CIVL2160

Modelling Systems with

Uncertainties
Lecture 6

Key statistics of a R.V.

Zhe (Walter) Wang
Department of Civil & Environmental Engineering
Motivation: describing a random variable

• The probabilistic properties of a R.V. would be completely described by

its distribution function, in terms of CDF or PDF/PMF and its associated
parameters, are specified.
• Sometimes,
• Distribution functions are unknown: you only have data but no model
• Humans prefer to use a scalar, rather than a function, to describe the
(probabilistic) behavior
• E.g., accuracy of a model, total score of the exam, …
• Main descriptors give useful information on the numerical properties
of the R.V.s that are of first importance in many practical applications.
• Mean/expectation – a measure of central tendency
• Standard derivation – a measure of dispersion
• Skewness – a measure of skewness
 01. Mean, or expectation

Agenda 02. Variance

03. Skewness

04. Summary
Expectation/mean

• Play a game 1000 times.

• Random gain at each play described by:
• “Average” gain?

• Expectation: the expected value in a large number of independent

repetitions of the experiment
Discrete random variable Continuous random variable
∞
𝐸 𝑋 = 𝜇𝑋 = ෍ 𝑥𝑖 𝑝𝑋 (𝑥𝑖 ) 𝐸 𝑋 = 𝜇𝑋 = න 𝑥𝑓𝑋 (𝑥) 𝑑𝑥
−∞
𝑎𝑙𝑙 𝑥𝑖
Examples

• Expected value of a Bernoulli r.v.

𝐸 𝑋 =1∙𝑝+0∙ 1−𝑝 =𝑝

• Expected value of a uniform r.v. (uniform on a, a+1, … b)

1 1 1 𝑎+𝑏
𝐸 𝑋 =𝑎∙ + (𝑎 + 1) ∙ +…+ 𝑏∙ =
𝑏−𝑎+1 𝑏−𝑎+1 𝑏−𝑎+1 2
Examples

• Expected value of a uniform continuous r.v. (uniform between a and b)

∞ 𝑏
1 𝑎+𝑏
𝐸 𝑋 = න 𝑥𝑓𝑋 (𝑥) 𝑑𝑥 = න 𝑥 𝑑𝑥 =
−∞ 𝑎 𝑏−𝑎 2

• Expected value of a normal r.v.

𝐸 𝑋 =𝜇
Expectation/mean

• Expectation/mean is a measure of central tendency of a random variable

• There are other measures of central tendency
• Median: the mid-value of the given set of data when arranged in a particular order
• Mode: the most frequent number occurring in the data set
Example

2 2
• Fine the mean, median and mode of 𝑓𝑋 𝑥 = − 𝑥
3 9
3
• Mean 2 2 fX (x) • Mode?
𝐸 𝑋 = න 𝑥( − 𝑥 ) 𝑑𝑥
0 3 9
1 2 2 3 3 2
= 𝑥 − 𝑥 | =1 3
3 27 0
• Median 3
2 2
න ( − 𝑥 ) 𝑑𝑥 = 0.5
𝑥𝑚 3 9
2 1 2
1− 𝑥𝑚 − 𝑥𝑚 = 0.5
3 9 x
0 1 3
• Solve for 𝑥𝑚 and take the root that lies between 0 and 3
Properties of expectation

• If 𝑋 ≥ 0, then E[𝑋] ≥ 0

• If 𝑎 ≤ 𝑋 ≤ 𝑏, then 𝑎 ≤ E 𝑋 ≤ 𝑏

• If c is a constant, E 𝑐 = c
Expected value rule
Example

• Suppose X is a r.v. about weeks of fX(x)

delay, and has a simple PDF as:
• The penalty of delay is directly 1
proportional to the no. of weeks,
following: 1 2 x weeks
𝑃𝑒𝑛𝑎𝑙𝑡𝑦 = 𝑔(𝑥) = 2000𝑥 2 (𝑖𝑛 $)

• Question: What is the expected penalty?

2
1 2000 3 2
𝐸 𝑝𝑒𝑛𝑎𝑙𝑡𝑦 = න 2000𝑥 2 × 1 𝑑𝑥 = 𝑥 | = $4667
1 2−1 3 1
Example

• Suppose X is a r.v. about weeks of fX(x)

delay, and has a simple PDF as:
• The penalty of delay is directly 1
proportional to the no, of weeks,
following: 1 2 x weeks
𝑃𝑒𝑛𝑎𝑙𝑡𝑦 = 𝑔(𝑥) = 2000𝑥 2 (𝑖𝑛 $)
2
1 2000 3 2
𝐸 𝑔(𝑥) = න 2000𝑥 2 × 1 𝑑𝑥 = 𝑥 | = $4667
1 2−1 3 1

• Let’s try another approach 𝑔 𝐸𝑥 = 2000[1.5]2 = 4500(≠ 4667)

𝑬 𝒈(𝒙) ≠ 𝒈 𝑬 𝒙
Properties of expectation

• If 𝑋 ≥ 0, then E[𝑋] ≥ 0

• If 𝑎 ≤ 𝑋 ≤ 𝑏, then 𝑎 ≤ E 𝑋 ≤ 𝑏

• If c is a constant, E 𝑐 = c

• 𝑬 𝒈(𝒙) ≠ 𝒈 𝑬 𝒙

• E 𝑎𝑋 + 𝑏 = 𝑎E 𝑋 + 𝑏
• Proof: based on the expected value rule
 01. Mean, or expectation

Agenda 02. Variance

03. Skewness

04. Summary
Variance

• Variance: a measure of the spread of the random variable

• Random variable 𝑋, with the mean 𝜇 = 𝐸[𝑋]

• Distance from the mean: 𝑋- 𝜇
• Average distance from the mean?
𝐸 𝑋−𝜇 =𝐸 𝑋 −𝜇 =𝜇−𝜇 =0
• Define:
var 𝑋 = E[(𝑋− 𝜇 )2 ]
Examples

• Variance of a Bernoulli r.v.

var(𝑋) = ෍ 𝑥 − E 𝑋 2 𝑝 (𝑥) = 𝑝(1 − 𝑝)

𝑋
𝑥
Properties of the variance

• var 𝑎𝑋 + b = 𝑎2 var 𝑋
• Review: E 𝑎𝑋 + b = 𝑎E 𝑋 + 𝑏

• var 𝑋 = E 𝑋 2 − (E[𝑋])2
Examples

• Variance of a Bernoulli r.v. var(𝑋) = ෍ 𝑥 − E 𝑋 2

𝑝𝑋 (𝑥) = 𝑝(1 − 𝑝)
𝑥

var 𝑋 = E 𝑋 2 − E 𝑋 2
= 𝑝 − 𝑝2

• Expected value of a uniform r.v. (uniform on a, a+1, … b)

var 𝑋 = E 𝑋 2 − E 𝑋 2
1
= (𝑏 − 𝑎)(𝑏 − 𝑎 + 2)
12
Other measures of spread

• Standard deviation, 𝜎𝑋
• Has the same unit of 𝑋

σ𝑋 = var 𝑋

• Coefficient of variance (c.o.v.), 𝛿𝑋

• A measure of dispersion relative to the central value
• Dimensionless, expressed in %
𝜎𝑋
𝛿𝑋 =
𝜇𝑋
Example

• X is a r.v. about the total load acting on a roof, with a PDF as follows
24 fX (x)
𝑓𝑋 = ቐ 𝑥 3 3≤𝑥≤6
0 𝑒𝑙𝑠𝑒𝑤ℎ𝑒𝑟𝑒

3 6 x
• Calculate the mean, variance, standard deviation, and c.o.v.

• Mean
6
24 −24 6
𝜇𝑋 = න 𝑥 3 𝑑𝑥 = | =𝟒
3 𝑥 𝑥 3
Example

• X is a r.v. about the total load acting on a roof, with a PDF as follows
24 fX (x)
𝑓𝑋 = ቐ 𝑥 3 3≤𝑥≤6
0 𝑒𝑙𝑠𝑒𝑤ℎ𝑒𝑟𝑒

6 3 6 x
• Variance 24
𝑉𝑎𝑟 𝑋 = න (𝑥 − 4)2 × 3
𝑑𝑥 = 𝟎. 𝟔𝟒
3 𝑥
• Standard deviation
𝜎𝑋 = 𝑉𝑎𝑟(𝑋) = 𝟎. 𝟖
• Coefficient of variance
𝜎𝑋 0.8
𝛿𝑋 = = = 𝟐𝟎%
𝜇𝑋 4
 01. Mean, or expectation

Agenda 02. Variance

03. Skewness

04. Summary
Skewness

• Skewness is a measure of the symmetry of PMF/PDF

• Skewness, 𝑆𝑘

𝑆𝑘 = ෍ (𝑥𝑖 − 𝜇𝑋 )3 𝑝𝑋 (𝑥𝑖 ) For discrete X

𝑎𝑙𝑙 𝑥𝑖

∞
𝑆𝑘 = න (𝑥 − 𝜇𝑋 )3 𝑓𝑋 (𝑥) 𝑑𝑥 For continuous X
−∞

• Skewness coefficient (dimensionless) 𝑆𝑘

𝜃= 3
𝜎
Skewness

• Sign of 𝑆𝑘 has a physical meaning

• 𝑆𝑘 = 0: Zero skewness, or symmetrical
• 𝑆𝑘 > 0: Positive skewness, right skew
• 𝑆𝑘 < 0: Negative skewness, left skew

Positive Negative
skewness skewness
 01. Mean, or expectation

Agenda 02. Variance

03. Skewness

04. Summary
 Key statistics of a random variable

Definition Other Implication Other measures

names
∞
Expectation First moment Central • Median
/mean 𝐸 𝑋 = න 𝑥𝑓𝑋 (𝑥) 𝑑𝑥 of inertia tendency • Mode
−∞

Variance var 𝑋 = E[(𝑋−𝜇𝑋 )𝟐 ] Second Spread • Standard deviation

= moment of • Coefficient of
∞
‫׬‬−∞(𝑥 − 𝜇𝑋 )2 𝑓𝑋 (𝑥) 𝑑𝑥 inertia variance
(dimensionless)
Skew 𝑆𝑘 = E[(𝑋−𝜇𝑋 )𝟑 ] = Third Skewness • Skewness coefficient
∞
3
moment of (dimensionless)
න (𝑥 − 𝜇𝑋 ) 𝑓𝑋 (𝑥) 𝑑𝑥 inertia
−∞
Summary

• Two scenarios to calculate the statistics

Scenario 1 Scenario 2:
You have the probability model, i.e., You have collected a bunch of data
you know the probability function
of the random variable, CDF, or PMF
for discrete variables/PDF for
continuous variables
σ𝑖 𝑥𝑖
𝐸 𝑋 = 𝜇𝑋 = ෍ 𝑥𝑖 𝑝𝑋 (𝑥𝑖 ) 𝐸 𝑋 =
𝑛
𝑎𝑙𝑙 𝑥𝑖
∞
𝐸 𝑋 = 𝜇𝑋 = න 𝑥𝑓𝑋 (𝑥) 𝑑𝑥
−∞
Summary

• Key statistics is not

adequate