.8351 to 0.9873.
Two reliability estimates are provided. The first uses the parameters estimated from the probability plot and
the second uses the maximum likelihood estimates. Confidence limits are calculated for the maximum
likelihood estimates. (They have not been derived for the probability plot estimates for all data situations).
The formulas used are as follows.

𝑅𝑅�𝐿𝐿 (𝑡𝑡) = 𝑅𝑅� �𝑢𝑢� − 𝑧𝑧1−𝛼𝛼⁄2 �𝑉𝑉�(𝑢𝑢�)�

𝑅𝑅�𝑈𝑈 (𝑡𝑡) = 𝑅𝑅� �𝑢𝑢� + 𝑧𝑧1−𝛼𝛼⁄2 �𝑉𝑉�(𝑢𝑢�)�

where

𝑡𝑡 − 𝑀𝑀�
𝑢𝑢� =
𝑆𝑆̂
� � + 𝑢𝑢�2 𝑉𝑉�𝑆𝑆̂� + 2𝑢𝑢�𝐶𝐶𝐶𝐶𝐶𝐶�𝑀𝑀
𝑉𝑉�𝑀𝑀 � , 𝑆𝑆̂�
𝑉𝑉� (𝑢𝑢�) =
𝑆𝑆̂ 2

550-27
© NCSS, LLC. All Rights Reserved.
NCSS Statistical Software NCSS.com

Distribution (Weibull) Fitting

Note that limits for the Weibull, lognormal, and log-logistic are found using the corresponding extreme
value, normal, and logistic probability functions using the substitution y=ln(t).

Parametric Percentiles

Weibull Percentiles
────────────────────────────────────────────────────────────────────────
Maximum Likelihood (MLE)
────────────────────────
95% Confidence
Probability Plot Interval Limits
Failure Time Estimate of Estimate of ────────────
Percentage Failure Time Failure Time Lower Upper
──────────────────────────────────────────────────────────────────────────────────────
5 26.9 33.4 14.2 78.4
10 47.4 53.8 28.5 101.4
15 66.8 71.6 42.7 120.3
20 85.7 88.4 56.5 138.3
25 104.7 104.5 69.7 156.8
30 124.1 120.5 82.2 176.7
35 144.0 136.5 93.9 198.6
40 164.7 152.8 104.8 222.9
45 186.5 169.6 115.0 250.1
50 209.5 187.0 124.7 280.5
55 234.3 205.4 134.0 314.8
60 261.1 225.0 143.1 353.7
65 290.7 246.1 152.1 398.3
70 323.8 269.5 161.3 450.3
75 361.9 295.8 170.9 512.1
80 407.1 326.5 181.3 588.2
85 463.5 364.1 193.0 686.7
90 540.0 413.9 207.4 826.0
95 664.5 492.6 227.8 1065.0
────────────────────────────────────────────────────────────────────────

This report displays failure time percentiles and, for the maximum likelihood estimates, confidence intervals
for those percentiles. For example, the estimated median failure time is 187 hours. The 95% confidence
limits for the median time are 124.7 to 280.5. Note that these limits are very wide for two reasons. First, the
sample size is small. Second, the shape parameter is less than 2.0.
The estimated 100pth percentile and associated confidence interval is computed using the following steps:
1. Compute 𝑤𝑤𝑝𝑝 = 𝐹𝐹 −1 (𝑝𝑝)
� + 𝑤𝑤𝑝𝑝 𝑆𝑆̂. Note that in the case of the Weibull and exponential distributions, we let
2. Compute 𝑦𝑦𝑝𝑝 = 𝑀𝑀
� = ln�𝐶𝐶̂ � and 𝑆𝑆̂ = 1 ∕ 𝐵𝐵�.
𝑀𝑀
3. Compute 𝑉𝑉�𝑦𝑦𝑝𝑝 � = 𝑉𝑉𝑉𝑉1,1 + 𝑦𝑦𝑝𝑝2 𝑉𝑉𝑉𝑉2,2 + 2𝑦𝑦𝑝𝑝 𝑉𝑉𝑉𝑉1,2 .

4. For the normal, extreme value, and logistic distributions, the confidence interval for the percentile is
given by

𝑇𝑇𝐿𝐿𝐿𝐿𝐿𝐿𝐿𝐿𝐿𝐿,𝑝𝑝 = 𝑦𝑦𝑝𝑝 − 𝑧𝑧1−𝛼𝛼⁄2 �𝑉𝑉�𝑦𝑦𝑝𝑝 � + 𝐷𝐷

𝑇𝑇𝑈𝑈𝑈𝑈𝑈𝑈𝑈𝑈𝑈𝑈,𝑝𝑝 = 𝑦𝑦𝑝𝑝 + 𝑧𝑧1−𝛼𝛼⁄2 �𝑉𝑉�𝑦𝑦𝑝𝑝 � + 𝐷𝐷

550-28
© NCSS, LLC. All Rights Reserved.
NCSS Statistical Software NCSS.com

Distribution (Weibull) Fitting

For the lognormal, exponential, Weibull, and log-logistic distributions, the confidence interval for the
percentile is given by

𝑇𝑇𝐿𝐿𝐿𝐿𝐿𝐿𝐿𝐿𝐿𝐿,𝑝𝑝 = exp �𝑦𝑦𝑝𝑝 − 𝑧𝑧1−𝛼𝛼⁄2 �𝑉𝑉�𝑦𝑦𝑝𝑝 �� + 𝐷𝐷

𝑇𝑇𝑈𝑈𝑈𝑈𝑈𝑈𝑈𝑈𝑈𝑈,𝑝𝑝 = exp �𝑦𝑦𝑝𝑝 + 𝑧𝑧1−𝛼𝛼⁄2 �𝑉𝑉�𝑦𝑦𝑝𝑝 �� + 𝐷𝐷

For the lognormal base 10 distribution, the confidence interval for the percentile is given by

𝑦𝑦𝑝𝑝 −𝑧𝑧1−𝛼𝛼⁄2 �𝑉𝑉�𝑦𝑦𝑝𝑝 �

𝑇𝑇𝐿𝐿𝐿𝐿𝐿𝐿𝐿𝐿𝐿𝐿,𝑝𝑝 = 10 + 𝐷𝐷

𝑦𝑦𝑝𝑝 +𝑧𝑧1−𝛼𝛼⁄2 �𝑉𝑉�𝑦𝑦𝑝𝑝 �

𝑇𝑇𝑈𝑈𝑈𝑈𝑈𝑈𝑈𝑈𝑈𝑈,𝑝𝑝 = 10 + 𝐷𝐷

Parametric Residual Life

Weibull Residual Life

────────────────────────────────────────────────────────────────────────
Residual Life Percentile
Failure Proportion ───────────────────────
Time Failing 25th 50th 75th 90th
──────────────────────────────────────────────────────────────────────────
8 0.0059 97.9 180.1 288.7 406.6
16 0.0167 92.5 174.0 282.2 399.9
24 0.0306 87.9 168.5 276.2 393.5
32 0.0469 83.8 163.5 270.6 387.5
40 0.0651 80.1 158.9 265.3 381.8
48 0.0849 76.9 154.5 260.2 376.3
56 0.1060 73.9 150.5 255.4 371.1
64 0.1281 71.3 146.7 250.9 366.0
72 0.1510 68.8 143.2 246.5 361.1
80 0.1747 66.6 139.9 242.4 356.4
88 0.1989 64.6 136.7 238.4 351.8
96 0.2235 62.7 133.8 234.5 347.4
104 0.2483 60.9 131.0 230.9 343.1
112 0.2734 59.3 128.3 227.3 339.0
120 0.2984 57.8 125.8 223.9 335.0
128 0.3234 56.4 123.4 220.7 331.1
136 0.3483 55.1 121.1 217.5 327.3
144 0.3730 53.8 118.9 214.5 323.6
152 0.3975 52.7 116.9 211.5 320.0
160 0.4216 51.6 114.9 208.7 316.6
────────────────────────────────────────────────────────────────────────

This report gives percentiles of the estimated life remaining after a certain time period. For example, the
estimated median remaining life of items reaching 80.0 hours is 139.9 hours.

550-29
© NCSS, LLC. All Rights Reserved.
NCSS Statistical Software NCSS.com

Distribution (Weibull) Fitting

The percentile and associated confidence interval of residual (remaining) life is computed using the
following steps:
𝑦𝑦𝑝𝑝 −𝑀𝑀�
1. Compute 𝑧𝑧𝑝𝑝 = . Note that in the case of the Weibull and exponential distributions, we let
𝑆𝑆̂
� = ln�𝐶𝐶̂ � and 𝑆𝑆̂ = 1 ∕ 𝐵𝐵�. Also note that for the normal, extreme value, and logistic distributions,
𝑀𝑀
𝑦𝑦𝑝𝑝 = 𝑡𝑡. For the lognormal, Weibull, and log-logistic distributions, 𝑦𝑦𝑝𝑝 = 𝑒𝑒 𝑡𝑡 . And for the lognormal base
10 distribution, 𝑦𝑦𝑝𝑝 = 10𝑡𝑡 .

2. Compute 𝑝𝑝0 = 𝐹𝐹�𝑧𝑧𝑝𝑝 �

3. Compute 𝑝𝑝1 = 𝑝𝑝0 (1 + 𝑃𝑃), where P is the percentile of residual life to be estimated.
� + 𝐹𝐹 −1 (𝑝𝑝1 )𝑆𝑆̂. Note that in the case of the Weibull and exponential distributions, we
4. Compute 𝑤𝑤𝑝𝑝 = 𝑀𝑀
� = ln�𝐶𝐶̂ � and 𝑆𝑆̂ = 1 ∕ 𝐵𝐵�.
let 𝑀𝑀

For the normal, extreme value, and logistic distributions, the estimate is given by

𝑇𝑇𝑝𝑝 = 𝑤𝑤𝑝𝑝

For the lognormal, exponential, Weibull, and log-logistic distributions, the estimate is given by

𝑇𝑇𝑝𝑝 = 𝑒𝑒 𝑤𝑤𝑝𝑝

For the lognormal base 10 distribution, the estimate is given by

𝑇𝑇𝑝𝑝 = 10𝑤𝑤𝑝𝑝

Survival Plot

Survival Plot
─────────────────────────────────────────────────────────────────────────

This plot shows the product-limit survivorship function (the step function) as well as the parametric survival
plot and associated confidence intervals. If there are several groups, a separate line is drawn for each group.

550-30
© NCSS, LLC. All Rights Reserved.
NCSS Statistical Software NCSS.com

Distribution (Weibull) Fitting

Hazard Function Plot

Hazard Function Plot

─────────────────────────────────────────────────────────────────────────

This plot shows the parametric and nonparametric cumulative hazard functions for the data analyzed.
Confidence limits for the parametric cumulative hazard function are also given.
If you have several groups, then a separate line is drawn for each group. The shape of the hazard function is
often used to determine an appropriate survival distribution.

Hazard Rate Plot

Hazard Rate Plot

─────────────────────────────────────────────────────────────────────────

550-31
© NCSS, LLC. All Rights Reserved.
NCSS Statistical Software NCSS.com

Distribution (Weibull) Fitting

This plot shows the parametric and nonparametric hazard rate plots with confidence limits for the
parametric hazard rate. This plot is especially useful for studying the shape of the nonparametric hazard
rate and comparing that with the parametric hazard rate. When selecting a probability distribution to
represent a set of data, it is important to determine if the parametric hazard rate plot has a general shape
that is consistent both with the nonparametric hazard rate and with your prior knowledge of the hazard
distribution. This plot allows you to make this comparison.
Note that the asymptotic confidence intervals are not well behaved for some distributions. If the confidence
intervals seem to have zero width at some point along the plot, you should realize that they fall into this
category and ignore them.

Probability Plot

Probability Plot(s)
─────────────────────────────────────────────────────────────────────────

This is the Weibull probability plot of these data. The expected quantile of the theoretical distribution is
plotted on the horizontal axis. The natural logarithm of the time value is plotted on the vertical axis. Note
that censored points are not shown on this plot. Also note that for tied data, only one point is shown for
each set of ties.
This plot lets you investigate the goodness of fit of the selected probability distribution to your data. If the
points seem to fall along a straight line, the selected probability model may be useful. If the plot shows a
downward curve, the value of the threshold parameter, D, may need to be increased. If the plot shows an
upward curve, the value of the threshold parameter may need to be decreased. Or you may need to select a
different distribution.
You must decide whether the probability distribution is a good fit to your data by looking at this plot and by
comparing the value of the log-likelihood to that of other distributions.

550-32
© NCSS, LLC. All Rights Reserved.
NCSS Statistical Software NCSS.com

Distribution (Weibull) Fitting

Multiple-Censored and Grouped Data

The case of grouped and multiple-censored data cause special problems when creating a probability plot.
Remember that the horizontal axis represents the expected quantile from the selected distribution for each
(sorted) failure time. In the regular case, we use the rank of the observation in the overall dataset. However,
in the case of grouped or multiple-censored data, we use a modified rank. This modified rank, 𝑂𝑂𝑗𝑗 , is
computed as follows
𝑂𝑂𝑗𝑗 = 𝑂𝑂𝑝𝑝 + 𝐼𝐼𝑗𝑗
where

(𝑛𝑛 + 1) − 𝑂𝑂𝑝𝑝
𝐼𝐼𝑗𝑗 =
1 + 𝑐𝑐

where 𝐼𝐼𝑗𝑗 is the increment for the jth failure; n is the total number of data points, both censored and
uncensored; 𝑂𝑂𝑝𝑝 is the order of the previous failure; and c is the number of data points remaining in the data
set, including the current data. Implementation details of this procedure may be found in Dodson (1994).
Left censored and interval censored data are treated as failures for making the probability plots.

550-33
© NCSS, LLC. All Rights Reserved.
NCSS Statistical Software NCSS.com

Distribution (Weibull) Fitting

Example 2 – Distribution Selection

This section presents an example of how to let the program help you pick an appropriate parametric
distribution. The data used were shown above and are found in the Weibull dataset.

Setup
To run this example, complete the following steps:

1 Open the Weibull example dataset

• From the File menu of the NCSS Data window, select Open Example Data.
• Select Weibull and click OK.

2 Specify the Distribution (Weibull) Fitting procedure options

• Find and open the Distribution (Weibull) Fitting procedure using the menus or the Procedure
Navigator.
• The settings for this example are listed below and are stored in the Example 2 settings file. To load
these settings to the procedure window, click Open Example Settings File in the Help Center or File
menu.

Variables Tab
___________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________

Time Variable................................................... Time

Censor Variable ............................................... Censor
Frequency Variable.......................................... Count
Distribution ....................................................... Find Best

Options Tab
___________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________

Derivatives ....................................................... 0.00006

Plots Tab
___________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________

Two Plots Per Line........................................... Checked

3 Run the procedure

• Click the Run button to perform the calculations and generate the output.

550-34
© NCSS, LLC. All Rights Reserved.
NCSS Statistical Software NCSS.com

Distribution (Weibull) Fitting

Data Summary

Data Summary
────────────────────────────────────────────────────────────────────────
Type of
Observation Rows Count Percent (%) Minimum Maximum
────────────────────────────────────────────────────────────────────────────────────────────────────
Failed 12 12 40% 12.5 152.7
Right Censored 1 18 60% 152.7 152.7
Left Censored 0 0
Interval Censored 0 0
Total 13 30 100% 12.5 152.7
────────────────────────────────────────────────────────────────────────

This report displays a summary of the data that were analyzed. Scan this report to determine if there were
any obvious data errors by double-checking the counts and the minimum and maximum.

Distribution Fit Summary

Distribution Fit Summary

────────────────────────────────────────────────────────────────────────
Distribution Log-Likelihood Shape Scale Threshold
─────────────────────────────────────────────────────────────────────────────────────────────
Weibull -80.05649 1.511543 238.3481 0
Loglogistic -80.11679 5.28008 0.5909371 0
Lognormal -80.38821 1.137753 5.349999 0
Lognormal10 -80.38821 0.4941201 2.323475 0
Exponential -81.04864 1 315.4667 0
Normal -81.24539 171.1062 84.88175 0
Logistic -81.74763 169.1118 49.77026 0
Extreme Value -82.11030 189.3399 57.44398 0
────────────────────────────────────────────────────────────────────────

This report displays the values of the log-likelihood for each distribution along with the estimated values of
its parameters. Since our desire is to maximize the likelihood, under normal circumstances, we would pick
the distribution at the top of the report since it has the largest likelihood value. In this example, we would
select the Weibull distribution.

550-35
© NCSS, LLC. All Rights Reserved.
NCSS Statistical Software NCSS.com

Distribution (Weibull) Fitting

Probability Plots

Probability Plot(s)
─────────────────────────────────────────────────────────────────────────

550-36
© NCSS, LLC. All Rights Reserved.
NCSS Statistical Software NCSS.com

Distribution (Weibull) Fitting

By studying these probability plots, we can determine which distributions fit the data the best. In this
example, since there are only a few observations, it is difficult to select one distribution over another. We
can see that our candidate from the last section, the Weibull distribution, certainly cannot be removed on
the basis of its probability plot. Without further information, our decision would be to select the Weibull
distribution to fit these data.

550-37
© NCSS, LLC. All Rights Reserved.
NCSS Statistical Software NCSS.com

Distribution (Weibull) Fitting

Example 3 – Readout Data

This section presents an example of how to analyze readout data. The data used are found in the
Readout105 dataset. The table below shows the results of a study to test the failure rate of a particular
machine. This study began with 40 test machines. After each time period (24, 72, 168, etc.) the number of
machines that had failed since the period began was recorded. This number is entered into the Count
variable. Hence, two machines failed during the first 24 hours, one machine failed between 24 and 72 hours,
and so on.
After 1500 hours, the study was terminated. Sixteen machines still had not failed. The data are entered in
the spreadsheet as shown below.
We have used obvious indicators for censoring. Since the first period begins with a zero time, this entry
represents left censored data. We indicate left censoring with an ‘L.’ The next eight rows represent interval
censored data. Both beginning and ending times are needed for these entries. We indicate interval
censoring with an ‘I.’ The last row corresponds to the sixteen machines that did not fail. These are entered
as right censored data, which is indicated with an ‘R.’

Readout105 Dataset

Time1 Time2 Censor Count

24 0 L 2
72 24 I 1
168 72 I 3
300 168 I 2
500 300 I 2
750 500 I 4
1000 750 I 5
1250 1000 I 1
1500 1250 I 4
1500 R 16

Setup
To run this example, complete the following steps:

1 Open the Readout105 example dataset

• From the File menu of the NCSS Data window, select Open Example Data.
• Select Readout105 and click OK.

550-38
© NCSS, LLC. All Rights Reserved.
NCSS Statistical Software NCSS.com

Distribution (Weibull) Fitting

2 Specify the Distribution (Weibull) Fitting procedure options

• Find and open the Distribution (Weibull) Fitting procedure using the menus or the Procedure
Navigator.
• The settings for this example are listed below and are stored in the Example 3 settings file. To load
these settings to the procedure window, click Open Example Settings File in the Help Center or File
menu.

Variables Tab
___________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________

Time Variable................................................... Time1

Start Time Variable .......................................... Time2
Frequency Variable.......................................... Count
Censor Variable ............................................... Censor
Failed ............................................................... F
Right .............................................................. R
Left................................................................. L
Interval ........................................................... I
Distribution ....................................................... Find Best

Options Tab
___________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________

Derivatives ....................................................... 0.00006

Plots Tab
___________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________

Two Plots Per Line........................................... Checked

3 Run the procedure

• Click the Run button to perform the calculations and generate the output.

Data Summary

Data Summary
────────────────────────────────────────────────────────────────────────
Type of
Observation Rows Count Percent (%) Minimum Maximum
────────────────────────────────────────────────────────────────────────────────────────────────────
Failed 0 0
Right Censored 1 16 40% 1500 1500
Left Censored 1 2 5% 24 24
Interval Censored 8 22 55% 24 1500
Total 10 40 100% 24 1500
────────────────────────────────────────────────────────────────────────

This report displays a summary of the data that were analyzed. We note that the number of rows and the
total count appear to be correct.

550-39
© NCSS, LLC. All Rights Reserved.
NCSS Statistical Software NCSS.com

Distribution (Weibull) Fitting

Distribution Fit Summary

Distribution Fit Summary

────────────────────────────────────────────────────────────────────────
Distribution Log-Likelihood Shape Scale Threshold
────────────────────────────────────────────────────────────────────────────────────────────
Weibull -79.42889 0.8222772 1746.067 0
Exponential -79.96207 1 1631.161 0
Loglogistic -80.27086 7.044066 1.030881 0
Lognormal10 -81.19075 0.8194178 3.046982 0
Lognormal -81.19075 1.886779 7.015936 0
Normal -81.44245 1213.697 913.7082 0
Logistic -82.05516 1199.686 563.52 0
Extreme Value -83.09204 1525.271 726.1455 0
────────────────────────────────────────────────────────────────────────

It appears that the Weibull distribution is a reasonable choice for the parametric distribution, although the
shape parameter is less than one. This may point to the need for a nonzero threshold value.
To finish this example, you would view the probability plots. Finally, you would try fitting the Weibull
distribution to these data. We will leave that to you to do. Simply change the Distribution box to Weibull and
rerun the procedure.

550-40
© NCSS, LLC. All Rights Reserved.
NCSS Statistical Software NCSS.com

Distribution (Weibull) Fitting

Example 4 – Engine Fan Data

Nelson (1982) gives data on the failure times of seventy diesel engine fans. Twelve of the fans failed during
the duration of the test. Fifty-eight of the fans completed the test without failure, so only their running times
were recorded. These data are contained in the FanFailure dataset. You can observe the data by opening
this dataset. Note that ‘F’ designates a failure and ‘C’ designates a censored (non-failed) fan.
Two questions were to be answered from these data. First of all, the warranty period for the fan is 8000
hours. Management wanted to know what percentage would fail on or before the warranty period ended.
Second, management wanted to know what happens to the failure rate as the fans age.
The following steps will set up the procedure to analyze these data and answer the two questions given
above.

Setup
To run this example, complete the following steps:

1 Open the FanFailure example dataset

• From the File menu of the NCSS Data window, select Open Example Data.
• Select FanFailure and click OK.

2 Specify the Distribution (Weibull) Fitting procedure options

• Find and open the Distribution (Weibull) Fitting procedure using the menus or the Procedure
Navigator.
• The settings for this example are listed below and are stored in the Example 4 settings file. To load
these settings to the procedure window, click Open Example Settings File in the Help Center or File
menu.

Variables Tab
___________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________

Time Variable................................................... Hours

Frequency Variable.......................................... Count
Censor Variable ............................................... Censor
Failed ............................................................... F
Right .............................................................. C
Distribution ....................................................... Weibull

Reports Tab
___________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________

Times ............................................................... 1000:15000(1000)

3 Run the procedure

• Click the Run button to perform the calculations and generate the output.

We will show only those portions of the printout that are necessary to answer the two questions that were
posed at the beginning of this section.

550-41
© NCSS, LLC. All Rights Reserved.
NCSS Statistical Software NCSS.com

Distribution (Weibull) Fitting

Parameter Estimation

Weibull Parameter Estimation

────────────────────────────────────────────────────────────────────────
Maximum Likelihood (MLE)
───────────────────────────────────
95% Confidence
Probability Interval Limits
Plot Standard ────────────────
Parameter Estimate* Estimate Error Lower Upper
─────────────────────────────────────────────────────────────────────────────────────────────────────────
B (Shape) 1.202295 1.058446 0.2682645 0.6440661 1.739429
C (Scale) 17283.81 26296.85 12252.48 10551.25 65539.57
D (Threshold) 0 0
Log-Likelihood -135.1527
Mean 16250.13 25715.61
Median 12742.28 18600.24
Mode 3925.094 1703.919
Sigma 13574.96 24306.58
─────────────────────────────────────────────────────────────────────────────────────────────────────────

Estimation Details
──────────────────────────────────────────────────
Differential Evolution Iterations 32
Newton-Raphson Restart 1
Newton-Raphson Iterations 7
User-Entered Random Seed 2928475
──────────────────────────────────────────────────
────────────────────────────────────────────────────────────────────────
* Probability plot estimates were generated with F(t) calculated using the approximate median and using the model Time = A +
B(F).

This report shows the estimated parameters. We are particularly interested to see that the shape parameter
is almost exactly one. The confidence limits for the estimated shape parameter include one between them.
Remember that when the shape parameter is one, the Weibull distribution reduces to the exponential
distribution, a distribution which ‘has no memory.’ From this, we get an indication that the failure pattern of
the fans does not change over time. That is, the failure rate does not change as the fans get older.

550-42
© NCSS, LLC. All Rights Reserved.
NCSS Statistical Software NCSS.com

Distribution (Weibull) Fitting

Parametric Failure Distribution

Weibull Failure Distribution

────────────────────────────────────────────────────────────────────────
Maximum Likelihood (MLE)
──────────────────────
95% Confidence
Probability Interval Limits
Failure Plot Estimate Estimate ────────────
Time of Failure of Failure Lower Upper
──────────────────────────────────────────────────────────────────────────
1000 0.0320 0.0309 0.0105 0.0895
2000 0.0721 0.0633 0.0289 0.1360
3000 0.1147 0.0956 0.0501 0.1782
4000 0.1581 0.1274 0.0719 0.2203
5000 0.2016 0.1585 0.0927 0.2636
6000 0.2444 0.1888 0.1121 0.3083
7000 0.2863 0.2184 0.1297 0.3539
8000 0.3270 0.2471 0.1459 0.3999
9000 0.3664 0.2749 0.1607 0.4456
10000 0.4043 0.3019 0.1743 0.4905
11000 0.4406 0.3280 0.1870 0.5340
12000 0.4753 0.3533 0.1987 0.5758
13000 0.5084 0.3778 0.2098 0.6156
14000 0.5399 0.4014 0.2202 0.6531
15000 0.5697 0.4242 0.2300 0.6883
────────────────────────────────────────────────────────────────────────

This report presents the estimated failure proportions at various time periods. We note that at 8000 hours,
the maximum likelihood estimate for the proportion failing is 0.2471. The 95% confidence limits are 0.1459
to 0.3999. That is, almost 25% of the fans can be expected to fail by 8000 hours—a very high failure rate.
Management will have to change the fans to decrease the proportion failing!

550-43
© NCSS, LLC. All Rights Reserved.
NCSS Statistical Software NCSS.com

Distribution (Weibull) Fitting

Hazard Rate Plot

Hazard Rate Plot

─────────────────────────────────────────────────────────────────────────

This plot shows both the parametric and nonparametric estimates of the hazard rates. First, we analyze the
nonparametric estimate. Notice that the line wanders up and then down, but it does not extend outside the
confidence limits of the parametric hazard rate. The sharp rise at the end of the plot is due to a lack of data
in this region and should be ignored. We see that the parametric estimate of the hazard rate, the middle
horizontal line, is a reasonable approximation for the nonparametric line. The above considerations again
lead us to conclude that the failure rates do not change with age.
This ends this example. Notice how quickly we have been able to answer the two questions posed by
management. The only task that we did not complete was to make sure that the Weibull distribution was
appropriate for these data. A quick look at the probability plot will show you that it is.

550-44
© NCSS, LLC. All Rights Reserved.
NCSS Statistical Software NCSS.com

Distribution (Weibull) Fitting

Example 5 – Adding an At-Risk Table to a Survival Plot and

a Hazard Function Plot
This section demonstrates how to add a table containing the number of subjects at risk, the cumulative
number of censored observations, and the cumulative number of events to the bottom of a survival plot
and a hazard function plot. The hazard function plot will be created with default settings, but we’ll modify
the survival plot to highlight some of the available options. The data used in this example are contained in
the Weibull dataset.

Setup
To run this example, complete the following steps:

1 Open the Weibull example dataset

• From the File menu of the NCSS Data window, select Open Example Data.
• Select Weibull and click OK.

2 Specify the Distribution (Weibull) Fitting procedure options

• Find and open the Distribution (Weibull) Fitting procedure using the menus or the Procedure
Navigator.
• The settings for this example are listed below and are stored in the Example 5 settings file. To load
these settings to the procedure window, click Open Example Settings File in the Help Center or File
menu.

Variables Tab
_________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________

Time Variable................................................... Time

Frequency Variable.......................................... Count
Censor Variable ............................................... Censor

Options Tab
_________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________

Derivatives ....................................................... 0.00006

Reports Tab
_________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________

All Reports ....................................................... Unchecked

550-45
© NCSS, LLC. All Rights Reserved.
NCSS Statistical Software NCSS.com

Distribution (Weibull) Fitting

Plots Tab
_________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________

Survival/Reliability Plot .................................... Checked

Hazard Function Plot ....................................... Checked

Survival/Reliability Plot Format (Click the Button)

_________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________

Survival Tab
Confidence Limits (Distribution Fit Line) .......... Checked

At-Risk Table Tab

General Sub-Tab
Show At-Risk Table...................................... Checked
Display ......................................................... Number At Risk (Number Censored)
(Number of Events)
Layout Sub-Tab
Top Outside Margin .................................... 0
Groups Sub-Tab
Value Colors From ..................................... Line Fill

Hazard Function Plot Format (Click the Button)

_________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________

Haz. Function Tab

Confidence Limits (Distribution Fit Line) .......... Checked

At-Risk Table Tab

General Sub-Tab
Show At-Risk Table .................................... Checked

3 Run the procedure

• Click the Run button to perform the calculations and generate the output.

550-46
© NCSS, LLC. All Rights Reserved.
NCSS Statistical Software NCSS.com

Distribution (Weibull) Fitting

Output

Survival Plot
────────────────────────────────────────────────────────────────────────

The modified survival plot is displayed with numbers at risk, cumulative censoring, and cumulative events at
each reference time point. You can further modify the plot as required to suit your needs.

550-47
© NCSS, LLC. All Rights Reserved.
NCSS Statistical Software NCSS.com

Distribution (Weibull) Fitting

Hazard Function Plot

────────────────────────────────────────────────────────────────────────

This plot is displayed with default at-risk table settings to give you an idea of what you get if you make no
modifications to the table.

550-48
© NCSS, LLC. All Rights Reserved.