LOG-NORMAL DISTRIBUTION

Rather than use “average” and “standard deviation” to measure the likely and the
spread of results, Luco and Cornell (1997, proposed that “median, x! ” and
“dispersion, "(ln #
i
” are more reasona\$le measures, given that the spread of data does
not necessarily follo% a normal distri\$ution and that there may \$e several e#treme
values in the data& ' log normal distri\$ution also restricts the varia\$le to positive
num\$ers, %hich is often the case in engineering applications&

Definition:

(he log)normal distri\$ution is a single)tailed pro\$a\$ility distri\$ution of any random
varia\$le %hose logarithm is normally distri\$uted& *f + is a random varia\$le %ith a
normal distri\$ution, then , - e#p(+ has a log)normal distri\$ution. like%ise, if , is
log)normally distri\$uted, then log(, is normally distri\$uted& ((he \$ase of the
logarithmic function does not matter/ if loga(, is normally distri\$uted, then so is
log\$(,, for any t%o positive num\$ers a, \$ 0 1&

1ro\$a\$ility density function/
π σ
σ µ
σ
µ
2
, . (
2
2
2
(ln
x
e
x f
x−

=

3or # 4 5, %here 6 and " are the mean and standard deviation of the varia\$le7s
logarithm (\$y definition, the varia\$le7s logarithm is normally distri\$uted&

8hen 6-5, for various σ/

Properties:

*f x has a log)normal distri\$ution, that is ln(x has a normal distri\$ution %ith mean µ
and standard deviation σ, then/

9ean/ 2 : e#p(
2
σ µ µ + =
x

;tandard deviation/ 1 e#p(
2
− × = σ µ σ
x x

9ode/ e#p( (
2
σ µ − = x Mode
9edian/
n
n
n
i
i
x x x x x
n
x &&& ln(
1
e#p e#p( !
< 2 1
1
=

∑ = =
=
µ
=ispersion/

=

= =
n
i
i x
x
n
1
2
ln
(ln
1
1
µ σ σ

*f you are given the normally distri\$uted mean 6
#
and standard deviation "
#
for x
%hich is log)normally distri\$uted, to find the dispersion and median for a log)normal
distri\$ution/

− =

+

=
2
ln
2
ln
2
1
ln( e#p !
1 ln
x x
x
x
x
x σ µ
µ
σ
σ

Confidence intervas:

;pecifying confidence intervals can \$e calculated using the log)space or the median
and dispersion (i&e& the geometric mean x! and geometric standard deviation
e#p(σ
lnx
/

1ercentile value Log)space 9edian > dispersion
99&9? < e#p( σ µ + [ ]
<
ln
e#p( !
x
x σ ⋅
97&7? 2 e#p( σ µ +
[ ]
2
ln
e#p( !
x
x σ ⋅
@A&1?
e#p( σ µ + [ ] e#p( !
ln x
x σ ⋅
B5&5?
e#p(µ
x!
1B&9?
e#p( σ µ −
[ ]
1
ln
e#p( !

x
x σ
2&<? 2 e#p( σ µ −
[ ]
2
ln
e#p( !

x
x σ
5&1? < e#p( σ µ −
[ ]
<
ln
e#p( !

x
x σ