Lognormal distribution
Lognormal distribution plays an important role in probabilistic design
because negative values of engineering phenomena are sometimes
physically impossible.
Some common applications of lognormal distributions include maintenance
data analysis (for example, the time it may take to repair a specific piece of
equipment) and economic and/or stock market data analysis (where positive
values may be required to determine future returns for a stock).
The lognormal distribution is an ideal model for processes where the
multiplication of effects results in time to failure. The model is considered
to be very useful in the fields of medicine, economics, and engineering.
Lognormal vs. Normal Distribution:
The key difference between the two is that lognormal distributions contain only
positive numbers, whereas normal distribution can contain negative values.
Another key difference between the two is the shape of the graph. Normally
distributed data forms a symmetric bell-shaped graph. In contrast, lognormally
distributed data does not form a symmetric shape but rather slants or skews more
towards the right.
Suppose X is a non-negative random variable of the continuous type. If
ln(X) is distributed as normal with mean μ and variance σ2, then X is said
to have a lognormal distribution with parameters μ and variance σ2 and its
probability density function is given by
• Definition
• Derivation/ relation (normal and lognormal)
• r th moment
• Mean, variance, skewness, kurtosis
• Median, mode
• #########
• HM
• GM
• Relationship: mean, median mode/ GM and median/ etc…
• Distribution function
• Problem related to distribution function
• Parameter estimation
For a lognormally distributed random variable X with parameter μ and σ,
the following are always true:
a) Mean (X) > Median (X)> Mode (X)
b) GM (X) = Median (X)
c) Mode (X)/ Median (X)={Median (X)/Mean(X)}2
1
d) { log[E(X)] + Log [HM(X)]}=E(log X)
2
e) log[E(X)]-Log [HM(X)]=Var(logX)
f) 𝛽1 (𝑋) > 0
Estimation of parameters of Lognormal Distribution
Method of Moments
We know that the mean and variance of Lognormal distributions are
Let x1, x2 ,…, xn be a random sample of size n from the lognormal distribution.
According to the method of moments, we equate the sample mean with
population mean and sample variance with population variance .
Then we have the following equations
Let and be the estimates of and respectively. Then
𝜎2
𝑥ҧ = 𝑒 𝜇ෝ + 2
(A)
2 2ෝ 2 𝑒 𝜎2 −1
𝜇+𝜎
𝑠 =𝑒 B
We need to solve these equations for and . Elimination of from (A) and
(B) gives
𝑠2 2
𝜎
2
=𝑒 −1
𝑥ҧ
𝑠 2
2
𝜎
⇒ 𝑒 =1+ 2
𝑥ҧ
2 2
2 = 𝐼𝑛 𝑥ҧ + 𝑠
𝜎
𝑥ҧ 2
From (5) we get, 2
𝜎
𝑒 𝜇ෝ = ҧ −2
𝑥𝑒
𝑥ҧ 2
𝜇Ƹ = 𝐼𝑛
𝑥ҧ 2 + 𝑠 2
Estimation of parameters of Lognormal Distribution
Method of Maximum Likelihood
Likelihood function:
Thus,
The difference between the two estimates arises because MLE maximizes
the likelihood function (focusing on probability), while MM equates
sample and theoretical moments (focusing on matching moments). This
difference in estimation methodology leads to different estimates, with
MLE being generally more efficient and preferred in practice.