CE1602 - 2015 Statistics

Statistical Analysis of Experimental Data
Statistical Analysis of Experimental

Course Content
Data – Why?
• Section 1: Least-Squares Regression Analysis --
• Engineers/Scientists frequently have to carry out
experiments. Data are collected during experiments. Curve Fitting
• To draw useful conclusions from experiments, some • Section 2: Linear Correlation of Experiment Data
forms of analysis must be performed on all data. • Section 3: Probability and Random Variables
• The analysis may be a simple verbal appraisal of the test • Section 4: Mean Interval Estimation
results, or it may take the form of a complex theoretical
• Section 5: Measurement Errors and Uncertainty
analysis of the errors involved in the experiment and
matching of the data with fundamental physical Analysis
principles.
• Mathematical models can then be extracted (developed)
from experiment data.
– New principles may be developed from models to explain
phenomenon observed.
Past Exam Questions Past Exam Questions
© Dr, J. Chen, School of Engineering and Design, Brunel University 1 ME1302 Analytical Methods and Skills
Section 1: Least-Squares Curve Fitting Problem 1: Simple Linear Problem

– Fit a Function to Data Set Collected • The following table represents the output (volts) of linear
• Engineers use models, which express the relationships variable differential transformer (LVDT; an electric output
among engineering characteristics, to solve problems. device used for measurement of displacement) for five
These models allow engineers to predict a characteristic length inputs.
of interest, called the response or dependent variables,
given the values of other characteristics, called the L (cm) 0.00 0.50 1.00 1.50 2.00 2.50
independent variables. V (Volt) 0.05
– Voltage and current when in finding characteristics of an electric device.
0.52 1.03 1.50 2.00 2.56
– Force and deflection when investigating stiffness of a beam.
– Voltage and temperature in calibrating a thermocouple. Find a, b for the calibration model: V = aL + b
• Straight line: y = ax + b • Parabola: y = ax 2 + bx + c
• Determine the best linear fit to these data, i.e.
y = a0 + a1 log x + a2 x + a3 sin( x) + a4 e
2 x
find parameters a and b.
• Other example:
y = a0 + a1 xz + a2 x 2 z + a3 sin( z ) + a4 e x cos( z )
Wind Speed Probability

Problem 2: Nonlinear Problem u (m/s) P (%)
0.5 1.03 Problem 3 (More complicated):
1.0 2.00
• The following table gives experimental values of the 1.5
2.1
5.24
6.69
Wind Speed distribution
pressure P of a given mass of gas corresponding to various 2.6
3.1
10.29
7.61
values of the volume V. 3.6
4.1
8.48
6.92
• To investigate the viability of a wind
V (m3) 54.3 61.8 72.4 88.7 118.6 194.0
4.6
5.1
8.98
6.72
turbine, the wind speed data over
5.7 7.07
certain period are collected.
P (N/m2) 61.2 49.2 37.6 28.4 19.2 10.1
6.2
6.7
5.09
6.13
7.2 3.79
• The task is to find the parameters K
and C in the following probability
7.7 3.68
• According to thermodynamic principles, a relationship 7.8 0.57
density function f(u) using the least

8.2 2.55
having the form should exist between the variables. 8.7 2.50
square curve fitting method.

9.3 1.43
9.8 1.22
10.3 0.49
PV γ = C 10.8
11.3
0.57
0.32 K
⎧⎪ ⎛ u ⎞ K ⎫⎪
11.8 0.14 K⎛u⎞
12.3 0.19 p = f (u ) = ⎜ ⎟ exp ⎨ − ⎜ ⎟ ⎬
C , γ are parameters to be found.
– where 　 12.9
13.4
0.09
0.07
u ⎝C ⎠ ⎪⎩ ⎝ C ⎠ ⎭⎪
13.9 0.05
14.4
15.4
0.03
0.01 • Need to find parameters K and C.
17.0 0.01
Linear Models Curve Fitting and the Method of Least Square

• Model: • Step 1: To build the model (determine an equation that connects
y = f(x) variables), the first step is to collect data show corresponding
values of the variables under consideration.
x independent variable
x: independent variable y: dependent variable
y dependent variable
• n pairs of data connected: (x1, y1), (x2, y2), ..., (xn, yn)
y • Step 2: Plot the data points on a rectangular coordinate system.
y = ax + b
Form the plot it is often possible to visualise a smooth curve that
approximate the data. A model candidate is then selected. The most
• Linear model: The Simplest and
the Most Commonly Used Relation common curve is the straight line, i.e. the relationship is linear and
a liner model will be used.
x
Model: y = ax + b
• Step 3: Estimate the two unknown model parameters a and b.
Linear Function: Draw a line of best fit Least Squares: Why?

• It is a common requirement to find the mathematical y = ax + b
function using measurement data. Experiment x y
(1) 1 6 (1) 6 = a + b
y
• One of the most (2) 2 7 (2) 7 = 2a + b
common functions
Δy (3) 3 9 (3) 9 = 3a + b
used is the straight
line, i.e. a linear Δx (4) 4 12 (4) 12 = 4a + b
function. Δy
a=
y=ax+b Δx
• n pairs of data (xi, yi) a =1
b Solution A (Use Solution C (Use a=3
– to fit a straight line experiments (1) & (2) b=5
x experiments (3) & (4) b = 0
through the data. Use
visual inspection and
draw the line by the Solution B (Use a=2
y = ax + b experiments (2) & (3) Solution D (Use a=2
hand. b=3 experiments (1) & (4) b = 4
Least Squares: Why? y = ax + b Least Squares: Why? y = ax + b

y Experiment value (1) 6 = a + b J (a, b) = (6 − (a + b)) 2 + (7 − (2a + b)) 2
yˆ = ax + b Prediction value J: Sum of Square (2) 7 = 2a + b + (9 − (3a + b)) 2 + (12 − (4a + b)) 2
e = y − yˆ Prediction error error
(3) 9 = 3a + b
∂J (a, b)
(4) 12 = 4a + b =0
x y y (A) e (A) y (B) e (B) y (C) e (C) ∂a
Find a and b to minimize J(a,b) ∂J (a, b)
(1) 1 6 6 0 5 1 3 3 =0
∂b
(2) 2 7 7 0 7 0 6 1
J (a, b) = 30a 2 + 20ab − 190a + 4b 2 − 68b + 310
(3) 3 9 8 1 9 0 9 0 ∂J (a, b)
= 60a + 20b − 190 = 0
(4) 4 12 9 3 11 1 12 0 ∂a
a=2
J ∂J (a, b) Solution (P)
10 2 10 = 20a + 8b − 68 = 0 b = 3.5
∂b
Optimisation and Derivative Optimisation and Partial Derivatives

y = f(x)
n
J ( a ,b ) = ∑ (axi + b − yi )
2 • Find minimum:
J i =1
Ø Fix b: dJ ( a ) = 0
da
J* b* Ø Fix a: dJ ( b )
dy b =0
= f ' ( x*) = 0 a* db
dx x*
a
x* x
• Find minimum with partial derivatives:
dy ∂J ( a ,b ) ∂J ( a ,b )
• Find minimum: = f ' ( x*) = 0 =0 =0
dx x* ∂a ∂b
Least Squares: Why? y = ax + b Least Squares for Curve Fitting: Concept

x y y (A) e (A) y (B) e (B) y (C) e (C) y (P) e (P)
• A systematic and appropriate approach is to use the method of least
(1) 1 6 6 0 5 1 3 3 5.5 0.5
Solution A Solution B squares or linear regression to fit the data.
(2) 2 7 7 0 7 0 6 1 7.5 -0.5
a =1 a=2
(3) 3 9 8 1 9 0 9 0 9.5 -0.5 • n pairs of data: (x1, y1), (x2, y2), ..., (xn, yn)
b=5 b=3 y
(4) 4 1
2
9 3 11 1 12 0 11.5 0.5
• The line of best fit: yn
y = ax + b
J 10 2 10 1 y = ax + b
Solution C Solution (P) yi
12 x • With the line of best fit, ei
(P)
a=3 a=2
11
we can predict the value y2 ŷi
10 b=0 b = 3.5 of yi for each value of xi. y1
9 x (A) yˆ i = axi + b xi
8 xn x
x1 x2
7 x • Prediction error
(residual):
6 x • The best fit should minimise the
ei = yˆ i − yi = axi + b − yi prediction error.
1 2 3 4
Least Squares: Optimisation Optimisation and Partial Derivatives

x1 → y1 x1 → yˆ 1 = ax1 + b ax1 + b − y1
d ( f ( a ))
2
data x2 → y2 J ( a ) = ∑ ( f ( a )) dJ ( a )
=∑ = ∑2 f (a )
df ( a )
2
prediction x2 → yˆ 2 = ax2 + b prediction ax2 + b − y2
da da da
# # error #
xn → y n xn → yˆ n = axn + b axn + b − yn ∂J ( a ,b ) n ∂ (axi + b − yi ) n
= ∑ 2(axi + b − yi ) = ∑ 2(axi + b − yi )xi = 0
• Optimisation Criterion for the Least Squares Curve ∂a i =1 ∂a i =1
Fitting: n n n
– The sum of the squared errors for all the data points a ∑ xi2 + b ∑ xi = ∑ xi yi ( A)
i =1 i =1 i =1
n n
J ( a ,b ) = e =
i
2
(axi + b − yi )2 ∂ J ( a ,b ) n ∂ (axi + b − yi ) n
i =1 i =1 = ∑ 2(axi + b − yi ) = ∑ 2(axi + b − yi ) = 0
∂b i =1 ∂b i =1
• We now choose a and b to minimize J .
⎧n 2⎫
n n
a ∑ xi + bn = ∑ yi (B)
min J ( a ,b ) = min ⎨ axi + b − yi ⎬ i =1 i =1
a ,b a ,b
⎩ i =1 ⎭
Least Squares Solution Least Squares Solution: Quality of Fit

a=
( ∑ x )( ∑ y ) − n ( ∑ x y ) = S
i i i i xy
y = ax + b • An appropriate measure of the quality of the fit is the
sum of square for the residuals (prediction errors), the
(∑ x ) − n (∑ x )
2 2
S xx
i i
sum of square error.
( ∑ x )( ∑ x y ) − ( ∑ x ) ( ∑ y ) = y − ax = ∑ y − a∑ x
2 n n n S xy2
( yˆi − yi ) = ( axi + b − yi ) = S yy −
2 2
b= SSE = ei2 =
i i i i i i i
(∑ x ) − n (∑ x )
2 2
i
n i i =1 i =1 i =1 S xx
1
S yy = ∑ ( yi − y ) 2 = ( ∑ yi ) − ( ∑ yi2 )
2
• The resulting line (linear function) is called the least- n
squares best fit to the data represented by (xi, yi).
• Another commonly used measure of how well the least-
S xy = ∑ ( xi − x )( yi − y ) = ∑ xi yi −
1
( ∑ xi )( ∑ yi ) x=
∑x i
squares line will represent the data is called the standard
n n error of estimate, given by:
S xx = ∑ ( xi − x ) 2 = ( ∑ xi ) −
2 1
( ∑ xi2 ) y=
∑y i
S=
SSE
=
yi2 − b yi − a xi yi
n n n−2 n−2
Least Squares Solution (b=0) Least Squares y = ax + b
y = ax ⎧ n 2⎫
Experiment x y ∑ x = 1 + 2 + 3 + 4 = 10
min J ( a ) = min ⎨∑ ( axi − yi ) ⎬
i
(1) 1 6
a a
⎩ i =1 ⎭
(2) 2 7
∑ y = 6 + 7 + 9 + 12 = 34
i
dJ ( a ) n d ( axi − yi ) n ∑ x y = 1× 6 + 2 × 7 + 3 × 9 + 4 ×12 = 95
= ∑ 2 ( axi − yi ) = ∑ 2 ( axi − yi ) xi = 0 (3) 3 9 i i
da da
∑x
i =1 i =1
(4) 4 12 2
= 12 + 22 + 32 + 42 = 30
i
n n
2a ∑ xi2 − 2∑ yi xi = 0 n=4
i =1 i =1
n
a=
( ∑ x )( ∑ y ) − n ( ∑ x y ) = 10 × 34 − 4 × 95 = 2
i i i i
∑yx i i (∑ x ) − n (∑ x ) i
2
10 − 4 × 30 2
i
2
a= i =1
n
∑x 2
i ∑ y − a∑ x 34 − 2 ×10
i =1 b= i i
= = 3.5
n 4
Simple Least Square Example Example: Solution

• The following table represents the output (volts) of linear xi xi2 yi xiyi yi2
variable differential transformer (LVDT; an electric output 0 0 0.05 0 0.0025
device used for measurement of displacement) for five 0.5 0.25 0.52 0.26 0.2704
length inputs. 1 1 1.03 1.03 1.0609
L (cm) 0.00 0.50 1.00 1.50 2.00 2.50 1.5 2.25 1.5 2.25 2.25
2 4 2 4 4.0
V (Volt) 0.05 0.52 1.03 1.50 2.00 2.56 2.5 6.25 2.56 6.4 6.5536
∑xi ∑xi2 ∑yi ∑xiyi ∑yi2
• Determine the best linear fit to these data. 7.5 13.75 7.66 13.94 14.137
• Solution: Define: x=L, y=V
a=
(∑ x )(∑ y ) − n(∑ x y ) = 7.5 × 7.66 − 6 × 13.94 = 0.9977
i i i i
Find a, b so: y = ax + b or V = aL + b (∑ x ) − n(∑ x )
i
2 2
i
7.5 − 6 × 13.75 2
Example Solution Example: Least Square Solution for Nonlinear Model
b=
(∑ x )(∑ x y ) − (∑ x )(∑ y ) = 7.5 × 13.94 − 13.75 × 7.66 = 0.0295
i i i
2
i i
• The following table gives experimental values of the
(∑ x ) − n(∑ x )
2 2
7.5 − 6 × 13.752 pressure P of a given mass of gas corresponding to various
values of the volume V.
i i
•The resulting best fit line is then: V (m3) 54.3 61.8 72.4 88.7 118.6 194.0
y = 0.9977x + 0.0295 P (N/m2) 61.2 49.2 37.6 28.4 19.2 10.1
or V = 0.9977L + 0.0295
• According to thermodynamic principles, a relationship
•The standard error for these data is obtained by:
having the form should exist between the variables.
(∑ y ) − b (∑ y ) − a (∑ x y )
2 PV γ = C
S=
i i i i
n−2 C, γ
– where 　 are constants.
14.137 − 0.0295 × 7.66 − 0.9977 × 13.94
= = 0.0278
6−2
Example Example
a) Find the values of 　 and C. 70
b) Write the equation connecting P and V.

c) Estimate P when V=100 m3. 60
• Solution: PV　 = C 50
log P + γ log V = log C 40
log P = −γ log V + log C 30
Define: x = log V y = log P

20
PV 1.4 = 16000
y = ax + b with a = −γ b = log C
10
• The least squares method can be used to find a and b, then 　 50 100 150 200
and C can be calculated.
Example Solution Example Solution

x = log V y = log P x2 xy Thus: y = -1.4x + 4.2
1.7348 1.7868 3.0095 3.0997
1.7910 1.6946 3.2077 3.0350
1.8597 1.5752 3.4585 2.9294
(a) γ = − a = 1.4 b = log C = 4.2 C = 1.60 × 10 4
1.9479 1.4533 3.7943 2.8309
(b) PV 1.4 = 16000
2.0741 1.2833 4.3019 2.6617
2.2878 1.0043 5.2340 2.2976
∑ x = 116953
. ∑ y = 87975
. ∑ x = 23.0059 ∑ xy = 16.8543
2 (c) When V=100
16000
= 16000 P= = 25.1
(∑ x )(∑ y ) − n(∑ xy ) = 11.6953 × 8.7975 − 6 × 16.8543 = −1.4
1 .4
P100 100 1.4
a=
(∑ x ) − n(∑ x )
2 2
11.6953 − 6 × 23.0059
2
b=
(∑ x )(∑ xy ) − (∑ x )(∑ y ) = 11.6953 × 16.8543 − 23.0059 × 8.7975 = 4.2
2
(∑ x ) − n (∑ x )
2 2
11.6953 − 6 × 23.0059
2
Example Solution Nonlinear functional relationship and

transformation
1.8
• Commonly, test data do not show an even approximately linear
1.7
y = −1.4 x + 4.2 relationship between the dependent and independent variables and
1.6
linear regression is not directly useful.
• In some cases, however, the data can be transformed into a form
1.5
which is linear and a straight line can then be fitted using the method
1.4 of linear regression.
1.3
y=cedx Taking the natural logarithms of each side, we obtain:
1.2
ln(y) = dx + ln(c) Define: y1 = ln(y); a=d; b = ln(c)

1.1
1 y1 = ax + b
0.9
1.7 1.8 1.9 2 2.1 2.2 2.3 2.4 We now have a form in which y1 (ln(y)) is a linear function of x.
Functions/Curves Can be Transformed into Linear Wind Speed

u (m/s)
Cummulative
Probability P
Wind Speed distribution Example
0.5 0.010
1 1
• Hyperbola: y = ax + b ⇒ y = ax + b ⎧⎪ ⎛ u ⎞ K ⎫⎪
1.0 0.030 K
1.5 0.083 K⎛u⎞
2.1 0.150 p = f (u ) = ⎜ ⎟ exp ⎨− ⎜ ⎟ ⎬
2.6
3.1
0.253
0.329
u ⎝C ⎠ ⎪⎩ ⎝ C ⎠ ⎭⎪
• Exponential curve y = ab x ⇒ log y = (log b) x + log a 3.6 0.414
⎧ K
⎫
Integration: P = F (u ) = 1 − exp ⎪⎨ − ⎛ u ⎞ ⎪⎬
4.1 0.483
4.6 0.573
y = ab x + 5 ⇒ log( y − 5) = (log b) x + log a 5.1 0.640 ⎜ ⎟
5.7
6.2
0.711
0.762
⎩⎪ ⎝ C ⎠ ⎭⎪
y = ax b ⇒ log y = b log x + log a
6.7 0.823
• Geometric curve: 7.2 0.861
⎛u⎞
K
7.7 0.898
ln(1 − P ) = − ⎜ ⎟
y = ax b + 9 ⇒ log( y − 9 ) = b log x + log a 7.8
8.2
0.903
0.929 ⎝C ⎠
8.7 0.954
• Geompertz curve: y = pq b
x
⇒ log y = b x log q + log p
9.3
9.8
0.968
0.980 ln( − ln(1 − P )) = K ln(u ) − K ln(C )
10.3 0.985
10.8 0.991
1 1 11.3 0.994
y a x b
• Logistic curve: y= ⇒ = ab x + 5 11.8 0.995
ab + 5
x
y 12.3
12.9
0.997
0.998
y = ax + b
13.4 0.999
y = a0 + a1 (log x) + a2 (log x)2 13.9

14.4
0.999
1.000
15.4 1.000
17.0 1.000
Section 2: Correlation of Experiment Data Model Examples

• An experiment to find relationship between x and y: • Straight line: y = ax + b
– independent variable x is varied systematically
• Parabola: y = ax 2 + bx + c
– the dependent variable y is then measured
y y
• Cubic: y = ax 3 + bx 3 + cx + d
y
• nth order curve: y = a0 + a1 x + a2 x 2 + " + an x n

?
• Other example: y = a0 + a1 log x + a2 x + a3 sin( x) + a4 e
2 x
x x x
(a) line of best fit: Yes (b) line of best fit: NO (c) line of best fit: UNSURE
y = a0 + a1 xz + a2 x 2 z + a3 sin( z ) + a4 e x cos( z )
y = f(x) • Voltage and current when in
finding characteristics of an
x independent variable electric device.
1 1
y dependent variable • Force and deflection when • Hyperbola: y = ax + b ⇒ y = ax + b
investigating stiffness of a beam.
• Voltage and temperature in
calibrating a thermocouple. • Exponential curve y = ab x ⇒ log y = (log b) x + log a
Linear Model: The Simplest and the Linear Correlation Coefficient

Most Commonly Used Model • A statistical parameter which can be used to determine if
an apparent trend is real or could simply be a
y ax b consequence of randomness. ( x1 → y1 )
•Variables: x y ( x2 → y 2 )
• We would like to determine if the value of y depends on a set of n data pairs [(xi, yi), i=1, ..., n] #
the value of x proportionally (linear). • The linear correlation coefficient: ( xn → y n )
y y y n
( xi − x )( yi − y )
n
∑x
n
rxy = i =1
i ∑y i
n n
x= i =1
y= i =1
? n n
x x x ( xi − x ) 2 ( yi − y ) 2
(a) i =1 i=1
line of best fit: Yes (b) line of best fit: NO (c) line of best fit: UNSURE
(a) A strong relationship between y and x. • Measure the strength and direction of a linear
(b) No functional relationship between y and x. relationship between two random variables.
(c) Uncertain.
Values of Correlation Coefficient

Correlation Coefficient Formula
n
• rxy lies in the range -1 to +1. − 1 ≤ rxy ≤ 1
( xi − x )( yi − y ) n
∑x
n
rxy = i =1 i ∑y i
• A value of +1 would indicate a perfectly linear relationship between
the variables with a positive slope (i.e., increasing x results in
n n x= i =1
y= i =1
n increasing y).
( xi − x ) 2
( yi − y ) 2 n
i =1 i =1 • A value of -1 indicates a perfectly linear relationship with negative
slope (i.e., increasing x produces decreasing y).
Alternative Formula: • A value of zero indicates that there is no linear correlation between
the variables.
n xy − x y – Even if there is no correlation, it is unlikely that rxy will be
rxy = exactly zero.
2 2
n x2 − x n y2 − y y y
rxy = 1 rxy = −1
x x
Verify Linear Relationship Critical Value of Correlation Coefficient

α
• For a given sample size, statistical theory can be used to • For two variables and
n
3
0.20
0.951
0.10
0.988
0.05
0.997
0.02
I.000
0.01
1.000
determine if a computed rxy is significant or could be a n data pairs, the
4
5
0.800
0.687
0.900
0.805
0.950
0.878
0.980
0.934
0.990
0.959
6 0.608 0.729 0.811 0.882 0.917
consequence of pure chance. critical values of 7
8
0.551
0.507
0.669
0.621
0.754
0.707
0.833
0.789
0.875
0.834
9 0.472 0.582 0.666 0.750 0.798
If rxy > rt rt is a critical value which correlation 10
11
0.443
0.419
0.549
0.521
0.632
0,602
0.715
0.685
0.765
0.735
can be found from a Table. coefficient, rt is a 12
13
0.398
0.380
0.497
0.476
0.576
0.553
0.658
0.634
0.708
0.684
14 0.365 0.458 0.532 0.612 0.661
– We can presume that y does depend on x in a non- function of the 15

16
0.351
0.338
0.441
0.426
0.514
0.497
0.592
0.574
0.641
0.623
random manner and can expect that a linear number of samples 17

18
0.327
0.317
0.412
0,400
0.482
0.468
0.558
0.543
0.606
0.590
19 0.308 0.389 0.456 0.529 0.575
relationship will offer some approximation of the true and the level of 20
25
0.299
0.265
0.378
0.337
0.444
0.396
0.516
0.462
0.561
0.505
functional relationship. significance, α. 30

35
0.241
0.222
0.306
0.283
0.361
0.334
0.423
0.392
0.463
0.430
40 0.207 0.264 0.312 0.367 0.403
45 0.195 0.248 0.294 0.346 0.380
• A value of |rxy| less than rt, implies that we cannot be confidence level = 1 − α
50
100
0.184
0.129
0.235
0.166
0.279
0.197
0.328
0.233
0.361
0.257
confident that a linear functional relationship exists. 200 0.091 0.116 0.138 0.163 0.180
Table: Minimum Values of the Correlation

Coefficient for Significance Level
Example x2 y2
Solution x y xy
• It is thought that lap times for a race car depend on the 40 65.3 2612.0 1600 4264.09
ambient temperature. The following data for the same car 47 66.5 3125.5 2209 4422.25
with the same driver were measured at different races: 55 67.3 3701.5 3025 4529.29
62 67.8 4203.6 3844 4596.84
Ambient 66 67 4422.0 4356 4489.00
40 47 55 62 66 88
temperature (°F)
88 66.6 5860.8 7744 4435.56
Lap time (s) 65.3 66.5 67.3 67.8 67.0 66.6 ∑ ∑ ∑ ∑ ∑
70 358.00 400.50 23925.4 22778 26737.03
• Does a linear
n ∑ xy − ( ∑ x )( ∑ y )
68
relationship exist
rxy =
between these two
66
n∑ x2 − (∑ x ) n∑ y 2 − (∑ y )
2 2
variables (with 95% 64
confidence)? 62 6 × 23925.4 − 358 × 400.5

Ambient temperature ( oF) = = 0.4015
60
30 50 70 90 (6 × 22778 − 358 ) (6 × 26737.03 − 400.5
2 2
Example: Solution Example

• For a 95% confidence level, α=1-0.95=0.05. • An experiment is used to investigate the relationship
between the spring vibration frequency and the mass
• For six pairs of data, from Table 1, we obtain a applied.
value of rt, of 0.811 .
• When different mass is applied
to the system, we can measure
rxy = 0.4013 < 0.811 = rt different spring vibration
Laser
Measure
vibration
frequency. frequency
• Since rxy. is less than rt, there is no linear • Examine the relationship
relationship exists between lap time and ambient between the vibration frequency
temperature. (f) with the mass (m).
Mass (m, kg) 2.0 7.6 13.2 18.8 24.4 30

Frequency (f, Hz) 8.5 4.3 3.2 2.8 2.4 2.3
Example Example x (m) y (f) xy x2 y2

• From the plot of f vs m, we can find that the relationship 2.0 8.5 17.0 4.0 72.25
between f and m is not linear, i.e. a straight-line cannot be 7.6 4.3 32.68 57.76 18.49
drawn across data points. 13.2 3.2 42.24 174.24 10.24
18.8 2.8 52.64 353.44 7.84
10 24.4 2.4 58.56 595.36 5.76
30.0 2.3 69.0 900.0 5.29
8
∑ ∑ ∑ ∑ ∑
Frequency (f, Hz)
6 96 23.5 272.12 2084.8 119.87
n ∑ xy − ( ∑ x )( ∑ y )
4
rxy =
n∑ x 2 − ( ∑ x ) n∑ y 2 − ( ∑ y )
2 2 2
0
0 10 20 30 6 × 272.12 − 96 × 23.5
= = −0.8406
(6 × 2084.8 − 96 ) ( 6 × 119.87 − 23.5
Mass (m, kg)
2 2
)
Example Example 1
x= y= f
• For a 99% confidence level, α=1-0.99=0.01. Transformation m
• For six pairs of data, from Table 1, we obtain a m 2.0 7.6 13.2 18.8 24.4 30.0
value of rt, of 0.917 . x=
1
0.7071 0.3627 0.2752 0.2306 0.2024 0.1826
m
rxy = 0.8406 < 0.917 = rt y=f 8.5 4.3 3.2 2.8 2.4 2.3
10.0
• Since rxy. is less than rt, a linear relationship • A linear

between f and m does not exist.
7.5
relationship
Frequency (f, Hz)
5.0
exists.
2.5
0
0.15 0.30 0.45 1 0.60 0.75
m
Example x ( m1 ) y (f) xy x2 y2 Example

0.7071 8.5 6.0104 0.5000 72.25 • For a 99% confidence level, α=1-0.99=0.01.
0.3627 4.3 1.5598 0.1316 18.49
0.2752 3.2 0.8808 0.0758 10.24
• For six pairs of data, from Table 1, we obtain a
0.2306 2.8 0.6458 0.0532 7.84 value of rt, of 0.917 .
0.2024 2.4 0.4859 0.0410 5.76
0.1826 2.3 0.4199 0.0333 5.29 rxy = 0.9996 > 0.917 = rt
∑ ∑ ∑ ∑ ∑
1.9607 23.5 10.0025 0.8348 119.87 • The relationship between y and x is linear, i.e. the
n ∑ xy − ( ∑ x )( ∑ y )
frequency f is linearly proportional to 1
rxy = m
n∑ x 2 − (∑ x ) n∑ y 2 − ( ∑ y )
2 2
G ⎛ 2 G ⎞ 1
6 × 10.0025 − 1.9607 × 23.5 f = kd 2 = ⎜ kd ⎟
= = 0.9996 mnD 3 ⎝ nD 3 ⎠ m
(6 × 0.8348 − 1.9607 ) ( 6 × 119.87 − 23.5
2 2
)
Correlation Coefficient: Comments Correlation Coefficient: Comments

• It is not necessary for the functional • A single bad data point can have a strong effect
relationship to be linear for a significant on the value of rxy
correlation coefficient to be found. – If possible, bad data point should be identified before
• On the other hand, some functional evaluating the coefficient.
relationships, while very strong (a multi-valued • It is a mistake to conclude that a significant value
circular function being an example) will result of the correlation coefficient implies that change
in a very low value of rxy. in one variable causes the other variable to
change.
– Causality should be determined from other
knowledge about the problem.
What type of problems do we want to solve?
– A production line produces millions of

3.1: Dealing with randomness (uncertainty):
similar ball bearings per day.
Repeated experiments/measurements
– How to find the mean (average) diameter?
3.2: Dealing with large number of data: Organise
• Measure all of them and calculate mean.
data by groups
• Measure 100 and calculate mean.
3.3: Probability, random variables and • Measure two dozen and calculate mean.
probability distribution – Are we confident with the result if not all of
them are measured? And in what degree?
What type of problems do we want to solve? Inevitability of Randomness (Uncertainty)

• A medical laboratory analyses specimens of a Example: Measure the temperature outside in an
pharmaceutical product to determine the concentration of Autumn day.
the active ingredient. 1st attempt: The temperature is about 11ºC
– The chemical analysis procedure is not precise so that repeated 2nd attempt: The temperature could be between 10ºC to 12 ºC
measurements on the same specimen will give different results.
3rd attempt: Measurement using a
– The standard deviation for the procedure is σ=0.02 grams per litre. thermal meter, 10.5 ºC
– The laboratory analyzes each specimen 36 times and reports the
mean result. More accurate: 10 ºC ≤ T ≤ 11 ºC
– The laboratory has been asked to evaluate the claim that the Or: T = 10.5 ± 0.5 ºC
concentration of the active ingredient is 0.86 grams per litre.
10.5 ºC one can ask why not 10.5001 ºC
36 test results, Accept/Reject claim of active
mean value is ingredient 0.86 with 99% confidence Other random factors (e.g. wind, small change
0.8404. level. in location etc) all contribute to the uncertainty.
Inevitability of Randomness (Uncertainty) 3.1: Dealing with randomness (uncertainty):

10 20 30 40 50 60
Repeated experiments/measurements
Measure length
• Random features are observed in virtually all
with a ruler
experiments.
Probable range: – Random variations in measurement data.
Best estimate of length 46.5 mm to 47.5mm
= 47 mm – Regardless of the care taken in obtaining a set of data
Different observers may not from independent measures under identical
agree with the precise estimate. conditions, random variations in the data values will
normally occur.
Best estimate of time
= 9:05:35 • An experiment or trail is described as random if the
Probable range: result or outcome of the experiment is not predictable or
9:05:34.5 to 9:05:35.5 contains uncertainty.
Uncertainty introduced by instrument limitation!!
Dealing with randomness (uncertainty): Measure of central tendency: Mean (Average)

Repeated experiments/measurements • A set of n data collected are:
• The best way to deal with randomness is to x1, x2, ..., xn
“repeat” experiments many times and collect a
• The mean is equal to the sum of all the observed values
large number of data. divided by the total number of the observations.
– Statistical measures are required to deal with data
n
collection. Mean
xi
x + x2 + xn
• To find the value for a variable x (e.g. x= 1 = i =1
component length, room temperature etc), it is n n

measured (tested) n times. • 5 different numbers: 8, 3, 5, 12, 10. Mean (Average):
• A set of n data collected are: 8 + 3 + 5 + 12 + 10 38
x= = = 7 .6
x1, x2, ..., xn 5 5
The Weighted Mean (Average) Example

• Sometimes we associate with the numbers x1, • If an examination in a course is weighted 3
x2, ..., xn certain weighting factors (or weights) times as much as a lab report and a student has
w1, w2, ..., wn, depending on the significance an examination mark of 85 and lab report
or importance attached to the number. The marks of 70 and 90, the final mark is:
weighted average:
1 × 70 + 1 × 90 + 3 × 85 415
n x= = = 83
w x + w2 x 2 + wn x n wi x i 1+1+ 3 5
x= 1 1 = i =1
w1 + w2 + + wn n
wi
i =1
Data Spread & Standard Deviation Standard Deviation: Definition

• The extent to which n measurement values are
• The degree to which numerical data tend to spread
about an average value is called the dispersion, or
spread about the average can be expressed by
variation, of the data. the standard deviation S:
• The most common measure of dispersion is the
standard deviation. ( x1 − x ) 2 + ( x2 − x ) 2 + + ( xn − x ) 2
S=
n −1
n
( xi − x ) 2 n( xi2 ) − ( xi ) 2
= i =1
=
Mean n −1 n( n − 1)
Experiment Example xi xi − x ( x i − x )2
5.3 -0.3130 0.0980
• The length of a steel bar is measured 10 5.73 0.117 0.0137
times and results are: 6.77 1.157 1.3386
5.26 -0.353 0.1246
x(cm) 5.30 5.73 6.77 5.26 4.33 5.45 6.09 5.64 5.81 5.75
4.33 -1.283 1.6461
5.45 -0.163 0.0266
• Average: xi 56.13 6.09 0.477 0.2275
x= = = 5.613 ( cm ) 5.64 0.027 0.0007
n 10
• Standard Deviation: 5.81 0.197 0.0388
5.75 0.137 0.0188
( xi − x ) 2 3.5334 ∑x ∑(x − x)
2
S= = = 0.627 (cm) i i
n −1 10 − 1 56.13 3.5334
Standard Deviation for Weighted Mean 3.2 Dealing with large number of data:
• The numbers x1, x2, ..., xn are associated with Organise data by groups
weighting factors (or weights) w1, w2, ..., wn.
n • When summarising large number of raw data, it is
wi xi often useful to distribute the data into groups (classes),
w x + w2 x2 + wn xn or categories, and to determine the number of
x= 1 1 = i =1
individuals belonging to each case, called the class
w1 + w2 + + wn n
wi frequency.
i =1
• A tabular arrangement of data by classes together with
n the corresponding class frequencies is called frequency
n1 wi ( xi − x ) 2 Here n1 is the number of distribution.
S= i =1
n
non-zero weights.
(n1 − 1) wi
i =1
Frequency Frequency distribution

• The frequency for particular class (category) is the • A frequency table or frequency distribution is a way
number of times the category appears in the data set. of summarizing a set of data.
• The relative frequency for particular category is the • It is a record of how often each value (or set of
fraction or proportion of the time that the category values) of the variable in question occurs. The table
appears in the data set.
displays the possible categories along with the
associated frequencies or relative frequencies.
frequency
Relative Frequency = • A frequency table can be used to summarize all types
total number of observation in the data set
of data.
• When the table includes relative frequencies, it is
sometimes referred to as a relative frequency
distribution.
Frequency Distribution (Temperature

Measurement Example) • The variable range defining a class, such as [1085, 1090), is called
a class interval.
Temperature is measured 60 times and the values are all
between 1085 to 1120. • The end numbers, 1085 and 1090 are called class boundaries; the
smaller number (1085) is the lower class boundary, and the larger
Temperature Interval (°C) Number of Measurement number (1090) is the upper class boundary.
(Frequency)
• The class mark (mid-point) is the middle point of the class interval
1085 ≤ T <1090 1
and is obtained by adding the lower and upper class boundaries and
1090 ≤ T <1095 3 dividing by 2.
1095 ≤ T <1100 12 Class interval
1100 ≤ T <1105 21 1085 1087.5 1090

1105 ≤ T <1110 14 Lower Class
Upper
class class
1110 ≤ T <1115 7 boundary
mid-point
boundary
1115 ≤ T <1120 2
lower class boundary + upper class boundary
Total 60 class mid-point =
2
Class Frequency
Class mid-
Number of Measurements
Class Frequency 1085 ≤ T <1090 1 20
point 12
1090 ≤ T <1095
5
3 16 P= = 0 .2
1085 ≤ T <1090 1087.5 1 60
1095 ≤ T <1100 12 12
1090 ≤ T <1095 1092.5 3
1095 ≤ T <1100 1097.5 12 1100 ≤ T <1105 21 8 12
1100 ≤ T <1105 1102.5 21 1105 ≤ T <1110 14 4
1105 ≤ T <1110 1107.5 14 1110 ≤ T <1115 7

1085 1090 1095 1100 1105 1110 1115 1120
1110 ≤ T <1115 1112.5 7 1115 ≤ T <1120 2 Temperature (° C )
1115 ≤ T <1120 1117.5 2 Sum Total 60
• A histogram is a tabulation of the frequency-

of-occurrence of a random variable within
present intervals.
• Example results of 21 students: 38, 39, 42, 44, 48,

• A histogram is a way of summarising data that are 49, 50, 51, 52, 54, 54, 56, 59, 60, 64, 65, 65, 65,
measured on an interval scale.
68, 72, 76.
• It divides up the range of possible values in a data set
– Plot histogram over the range [30 80] with 5 classes.
into classes or groups.
7
• The histogram is only appropriate for variables whose 6
values are numerical and measured on an interval scale. C la ss C la ss
F re. 5
It is generally used when dealing with large data sets. in terva ls m a rk
4
• A histogram can also help detect any unusual 3 0 -4 0 35 2
4 0 -5 0 45 5 3
observations (outliers), or any gaps in the data set.
5 0 -6 0 55 7 2
6 0 -7 0 65 5 1
7 0 -8 0 75 2
0
35 45 55 65 75
Class Rel. Fre. • The normalised relative frequency is the relative

1085 ≤ T <1090 1.67% 33.3% frequency divided by the class interval.
Relative Frequency
1090 ≤ T <1095 5% 26.7%
P = 0. 2
5
1095 ≤ T <1100 20% 20%

0.0667 P = 0. 2 • The probability of
Relative Frequency
1100 ≤ T <1105 35% 0.0533
5
the random
Normalised
13.3% Area=0.2
1105 ≤ T <1110 23.33% 6.7%
12
60
= 20% 0.004 variable in each
1110 ≤ T <1115 11.67% 0.0267 class is the area of
1115 ≤ T <1120 3.33% 1085 1090 1095 1100 1105 1110 1115 1120
the corresponding
Temperature (° C ) 0.0133
Sum Total 1 bar in the
• The relative frequency of a class is the frequency
1085 1090 1095 1100 1105 1110 1115 1120
Temperature (° C )
histogram.
of the class divided by the total frequency of all
classes and generally expressed as a percentage.
Frequency Polygon Mean and Standard Deviation of Grouped

• A frequency polygon is a line graph of the class Data
frequency plotted against the class mark. • When the experiment data are presented in the form
of grouped data, we need to find the way to
– It can be obtained by connecting the midpoints of the determine mean and standard deviation.
tops of the rectangles in the histogram.
• An experiment which tests breaking strength of 200
0.0667 P = 0.2 pieces of cord.
Relative Frequency
0.0533 Aera=0.2
Normalised
D a ta R anges N um bers
0.004
205 – 209 2
0.0267
210 – 214 8
0.0133 215 – 219 20
220 – 224 39
1085 1090 1095 1100 1105 1110 1115 1120
Temperature (° C ) 225 – 229 42
Group Midpoint Frequency

Example: Breaking Strength of Cord
Group 1
Mean and Standard Deviation of Grouped Data
m1 f1
• An experiment which tests breaking strength
of 200 pieces of cord.
Group j mj fj
fm 45745
• Average: x= = = 228.73 ( N )
Group K mK fK
f 200
• Standard Deviation:
K
2
f jm j f m x f (m − x ) 2 18429.88
j =1 fm S S= = = 9.62 ( N )
x= = f −1 200 − 1
n n n
n = f 1+ f 2 + + fK
Group m f m−x f (m − x ) 2 3.3 Probability and Random Variables

fm
200 – 204 202 1 202 -26.73 714.49
205 – 209 207 2 414 -21.73 944.39
210 – 214 212 8 1696 -16.73 2239.14 • An experiment or trail is described as random if the
215 – 219 217 20 4340 -11.73 2751.86 result or outcome of the experiment is not predictable or
220 – 224 222 39 8658 -6.73 1766.42 contains uncertainty.
225 – 229 227 42 9534 -1.73 125.70 • The theory of probability is essential in the modelling
230 – 234 232 35 8120 3.27 374.25 and analysis of random experiments.
235 – 239 237 26 6162 8.27 1778.22
• Probability is a measure of the likelihood that a
240 – 244 242 15 3630 13.27 2641.39
particular event will occur in any one trail (or
245 – 249 247 8 1976 18.27 2670.34
experiment), carried out in prescribed condition.
250 – 254 252 3 756 23.27 1624.48
255 – 259 257 1 257 28.27 799.19
∑ 200 45745 18429.88
Sample Space and Events

• The first task with random experiments is to define the • The outcome (X) of throwing a
list, or set, of all possible outcomes which is know as honest dice.
the sample space. Sample Space (All possible outcomes):
• A collection of elements satisfying a common
requirement in a sample space is known as an event. S = { X = 1, 2,3, 4,5, 6}
An event is a subset of the sample space.
Events A = X | X = 2 = 2 A2 = X | X ≥ 5 = {5, 6}
1
S A3 = X | X is even = {2, 4, 6} A4 = X | X > 10 = {}
A A5 = X |1 ≤ X ≤ 6 = {1, 2,3, 4,5, 6}
Determination of Probability
• Probability is a numerical value expressing the • Probabilities: S = { X = 1, 2,3, 4,5, 6}
likelihood of occurrence of an event relative to all
possibilities (sample space) for an experiment. 1
• Probability is a mathematical quantity that is linked to A1 = X | X = 2 = 2 P ( X = 2) = ,
6
the frequency with which a certain phenomenon
occurs after a large number of trial. 2
A2 = X | X ≥ 5 = {5, 6} P( X ≥ 5) = ,
P(A ) =
number of possible outcome favorable to the event A 6
total number of possible equally likely outcomes
3
An event (A) is an subset of sample space S.
A3 = X | X is even = {2, 4, 6} P ( X is even) =
6
• Probabilities: S = { X = 1, 2,3, 4,5, 6} 349

P1 =
0 400
A4 = X | X > 10 = {} P( X > 10) = =0 π = 4 P1 = 3.49
6
πr 2 π
P= = 1983
A5 = X |1 ≤ X ≤ 6 = {1, 2,3, 4,5, 6} πr 2
4r 2
4 P2 =
2500
r π = 4P π = 4 P2 = 3.1728
6
P( 1 ≤ X ≤ 6 ) = = 1 7858
6 4r 2 P3 =
10000
π = 4 P3 = 3.1432
• Probability is the chance of occurrence of an event in • The probability of an event happening usually lies
an experiment. somewhere between two extreme values, 0 and 1.
• The probability is obtained by dividing the number 0≤ P≤1
of successful occurrence by the total number of trails.
• Probability is based upon relative frequency. • Success: an event occurs A e.g . x ≥ 0
• Values of probability lie between 0 and 1, where 0 • Fail: an event fails to occur A e.g . x < 0
represents an absolute impossibility and 1 represents
an absolute certainty. P( A) + P ( A) = 1
• The probability of an event happening usually lies
somewhere between two extreme values, 0 and 1. P ( A) = 1 This event A definitely occurs A S
0≤ P≤1 P( A) = 0 This event A cannot happen

A
• Probability Addition Law: A and B are mutually exclusive

events, the probability of either event A or event B occurs Dice example: S = { X = 1, 2,3, 4,5, 6}
(not both) is: P ( A or B ) = P ( A) + P ( B ) A: X = 2 = 2 The outcome of dice throw is 2.
Mutually exclusive
The outcome of dice throw is not less
events are events which B B : X ≥ 5 = 5, 6 than 5.
cannot occur together. S
A (A or B): The outcome of dice throw
Dice example: ( A or B) = 2,5, 6 is either 2 or not less than 5.
A: X = 2 = 2 The outcome of dice throw is 2.
P( A or B) =
3
mutually
6 exclusive:
The outcome of dice throw is not less
B : X ≥ 5 = 5, 6 than 5. no
Check 1 2 overlapping
P ( A or B) = P( A) + P( B) = +
mutually exclusive: no overlapping 6 6
S = { X = 1, 2,3, 4,5, 6}
• Probability Addition Law: A B A : X is odd= 1,3,5
and B are NOT mutually
B : X ≥ 5 = 5, 6
exclusive events, the
S A
probability of either event A ( A or B) = ( X ≥ 5) or ( X is odd) = 1,3,5, 6
or event B occurs (not both) The outcome of dice throw is either odd or not less than 5.
is: ( A and B ) = ( X ≥ 5) and ( X is odd) = 5
P ( A or B ) = P ( A) + P( B ) − P( A and B)
The outcome of dice throw is odd and not less than 5.
The outcome of dice throw is either odd or not less than 5. 4 1
P ( A or B) = P( A and B) =
A : X is odd P( A and B) = P(outcome is odd and not less than 5) = 1 6 6
6 Check:
B: X ≥5
3 2 1 4 3 2 1 4
P ( A or B) = + − = A and B = 5 P( A) + P( B) − P( A and B) = + − = = P( A or B)
6 6 6 6 6 6 6 6
A or B = {1,3,5, 6}
• Independent event: Events are independent when the • Independent event: Events are independent when the
occurrence of an event does not affect the occurrence occurrence of an event does not affect the occurrence
of the second event. of the second event.
Event A = {throwing a six} when rolling a die A bag of 100 balls (80 white and 20 black). Student 1
Event B = {drawing an ace} from a pack of 52 cards pick up one ball randomly and then Student 2 randomly
pick one ball.
1 4 1
P ( A ) = P( X = 6) = P ( B ) = P(Y = ace) = = Event A = {Student 1 picked black ball}
6 52 13
Event B = {Student 2 pick black ball}
• Multiplication law of probabilities: The probability of Event A and Event B are independent if Student 1 puts back
the occurrence of both events A and B is the ball before Student 2.
1 1 1 Event A and Event B are not independent if Student 1 does not
P ( A and B ) = P( A) P( B) = × = put back the ball when Student 2 picks the ball.
6 13 78
“Experiment” Outcomes and

Randomness and Random Variables in
Random Variables
Experiments • The outcome of an experiment need not be a
• Random features are observed in virtually all number, for example, the outcome when a coin
is tossed can be 'heads (H)' or 'tails (T)'.
experiments. – Experiment is a term used in statistics, e.g., through
– Random variations in measurement data. a coin is an experiment, the process of taking a
• Regardless of the care taken in obtaining a measurement in a lab test is also an experiment.
set of data from independent measures under
identical conditions, random variations in the
Toss two coins: S = HH HT TH TT
data values will normally occur.
– The measured variable is treated as a random Throw a dice: S=
variable.
Random Variables Random Variables

• We often want to represent outcomes as numbers.
• A random variable is a function that associates a Toss two coins: S = HH HT TH TT
unique numerical value with every outcome of an
experiment.
– The value of the random variable will vary from trial to trial Random variable: X=
as the experiment is repeated.
– A random variable has a single numerical value for each
outcomes of probability experiment.
Throw a dice: S=
S X Random variable: a
mapping from sample
space S to real line
s Random variable: Y=
0 1 2 3 4 5
Random variables: X, Y, Z…….
Their values: x, y, z…….
Discrete Random Variables Continuous Random Variables

• A discrete random variable is one in which all • A continuous random variable is a random
possible values can be counted or listed.
variable capable of assuming all the values in an
– The outcome of throwing a dice:
interval or several of real numbers.
X = x = {1, 2,3, 4,5, 6} – Temperature measurement: T ∈ [19.5, 21]
• The probability distribution of a discrete random • The probability of a continuous random variable x
variable x is a list or table of the distinct values of x is described by the probability density function
and the probabilities associated with those values. (PDF).
x 1 2 3 4 5 6
P( x ) = 1
P(x) 1/6 1/6 1/6 1/6 1/6 1/6
Probability as An Area under the Probability

Density Function
• The probability of the random variable X within the
• A typical problem associated with continuous variables interval [x1, x2] x2
would be to determine the probability of a random P( x ≤ X ≤ x ) =
1 f ( x)dx
2 x1
variable (X) within certain interval (x1, x2).
– Probability as an area under the probability density curve. • When the limits of integration are extended to negative
Temperature and positive infinity: +∞
f(x)
between 19.5°C to P( −∞ ≤ X ≤ +∞) = f ( x) dx = 1
P ( x1 ≤ X ≤ x2 ) −∞
20.5°C
This covers all possibilities!! • The total area under the
x2
P ( x1 ≤ X ≤ x2 ) = f ( x)dx pdf curve is 1.
x1
• The probability density function is not itself a
pdf
Voltage between probability, only areas under the probability
x1 x2
x 225V to 235V density curve are probabilities.
• For entire range of possible value of X (from a to b), the • Example: The lifetime of an electronic component (in
probability density function is constant. thousands of hours) is a continuous random variable
1 with pdf: ⎧ 1 − 2x
f ( x) = ⎧ 1 1
b−a ⎪ a≤ x≤b ⎪ e x≥0
f ( x) = ⎨ b − a f ( x) = ⎨ 2
⎪⎩ 0 ⎪ 0 x<0
otherwise ⎩ 0 x
0 a b Find the proportion of components that last longer than 6000 hours.
Random variable’s value, x 6
P ( X > 6) = 1 − P ( X ≤ 6) = 1 − f ( x) dx
The measurement of length use an ordinarily ruler, the round- −∞
off error X, could take any value between -0.05 cm to 0.05cm. 6
6 1 − 2x ⎡ − 2x ⎤
10 20 30 40 50 = 1− e dx = 1 − ⎢ −e ⎥
60 0 2 ⎣ ⎦0
⎧10 −0.05 ≤ x ≤ 0.05
f ( x) = ⎨ 6
= 1 − ⎡⎣ −(e −3 − e0 ) ⎤⎦ = e −3 ≈ 0.05
⎩0 otherwise 0
Approximately, 1 in 20 components lasts longer than 6000 hours.
• For a random variable with the pdf f(x), the man and ⎧α e −α x x≥0
f ( x) = ⎨
standard deviation can be determined as: ⎩ 0 x<0
Mean: +∞ +∞ +∞ +∞ de −α x
μ= xf ( x) dx μ= xf ( x)dx = α xe −α x
dx = − x dx
−∞ −∞ 0 0 dx
Standard +∞ Mean: +∞ +∞ dx
deviation: σ= ( x − μ ) 2 f ( x) dx = − ⎡⎣ xe −α x ⎤⎦ + e −α x dx (integrating by
−∞ 0 0 dx parts)
+∞
= 0+ e −α x dx
0
1 +∞ 1
=− ⎡⎣e −α x ⎤⎦ =
α 0 α
+∞ 1
σ= ( x − μ ) 2 f ( x) dx =
−∞ α
0.8 σ=0.5
• The normal (or Gaussian) distribution function is a mean
σ=0.6
0.6 μ=2
simple distribution function that is useful for many
σ=1
common problems involving random variables.
f(x)
0.4
σ=2
1 ⎡ − ( x − μ )2 ⎤ μ mean 0.2
f (x) = exp ⎢ ⎥
2 πσ ⎣ 2σ
2
⎦ σ s tan dard deviation
0
-1 0 1 2 3 4 5
Random variable, x
1 X ~ N (μ, σ 2 ) • There is only one maximum and it occurs at the mean value.
f(x)
2π σ • The density function is symmetrical about the mean value.
• The width of the density function is directly proportional to the
standard deviation, σ.
mean μ • The maximum value of the density function is inversely
Random variable, x
proportional to the standard deviation σ.
t F
• The probability that a random variable falls within one or more ν=∞
standard deviations (σ’s) of the mean. 0.4 N (0,1) ν=5
ν=1 v is degree of freedom
f(v,t)
Range Probability f (v, t )
f(x) 1 0.2
1σ 68.26% 2πσ
t the value for a
2σ 95.44% random variable
0
0 t
3σ -3 -2 -1 1 2 3
99.74%
1.645σ 90% x
1.96σ 95% μ−3σ μ−2σ μ−σ μ μ+σ μ+2σ μ+3σ
68.26%
2.33σ 98% 95.44% d1 and d2 are
99.74% degrees of
2.58σ 99%
freedom
3.3σ 99.9%
Determine Probability using Normal Distribution Standard

Normal Distribution
Function Normal Distribution
• The probability of the random variable X falls between x1 X ~ N (μ, σ 2 ) Z ~ N (0,1)
and x2 is: 1 ⎡ −( x − μ ) 2 ⎤ 1 ⎡ − ( x − μ )2 ⎤ 1 ⎡ − z2 ⎤
P ( x1 ≤ X ≤ x2 ) = ∫ f ( x) dx = ∫
x x
f (x) = f ( z) =
2 2
exp ⎢ ⎥ dx exp ⎢ ⎥ exp ⎢ ⎥
2πσ 2σ 2 2 πσ ⎣ 2σ 2π
2
x1 x1
⎣ ⎦ ⎦ ⎣ 2 ⎦
when x ~ N ( μ ,σ 2 ) • The integral cannot
f(x)
P ( x1 ≤ X ≤ x2 )
0.4
P( z1 ≤ Z ≤ z 2 )
f(z)
f(x) be evaluated 0.2
P ( x1 ≤ X ≤ x2 ) analytically, and as z1 z2
a result, the pdf 0
-3 -2 -1 0 1 2 3
x
integration can be x1 x2 Random variable, z
performed • Probabilities standardized

• Probabilities relate to μ and
pdf numerically. σ.
and can be listed by a
standard normal distribution
x1 x2
x table.
Uses of Standard Normal Distribution Table

0.00 0.01 0.02 0.03 0.04 0.05 0.06 0.07 0 08 0 09
Z ~ N (0,1)
z
0.0 .0000 .0040 .0080 .0120 .0160 .0199 .0239 .0279 .0319 .0359
• The values of random 0.1

0.2
.0398
.0793
.0438
.0832
.0478
.0871
.0517
.0910
.0557
.0948
.0596
.0987
.0636
.1026
.0675
.1064
.0714
.1103
.0753
.1141
0.3 .1179 .1217 .1255 .1293 .1331 .1368 .1406 .1443 .1480 .1571
variable z are shown in the 0.4
0.5
.1554
.1915
.1591
.1950
.1628
.1985
.1664
.2019
.1700
.2054
.1736
.2088
.1772
.2123
.1808
.2157
.1844
.2190
.1879
.2224
• Find: P (0 ≤ Z ≤ 2.24)
left column and the top
0.6 .2257 .2291 .2324 .2357 .2389 .2422 .2454 .2486 .2517 .2549
0.7 .2580 .2611 .2642 .2673 .2704 .2734 .2764 .2794 .2823 .2852
0.8 .2881 .2910 .2939 .2967 .2995 .3023 .3051 .3078 .3106 .3133
row. 0.9 .3159 .3186 .3212 .3238 .3264 .3289 .3315 .3340 .3365 .3389
1.0
1.1
.3413
.3643
.3438
.3665
.3461
.3686
.3485
.3708
.3508
.3729
.3531
.3749
.3554
.3770
.3577
.3790
.3599
.3810
.3621
.3830
z 0.04 0.05 P (0 ≤ Z ≤ 2.24)
• The top row serves as the 1.2
1.3
.3849
.4032
.3869
.4049
.3888
.4066
.3907
.4082
.3925
.4099
.3944
.4115
.3962
.4131
.3980
.4147
.3997
.4162
.4015
.4177
second decimal point of

1.4
1.5
1.6
.4192
.4332
.4452
.4207
.4345
.4463
.4222
.4357
.4474
.4236
.4370
.4484
.4251
.4382
.4495
.4265
.4394
.4505
.4279
.4406
.4515
.4292
.4418
.4525
.4306
.4429
.4535
.4319
.4441
.4545
= 0.4875
1.7 .4554 .4564 .4573 .4582 .4591 .4599 .4608 .4616 .4625 .4633
the first column. 1.8
1.9
.4641
.4713
.4649
.4719
.4656
.4726
.4664
.4732
.4671
.4738
.4678
.4744
.4686
.4750
.4693
.4756
.4699
.4761
.4706
.4767
2.0
2.1
.4772
.4821
.4778
.4826
.4783
.4830
.4788
.4834
.4793
.4838
.4798
.4842
.4803
.4846
.4808
.4850
.4812
.4854
.4817
.4857
2.0 .4793 .4798
2.2
2.3
2.4
.4861
.4893
.4918
.4864
.4896
.4920
.4868
.4898
.4922
.4871
.4901
.4925
.4875
.4904
.4927
.4878
.4906
.4929
.4881
.4909
.4931
.4884
.4911
.4932
.4887
.4913
.4934
.4890
.4916
.4936
2.1 .4838 .4842
2.2 .4875 .4878
a = 2.15
2.5 .4938 .4940 .4941 .4943 .4945 .4946 .4948 .4949 .4951 .4952
2.6 .4953 .4955 .4956 .4957 .4959 .4960 .4961 .4962 .4963 .4964
2.7 .4965 .4966 .4967 .4968 .4969 .4970 .4971 .4972 .4973 .4974
2.8
2.9
.4974
.4981
.4975
.4982
.4976
.4982
.4977
.4983
.4977
.4984
.4978
.4984
.4979
.4985
.4979
.4985
.4980
.4986
.4981
.4986
2.3 .4904 .4906
3.0 .4987 .4987 .4987 .4988 .4988 .4989 .4989 .4989 .4990 .4990
3.1 .4990 .4991 .4991 .4991 .4992 .4992 .4992 .4992 .4993 .4993
3.2 .4993 .4993 .4994 .4994 .4994 .4994 .4994 .4995 .4995 .4995
3.3 .4995 .4995 .4995 .4996 .4996 .4996 .4996 .4996 .4996 .4997
3.4 .4997 .4997 .4997 .4997 .4997 .4997 .4997 .4997 .4997 .4999
3.5 .4998 .4998 .4998 .4998 .4998 .4998 .4998 .4998 .4998 .4998
Table Provide:
3.6 .4998 .4998 .4999 .4999 .4999 .4999 .4999 .4999 .4999 .4999
• When: P (0 ≤ Z ≤ a ) = 0.4842 • Find: a

3.7 .4999 .4999 .4999 .4999 .4999 .4999 .4999 .4999 .4999 .4999
3.8 .4999 .4999 .4999 .4999 .4999 .4999 .4999 .4999 .4999 .4999
P (0 ≤ Z ≤ a )
3.9 .5000 .5000 .5000 .5000 .5000 .5000 .5000 .5000 .5000 .5000
4.0 .5000 .5000 .5000 .5000 .5000 .5000 .5000 .5000 .5000 .5000
0.4 P( z1 ≤ Z ≤ 0) P(0 ≤ Z ≤ z1 )
P( z1 ≤ Z ≤ z 2 )
f(z)
0.2
z2 z1 z1 z1
z1 z z
0 z -2 -1 0 1 2 3 -2 -1 0 1 2 3
-3 -2 -1 0 1 2 3
P( z1 ≤ Z ≤ 0) = P(0 ≤ Z ≤ z1 )
P (0 ≤ Z ≤ z2 ) P( z1 ≤ Z ≤ 0)
• Standard Normal Distribution Table:
– It is practical to perform the integration numerically and tabulate
z2 z2 the results as a function of z.
z1 z1
z z
-2 -1 0 1 2 3 -2 -1 0 1 2 3
P( z1 ≤ Z ≤ z2 ) = P(0 ≤ Z ≤ z2 ) + P( z1 ≤ Z ≤ 0)
Normal Distribution Standard Standard

X −μ Normal Distribution
Z= Normal Distribution Normal Distribution
σ
X ~ N (μ, σ 2 ) Z ~ N (0,1) X ~ N (μ, σ 2 ) Z ~ N (0,1)
1 ⎡ − ( x − μ )2 ⎤ 1 ⎡ − z2 ⎤ 1 ⎡ − ( x − μ )2 ⎤ 1 ⎡ − z2 ⎤
f (x) = exp ⎢ ⎥ f ( z) = exp ⎢ ⎥ f (x) = exp ⎢ ⎥ f ( z) = exp ⎢ ⎥
2 πσ ⎣ 2σ 2π 2 πσ ⎣ 2σ 2π
2 2
⎦ ⎣ 2 ⎦ ⎦ ⎣ 2 ⎦
X −μ
f(x)
P( x1 ≤ X ≤ x2 )
0.4
P ( z1 ≤ Z ≤ z 2 )
Z=
σ
f(z)
0.2
x1 − μ
z2 z1 =
pdf 0
z1
σ
-3 -2 -1 0 1 2 3 P ( x1 ≤ X ≤ x2 ) P ( z1 ≤ Z ≤ z2 )
x1 x2
x
Random variable, z x2 − μ
x1 − μ z2 =
z1 = σ
σ
P ( x1 ≤ X ≤ x2 ) P ( z1 ≤ Z ≤ z2 )
x2 − μ • Convert Normal Distribution into Standard Normal
Convert Normal Distribution z2 = Standard Normal Distribution
σ Distribution
Example
• The random variable x follows normal distribution
μ = 10.0; σ = 1
z1 =
x1 − μ
z2 =
x2 − μ Mean to be found: μ
• Find probability: P(9 ≤ X ≤ 12); σ σ
9 − 10 12 − 10 Mean found from x1 + x2 + xn
z1 =
1
= −1; z2 == 2; experiment: x=
1 f(z)
n
P (9 ≤ X ≤ 12) = P ( −1 ≤ Z ≤ 2)
=
μ = x ±δ
z2
P(−1 ≤ Z ≤ 2)
z1 z
-3 -2 -1 0 1 2 3 Mean estimation:
= P(−1 ≤ Z ≤ 0) + P(0 ≤ Z ≤ 2)
f(z) f(z)
= P(0 ≤ Z ≤ 1) + P(0 ≤ Z ≤ 2) +
δ
z2
z z1 z
= 0.3413 + 0.4772 -3 -2 -1 0 1 2 3 -3 -2 -1 0 1 2 3 Mean interval???

f(z)
= 0.8185 ; 81.85%
Table |z1| z
-3 -2 -1 0 1 2 3
• Find out the mean wind speed in wind tunnel over 30

– A production line produces millions of minutes.
similar ball bearings per day. – Measure continuously and find the mean value of a
continuous signal.
– How to find the mean (average) diameter? – Measure every minute and calculate mean of 30
• Measure all of them and calculate mean. measurements.
• Measure 100 and calculate mean. – Measure every 5 minutes and calculate mean of 6
measurements.
• Measure two dozen and calculate mean. wind speed
– Are we confident with the result if not all of • Are we confident

with the result if
them are measured? And in what degree? only measured
periodically?
Time
Population Sample
• The population • A sample is a
• A medical laboratory analyses specimens of a
comprises the entire representative subset of a
pharmaceutical product to determine the concentration of
collection of objects, population on which an
the active ingredient.
measurements, experiment is performed
– The chemical analysis procedure is not precise so that repeated
observations, and so on. and numerical data are
measurements on the same specimen will give different results.
– All of ball bearings obtained.
– The standard deviation for the procedure is σ=0.02 grams per litre.
produced. – 100 (or two dozen) ball
– The laboratory analyzes each specimen 36 times and reports the
– Wind speeds at every bearings selected.
mean result.
instant in 30 minutes. – wind speed sampled once
– The laboratory has been asked to evaluate the claim that the
each minutes over the 30
concentration of the active ingredient is 0.86 grams per litre.
minutes period.
36 test results, Accept/Reject claim of active
ingredient 0.86 with 99% confidence • Different samples of the same population may be
mean value is
0.8404. level. chosen for experimentation.
Population (N data) Sample (n data) Estimation of Population Mean and Std

• The principle intention of a statistical
x1, x2, ..., xN x1, x2, ..., xn experiment is to find the population mean and
• Population mean: • Sample mean (average): standard deviation.
– Sometimes, it is not practical or even possible to find
x + x 2 +" x N 1 N
x + x 2 + " xn 1 n
μ= 1
N
=
N
∑x i x= 1 = ∑ xi population mean because the population size is too
i =1 n n i =1 big or infinite.
• Population standard • Sample standard – It is more practical to find sample mean and sample
deviation: deviation: standard deviation.
N
– Take a sample of finite size from the population,
1
σ= ∑(x i − μ)
2
S=
1 n
∑ ( xi − x )2 carry out test (measurement) and then calculate the
N i =1 n − 1 i =1 sample mean and standard deviation.
Keys on σn σ n −1
Calculators
Estimation of Population Mean and Std Estimation of Population Mean and Std
• The sample mean can be used as an estimation of the • Population: 1000 ball bearings μ = 0.5021cm σ = 0.2892cm
population mean.
Sample Mean Standard 90%
x ( x − δ ,x + δ )
of 100 x deviation S interval δ
Sample 1 0.5286 0.2818 0.0464 (0.4822, 0.5750)
• The sample standard deviation can be used as an Sample 2 0.4652 0.2790 0.0459 (0.4193, 0.5111)
estimation of the population standard deviation. Sample 3 0.5312 0.2715 0.0447 (0.4865, 0.5759)
Sample 4 0.5150 0.2633 0.0433 (0.4717, 0.5583)
Sample 5 0.4933 0.3153 0.0519 (0.4414, 0.5452)
S
• Different samples from the same population will yield
different values.
• However, x − δ ≤ μ ≤ x + δ
• Simple estimation of mean is not sufficient.
Mean Interval Estimation Confidence Interval and Confidence Level

• It is necessary to determine uncertainty interval for the • The confidence interval depends on a concept called the
estimated population mean. confidence level, sometimes called the degree of
μ = x ± δ ; or x −δ ≤ μ ≤ x +δ | μ − x |≤ δ confidence.
x +δ • The confidence level is the probability that the population
x −δ δ δ • Population mean: μ
mean will fall within the specified interval:
μ x • Sample mean: x confidence level = Probability ( x − δ ≤ μ ≤ x + δ )
= P (| x − μ |≤ δ )
• The uncertainty interval for the population mean: δ • Higher confidence level: large interval.
• The confidence interval on the mean: • Lower confidence level: small interval.
( x −δ , x +δ ) mean=11±5 with 95% confidence level
mean=11±3 with a 80% confidence level
Confidence Level and Level of Significance Mean and Standard Deviation of Sample Mean
• A population of the random variable x:
• The confidence level is normally expressed in terms of a μ
– Population mean
variable α called the level of significance. – Population standard deviation σ
confidence level = P (| x − μ |≤ δ ) = 1 − α • Take a sample of size n:
– Sample mean x
• α is then the probability that the mean will fall outside S
– Sample standard deviation
the confidence interval.
• Different sample yields different sample mean (There are
P (| x − μ |> δ ) = α
random variations in the sample mean).
• Examples: • The sample mean is also a random variable! x
α = 2%; P (| μ − x |≤ δ ) = 0.98 = 98%
• Mean and standard deviation of the sample mean.
α = 5%; P (| μ − x |≤ δ ) = 0.95 = 95% σ
μx = μ σx =
α = 10%; P (| μ − x |≤ δ ) = 0.9 = 90% n
Sample Mean Probability Distribution Sample Mean Distribution Standard Normal Distribution
⎛ σ ⎞ z ~ N (0,1)
• The sample mean approaches to a normal distribution if x ~ N⎜μ, ⎟
⎝ n⎠ x −μ
the sample size n is sufficiently large. z=
⎛ σ ⎞ σx σ
x ~ N⎜μ, ⎟ σx =
⎝ n⎠ n
• If the original population is normal, the distribution for P (x − δ ≤ μ ≤ x + δ ) = 1 − α P ( − zc ≤ z ≤ z c ) = 1 − α

the sample mean x is normal.
• If the original population is not normal and n is large P (μ − δ ≤ x ≤ μ + δ ) = 1 − α
(n>30), the distribution for the sample mean x is
normal. (μ ± δ)− μ ±δ σ
= = ± zc δ = zc
σ σ n
• If the original population is not normal and if n<30, the
n n σ
sample mean x follows a normal distribution only μ = x ± δ = x ± zc
approximately. n
Use Standard Normal zc
The Central Limit Theorem.
Distribution Table to find:
Mean Interval Estimation Application of Standard Normal Distribution Table

σ S
• Estimate δ δ = zc ≈ zc σ ≈S • Confidence Level: 1 − α = 95% α = 0.05
n n α
P( 0 < z < zc ) = 0.5 − = 0.475 zc = 1.96
P (x − δ ≤ μ ≤ x + δ ) = 1 − α 2
Standard Normal
Distribution Table
0.04 0.05 0.06
P ( − zc ≤ z ≤ zc ) = 1 − α z ~ N ( 0, 1) z
P( 0 < z < zc )
1.6 0.4495 0.4505
α
P( z > zc ) =
2 zc z
1.9 0.4750
f(z)
α • Confidence Level: 1 − α = 90% α = 0.1
P( 0 < z < zc ) = 0.5 − α
Area = α/2
confidence 2 P( 0 < z < zc ) = 0.5 − = 0.45
interval Area = α/2 2
1-α Use Standard Normal Table: P(0 < z < 1.64 ) = 0.4495 P (0 < z < 1.65) = 0.4505
Distribution Table to find:
-zc z=0 zc zc zc ≈ 1.645
Example
• We would like to determine the confidence interval of
the mean of a batch of resistors made in a certain
Confidence Confidence Confidence Confidence process.
level coefficient level coefficient
99% 2.58 68.26% 1 • Based on 36 samples, the average resistance is 25Ω. and
98% 2.33 95.44% 2 the sample standard deviation is 0.5Ω.
97% 2.17 99.74% 3
96% 2.05 99.9% 3.3 • Determine the 90% confidence interval of the mean
95% 1.96 99.96% 3.5 resistance of the batch.
90% 1.645
80% 1.28 70% 1.036 μ = x ±δ σ
• Solution: δ = zc
50% 0.6745 60% 0.842 n
x = 25Ω S = 0.5Ω
S
μ = x ± zc • The desired confidence level is 90%:
n
1-α=0.90 and α=0.1
Example: Solution Pharmaceutical Product Problem Solution

• n=36>30, sample mean follows normal distribution & • n=36>30, sample mean follows normal distribution &
standard normal distribution table is used to determine zc normal distribution function is used to determine zc
α α 0.01
P( 0 < z < zc ) = 0.5 − = 0.45 P( z < zc ) = 1 −
= 1− = 0.995
2 2 2
• From the Table 2, we get: zc ≈ 1.645 • From the table, we get: zc = 2.58 (99% confidence)
μ = x ±δ σ S 0.5
δ = zc ≈ zc = 1.645 × = 0.14 μ = x ±δ σ 0.02
n n 36 δ = zc = 2.58 = 0.0086
x = 25Ω n 36
x = 0.8404
μ = 25 ± 0.14Ω ( 90%) μ = 0.8404 ± 0.0086 = (0.8318, 0.849)
S = 0.5Ω or P (24.86 Ω ≤ μ ≤ 25.14Ω ) = 0.9 σ = 0.02
• The claim of 0.86 is outside of this range, therefore the

• This can be stated that the average resistance is expected
claim cannot be proved to be true.
to be 25±0.14Ω with a confidence level of 90%.
Mean Interval Estimation (n>30): Summary

S
• Estimate δ δ = zc μ = x ±δ
n • n ≤ 30: The sample mean x is not normally distributed.
• A new random variable is defined:
• where P (| μ − x |≤ δ ) = 1 − α x −μ
t= Sample mean x
P (| z |≤ zc ) = 1 − α
S
n Sample standard deviation S
ν=∞
α
P( 0 < z < zc ) = 0.5 − • The random 0.4 N (0,1) ν=5
2 ν=1
variable t follows
f(v,t)
Standard Normal
Use Standard Normal Distribution Table
Student's t 0.2
Distribution Table to P( 0 < z < zc ) distribution: f(t, ν).
find confidence
ν = n −1
coefficient: 0
zc • Degree of freedom -3 -2 -1 0 t 1 2 3
zc z
t Mean Interval Estimation using Student’s t distribution

Given the confidence level: 1-α .
• Given the confidence level: 1-α .
x −μ Sample mean x
• Find a value tc so that the probability of z lies t=
S Sample standard deviation S
somewhere in the confidence interval ±tc is 1-α. n
P (| t |≤ tc ) = 1 − α α P ( − tc ≤ t ≤ t c ) = 1 − α
P( t > tc ) =
2
• tc can be found
f(t,v) ⎛ x −μ ⎞
from Student’s t P ⎜ − tc ≤ ≤ tc ⎟
⎝ S/ n ⎠
Distribution Table.
Area = α/2 confidence Area = α/2 ⎛ S S ⎞
interval = P ⎜ x − tc ≤ μ ≤ x + tc ⎟ = 1−α
α ⎝ n n⎠
P( t > tc ) = 1-α
2
S S
-tc t=0 tc δ = tc μ = x ± δ = x ± tc
n n
Student’s t distribution Table

Example
α/2
• Student’s t as a ν
1
0.100
3.078
0.050
6.314
0.025
12.706
0.010
31.823
0.005
63.658
• A manufacturer of VCR systems would like to estimate
function of α and ν. 2
3
1.886
1.638
2.920
2.353
4.303
3.182
6.964
4.541
9.925
5.841 the mean failure time of a VCR brand with 95%
4 1.533 2.132 2.776 3.747 4.604
5
6
1.476
1.440
2.015
1.943
2.571
2.447
3.365
3.143
4.032
3.707
confidence.
7 1.415 1.895 2.365 2.998 3.499
8
9
1.397
1.383
1.860
1.833
2.306
2.262
2.896
2.821
3.355
3.250
• Six systems are tested to failure and the following data
l0
11
1.372
1.363
1.812
1.796
2.228
2.201
2.764
2.718
3.169
3.106 (in hours of playing time) are obtained: 1250, 1320,
12 1.356 1.782 2.179 2.681 3.054
tc
13
14
1.350
1.345
1.771
1.761
2.160
2.145
2.650
2.624
3.012
2.977
1542, 1464, 1275, and 1383.
15 1.341 1.753 2.131 2.602 2.947
α 16 1.337 1.746 2.120 2.583 2.921 • Estimate the population mean and the 95% confidence
P( t > tc ) = 17 1.333 1.740 2.110 2.567 2.898
2 l8
19
1.330
1.328
1.734
1.729
2.101
2.093
2.552
2.539
2.878
2.861 interval on the mean.
20 1.325 1.725 2.086 2.528 2.845
1-α=90% ; n=15
21 1.323 1.721 2.080 2.518 2.831
22 1.321 1.717 2.074 2.508 2.819
23 1.319 1.714 2.069 2.500 2.807
α=0.1; v=n-1=14 24
25
26
1.318
1.316
1.315
1.711
1.708
1.706
2.064
2.060
2.056
2.492
2.485
2.479
2.797
2.787
2.779
27 1.314 1.703 2.052 2.473 2.771
α
tc = 1.761
28 1.313 1.701 2.048 2.467 2.763
= 0.05 29 1.311 1.699 2.045 2.462 2.756
2 30 1.310 1.697 2.042 2.457 2.750
∞ 1.283 1.645 1.960 2.326 2.576
Example: Solution Example: Solution

• Data: 1250 1320 1542 1464 1275 1383 • n=6<30, the t-distribution to estimate the
• Sample Mean: confidence interval.
1250 + 1320 + 1542 + 1464 + 1275 + 1383 μ = x ±δ S
x= = 1372h δ = tc
6 n
• Sample Standard Deviation: • 95% confidence corresponds to α of 0.05.
S=
• Using the t-distribution table:
α
(1250-1372) 2 +(1320-1372) 2 +(1542-1372) 2 +(1464-1372) 2 +(1275-1372) 2 +(1383-1372)2 ν = n −1= 5 ; = 0.025 ; tc = 2.571
6-1 2
64482 S 114
=
5
= 114 h δ = tc = 2.571 × = 120 μ = x ± δ = 1372 ± 120h (95%)
n 6
P (1252h ≤ μ ≤ 1492h ) = 0.95
Mean Interval Estimation (n≤30): Summary Section 5: Experiment Errors and

• Estimate δ δ = tc
S
μ = x ±δ Uncertainty Analysis
n
• Measurement Error:
• where P (| μ − x |≤ δ ) = 1 − α
error true value measured value
P (| t |≤ tc ) = 1 − α
α • Error in a measurement means the inevitable uncertainty
P( t > tc ) = ν = n −1
2 that presents in all measurements.
– Error in technical usage does not imply that there is any
mistake in the measurement, although mistakes can cause
Use Student’s t
errors.
Distribution Table to
– One can not eliminate errors by being very careful in
find: tc
measurement.
– The best one can hope to do is to ensure that errors are as
small as reasonably possible and to have a reliable estimate of
tc how large they are.
Inevitability of Uncertainty Measurement Uncertainty Examples

Example: Measure the temperature outside in an 10 20 30 40 50 60
Autumn day. Measure length

1st attempt: The temperature is about 11ºC
with a ruler
2nd attempt: The temperature could be between 10ºC to 12 ºC Probable range:
Best estimate of length 46.5 mm to 47.5mm
3rd attempt: Measurement using a = 47 mm
thermal meter, 10.5 ºC Different observers may not
agree with the precise estimate.
More accurate: 10 ºC ≤ T ≤ 11 ºC
Best estimate of time
Or: T = 10.5 ± 0.5 ºC
= 9:05:35
10.5 ºC one can ask why not 10.5001 ºC Probable range:
Other random factors (e.g. wind, small change 9:05:34.5 to 9:05:35.5
in location etc) all contribute to the uncertainty. Uncertainty introduced by instrument limitation!!
How to report and use uncertainties? Measurement Error and Uncertainty

• The correct way to state the result of measurement is to • The magnitude of the error in any measurement
give a best estimate of quantity and the range within can only be estimated.
which you are confident the quantity lies.
• One refers to an estimate of the error in the
Best estimate of length = 47 mm
measurement as the uncertainty presented in the
Probable range: 46.5 mm to 47.5mm measured value.
• The measurement of the length is usually state as:
Measured value of length = 47±0.5 mm Uncertainty: the bound of the error
• In general, the result of any measurement of a
quantity x is stated as: x − wx wx wx x + wx
Measured value of x = xbest ± δ

error ≤ wx
x x
True value Estimated value
Uncertainty
• As the uncertainty is an estimate, the confidence level is
Uncertainty and Percentage Uncertainty
normally associated with the uncertainty. • Uncertainty: V = 10 (V ) wV = 1 (V )
• General, a variable to be measured can be described as:
Voltage V 10 1 V 95
x x wx P
• In some cases, the uncertainty is quoted as
x Measured (or estimated) value. percentage or fractional uncertainty:
wx Uncertainty. P Confidence level.
wx
fx
x
Resistance is: 25 ± 0.14 Ω (90%)
Voltage example: wV 1
VCR average life: 1372 ± 120 h (95%) fV = = = 10%
V 10
Random Uncertainty Systematic Uncertainty

• Systematic (Bias) uncertainty is fixed error which
• Random uncertainty can occur for a variety of reasons, all of
causes the experimental results to be consistently too
which lead to the measurements fluctuating about a mean value.
high or too low by roughly the same amount.
• Random errors may be reduced by improving the measurement
apparatus (like getting a more precise voltmeter) or the technique • The systematic error can be corrected if it is identified.
(reading the scale with a magnifying glass), but they cannot be However, it is often difficult to uncover and explain the
eliminated. cause.
• The size of the random uncertainty may be obtained only by • Common cause of systematic errors are:
making a set of repeated, independent observations. – Incorrect calibration of measurement instruments
– Change of measurement conditions
– Human errors
True value True value
Measurable values being
Independent corroboration
inconsistent when
of experiment results is Bias error
repeated measures of a Mean x required before a critical Average of
constant.
experiment can be trusted. measured values
Problem to be solved, 1 Problem to be solved, 2

• A certain obstruction-type flowmeter (orifice, venturi, nozzle) is
• Manometers used to measure the flow of air at low velocities. The relation
– determine the pressure by measuring the height of a fluid column. describing the flow rate is:
P = ρgh 2 g c p1 p
v CA
ρ: the manometer fluid density h: the fluid column height
RT1
g: the acceleration of gravity with uncertainty 0.1mm C: empirical-discharge coefficient A: flow area
• Requirement: p1: upstream pressure R: gas constant for air
– to achieve an accuracy of 0.1% of at the maximum reading, Δp: the difference between upstream and downstream pressures
10000Pa.
• Experiment results: C = 0.92 ± 0.005 p1 = 25 ± 0.5
• Problem to solve: Estimate the uncertainty that can be tolerated T1 = 530 ± 15 Δp = 1.4 ± 0.005 A = 1.0 ± 0.001
in the density of a gage fluid that has a nominal value of 2500 kg/m3. gc = 32.17 R = 53.35
• Problem to solve: Find v ± wv.
Propagation of Uncertainties Propagation of Uncertainties

• Example: Determine the electrical power using P = VI = (100 ± 2)(10 ± 0.2)
voltage and current: ⎧(100 + 2)(10 + 0.2) = 1040.4 = 1000 + 40.4
V 100 2 V ⎪ (100 + 2)(10 − 0.2) = 999.6 = 1000 − 0.4
⎪
=⎨
I 10 0 2 A ⎪ (100 − 2)(10 + 0.2) = 999.6 = 1000 − 0.4
⎪⎩ (100 − 2)(10 − 0.2) = 960.4 = 1000 − 39.6
• What will be the value of the power P and its
uncertainty? • Maximum error: ( ΔP )max = 40.4 ( W )
• Here the uncertainties in the measurements V
and I propagate into the result P. P = 1000 ± 40.4 (W )
• It is a worst case estimation, not a good measure
P = P ± wp P 100 10 1 kW of uncertainty (error bound) in P.
Propagation of Uncertainties Error Propagation:

Single Variable Function Case
• Measurement equation: m f x y z
y Tangent at P
Variable Estimate Uncertainty Q
The measurements y + Δy
x x wx and uncertainties for Δy
x, y and z are Δxf ' ( x )
y y wy assumed to be P
y
z z wz independent. Δx dy
f ' ( x) =
dx
• Estimate of m: m f x y z y=f(x)
x + Δx
x
x
dy
• We need to find the uncertainty of m: wm Δy ≈ Δx
dx
Error Propagation: Error Propagation: General Cases

Two Variables Function Case
∂m ∂m ∂m
m f x y z Δm ≈ Δx + Δy + Δz
∂x ∂y ∂z
Δx: change in x Δy: change in y
Δz: change in z Δm: change in m
∂m
partial derivative of m with respect to x
∂x
∂m
partial derivative of m with respect to y
∂z ∂z ∂y
z = f ( x, y ) Δz ≈ Δx + Δ y ∂m
∂x ∂y partial derivative of m with respect to z
∂z
Error Propagation: Simple Example Error and Uncertainties

m f x y z
y = x1 + x2 x1 = 200 ± 2 x2 = 10 ± 0.5
∂m ∂m ∂m
y = 210 ± Δy Δm = Δx + Δy + Δz
∂x ∂y ∂z
∂y ∂y ∂y ∂y Δy = ± w y Δz = ± wz
Δy = Δx1 + Δx 2 =1 =1 Δx = ± wx
∂x1 ∂x2 ∂x1 ∂x2
∂m ∂m ∂m
Δy = ( ±2) + ( ±0.5) Δm = ± w x ± wy ± wz
∂x ∂y ∂z
• Question: • Question:
– What sign (+ or -) to use in the above equation? – What sign (+ or -) to use in the above equation?
Uncertainty Error Propagation: Simple Example

• The most probable error:
– Square root of the sum of the squares (RSS) . y = x1 + x2 x1 = 200 ± 2 x2 = 10 ± 0.5
2 y = 210 ± Δy Δy = ( ±2) + ( ±0.5)
⎛ ∂m ⎞ ⎛ ∂m ⎞ ⎛ ∂m ⎞
2 2
wm = ⎜ wx ⎟ + ⎜⎜ w y ⎟⎟ + ⎜ wz ⎟
⎝ ∂x ⎠ ⎝ ∂y ⎠ ⎝ ∂z ⎠ • Uncertainty: wy = 2 + 0.5
2 2
= 2.062
– Uncertainty.
• The fractional (percentage) uncertainty. • Result: y = 210 ± 2 062
wm
fm = ; m = f ( x , y, z )
m
Example Example: Solution

• To calculate the power consumption in a resistive electric P VI ∂P = I = 10 ∂P = V = 100
circuit: ∂V ∂I
V = 100 ± 2 (V )
P VI
I = 10 ± 0.2 ( A) ⎛ ∂P ⎞ ⎛ ∂P ⎞
2 2
wP = ⎜ wv ⎟ + ⎜ wI ⎟
• The maximum possible error in P? ⎝ ∂V ⎠ ⎝ ∂I ⎠
• The best estimation of uncertainty in P?
= 10 × 2 + 100 × 0.2 = 28.3W
2 2
• Solution: P V I 100 10 1 kW wP 28.3

fP = = = 2.83%
P 1000
• To calculate (ΔP)max and wp , we need to find the
partial derivatives of P with respect to V and I: • Result: P = (1000 ± 28.3) W
or P = (1000 ± 2.83%) W
Example Example: Solution

• Find the the uncertainty
when the area of the shape c
shown in right is
A ac 2bc - 2c 2
determined by the b
∂A
measurements of a, b and
= c = 5 ∂A = 2c = 10 ∂A = a + 2b − 4 c = 60
c.
A ac cb c ∂a ∂b ∂c
a
ac bc c2
A 30 5 2 25 5 ⎛ ∂A ⎞ ⎛ ∂A ⎞ ⎛ ∂A ⎞
2 2 2
a = 30 ± 0.1 ( mm ) wA = ⎜ wa ⎟ + ⎜ wb ⎟ + ⎜ wc ⎟
b = 25 ± 0.1 ( mm ) - 2 52 ⎝ ∂a ⎠ ⎝ ∂b ⎠ ⎝ ∂c ⎠
c = 5 ± 0.1 ( mm ) 350 ( mm 2 ) = 0.5 + 1 + + 6
2 2 2
= 6.1mm 2
Propagation of Uncertainties Example

• Manometers
V = 100 ± 2 (V ) – determine the pressure by measuring the height of a fluid column.
P = ρgh
ρ: the manometer fluid density h: the fluid column height
I = 10 ± 0.2 ( A) g: the acceleration of gravity with uncertainty 0.1mm
• Requirement:
P = P ± w p = 1000 ± 28 3 W – to achieve an accuracy of 0.1% of the maximum reading, 10000Pa.
h=0.408m δh=0.0001m g=9.81m/s2

P = 100 × 10 = 1000(W )
• Problem to solve: Estimate the uncertainty that can be tolerated
in the density of a gage fluid that has a nominal value of 2500 kg/m3.
wP = 10 × 2 + 100 × 0.2 = 28.3W
2 2
ρ = 2500 kg / m 3 wρ ??
Example: Solution Example Solution P = ρgh

P = ρgh • Uncertainty equation:
2
g = 9.81 ρ = 2500 kg / m 3 ⎛ ∂P ⎞ ⎛ ∂P ⎞
2 2
⎛ ∂P ⎞ ⎛ ∂ P ⎞
2
P = 10 kPa full scale wP = ⎜⎜ wρ ⎟⎟ + ⎜ wh ⎟ w = ⎜⎜
2
wρ ⎟⎟ + ⎜ wh ⎟
⎝ ∂ρ ⎠ ⎝ ∂h ⎠ ⎝ ∂ρ ⎠ ⎝ ∂h ⎠
P
P 10000
h= = = 0.408 m
ρ g 2500 × 9.81 ∂P ∂ d
= {ghρ} = gh {ρ} = gh = 9.81 × 0.408
The maximum uncertainty in ∂ρ ∂ρ dρ
wP = 10 kPa × 0.1% = 10.0Pa
pressure is 0.1% of 10 kPa = 4.00 m 2 / s 2
wh = 0.1 mm = 0.0001 m
∂P ∂ d
= {gρh} = gρ {h} = gρ = 9.81 × 2500.0
• Problem to solve: find wρ wρ ?? ∂h ∂h dh
= 24525.0 kg / m 2 s 2
Example: Solution Question

2
⎛ ∂P ⎞ ⎛ ∂ P ⎞
2
• Can we obtain uncertainty and/or maximum
w = ⎜⎜
2
wρ ⎟⎟ + ⎜ wh ⎟ error without the involvement of partial
∂ρ ∂
P
⎝ ⎠ ⎝ h ⎠
derivatives?
100 = 4.00 × wρ + (24525 × 0.0001)
2 2
wρ = 2.4 kg / m 3
• Problem Type 1: R = R1 + R2
m = ax + by + cz
• The maximum allowable uncertainty in the gage fluid
density is f mnD
• Problem Type 2: K=
fρ =
wρ
=
2.4
≈ 0.1%
d2 G
± 2.4 kg / m 3 or
ρ 2500
m = kx a y b z c
Problem Type 1: m = ax + by + cz Problem Type 1: Uncertainty

• a, b, c are known constants without error. m = ax + by + cz m = ax + by + cz
Variable Estimate Uncertainty Percentage Uncertainty ∂m ∂m ∂m
=a =b =c
x x wx fx ∂x ∂y ∂z
y y wy fy 2
⎛ ∂m ⎞ ⎛ ∂m ⎞ ⎛ ∂m ⎞
2 2
z z wz fz wm = ⎜ wx ⎟ + ⎜⎜ w y ⎟⎟ + ⎜ wz ⎟
⎝ ∂x ⎠ ⎝ ∂y ⎠ ⎝ ∂z ⎠
• Estimate of m: m = ax + by + c z
wm = ( awx )2 + ( bw y )2 + ( cwz )2
• Uncertainty and Percentage Uncertainty? wm fm
wm
fm =
m
Problem Type 1: Example Example: Prove Repeated

Measurements reduced uncertainty
R = R1 + R2 m = ax + by + cz R1 = 100 ± 5 Ω The accuracy of all n measurements are the same:
Measurement 1: m1 = x1 ± δ
wR (1 wR1 ) 2 (1 wR2 ) 2 R2 900 63 Measurement 2: m2 = x2 ± δ
……………
52 632 63.2 Measurement n: mn = xn ± δ
m1 + m2 + " + mn 1 1 1
Average: m = = m1 + m2 + " + mn
R = R1 + R2 = 1000 Ω R = 1000 ± 63.2 Ω n n n n
x1 + x2 + " xn
Best estimate: mbest =
wR ΔR max n
fR = = 6.32% f ΔR max
= = 6.8%
R R ⎛1 ⎞ ⎛1 ⎞
2 2
⎛1 ⎞ 1
2
Uncertainty: δ m = ⎜ δ ⎟ + ⎜ δ ⎟ + " + ⎜ δ ⎟ = δ <δ

⎝n ⎠ ⎝n ⎠ ⎝n ⎠ n
Problem Type 2: m = kx a y b z c Problem Type 2: Partial Derivatives

• a, b, c and k are known constants without error. m = kx a y b z c
Variable Estimate Uncertainty Percentage Uncertainty ∂m ∂ d a
= kx a y b z c = ky b z c x
x x wx fx ∂x ∂x dx
y y wy fy m
= ky b z c ax a −1 = kx a y b z c ax −1 = a
z z wz fz x
• Estimate of m: m = kx a y b z c
∂m m
= kx a z c by b−1 = kx a y b z c by −1 = b
∂y y
• Uncertainty and Percentage Uncertainty? wm fm
∂m m
= kx a y b bz c −1 = kx a y b z c cz −1 = c
∂z z
Problem Type 2: Uncertainty Manometer Example Revisited

2
⎛ ∂m ⎞ ⎛ ∂m ⎞ ⎛ ∂m ⎞ P = ρgh = gρh
2 2
wm = ⎜ wx ⎟ + ⎜⎜ w y ⎟⎟ + ⎜ wz ⎟ m = kx a y b z c
⎝ ∂x ⎠ ⎝ ∂y ⎠ ⎝ ∂z ⎠ P = 1000; wP = 10
fm = + bf y + cf z
2 2 2
2 h = 0.408; wh = 0.0001 af x
m⎞ ⎛ m⎞ ⎛
2 2
⎛ m⎞
= ⎜ awx ⎟ + ⎜⎜ bw y ⎟⎟ + ⎜ cwz ⎟ g = 9.81; ρ = 2500
⎝ x⎠ ⎝ y⎠ ⎝ z⎠
wρ = ? fP = 1× fρ
2
+ 1 × fh
2
2
⎛ w ⎞ ⎛ w ⎞ ⎛ w ⎞
2 2
= m ⎜ a x ⎟ + ⎜⎜ b y ⎟⎟ + ⎜ c z ⎟ 2 f P2 = f ρ2 + f h2
⎝ x ⎠ ⎝ y ⎠ ⎝ z ⎠ ⎛ wP ⎞ ⎛ wρ ⎞ ⎛ wh ⎞
2 2
⎜ ⎟ = ⎜⎜ ⎟⎟ + ⎜ ⎟
⎝ P ⎠ ⎝ ρ ⎠ ⎝ h ⎠
fm = af x
2
+ bf y + cf z
2 2
wm = mf m wρ = 2.4 kg / m 3
Example Example
• A certain obstruction-type flowmeter (orifice, venturi, nozzle) is C = 0.92 ± 0.005 p1 = 25 ± 0.5 Δp = 1.4 ± 0.005
used to measure the flow of air at low velocities. The relation
describing the flow rate is:
T1 = 530 ± 15 A = 1.0 ± 0.001 gc = 32.17 R = 53.35
2 g c p1 p
v CA 2 g c p1 p
RT1 v CA
RT1
C: empirical-discharge coefficient A: flow area
p1: upstream pressure R: gas constant for air
2 32.17 25 1.4
0.92 1 0.2596
Δp: the difference between upstream and downstream pressures 53.35 530
1 1 1
• Experiment results: C = 0.92 ± 0.005 2 g c p1 p 2 gc −
p1 = 25 ± 0.5 v CA CA( p1 ) ( p ) (T1 ) 2
2 2
T1 = 530 ± 15 Δp = 1.4 ± 0.005 A = 1.0 ± 0.001

RT1 R m = kx a y b z c
gc = 32.17 R = 53.35 2 2 2
2 2 1 1 1
fv fC fA fp f Δp fT
• Problem to solve: Find v and wv. 2 1 2 2 1
Example Example
2 2 1 2 1 2 1 2 2 21 2 1 2 1 2
fv fC fA fp f Δp fT fv fC fA fp f Δp fT
4 1 4 4 1 4 1 4 4 1
C = 0.92 ± 0.005 p1 = 25 ± 0.5 Δp = 1.4 ± 0.005 0.01827 1.827%
T1 = 530 ± 15 A = 1.0 ± 0.001 gc = 32.17 R = 53.35 v 0.2596

wC 0.005 wp1 0.5
fC 0.00543 f p1 0.02 wv vf v 0.2596 0.01827 0.00474
C 0.92 p1 25
wΔp 0.005 wT1 15 • Finally: v 0.2596 0.00474
f Δp 0.00357 f 0.0283
p 1.4 T1
T1 530
wA 0.001
fA 0.001
A 1
Practice Questions
• Question 1: Suppose you measure three numbers as
x=200±2, y=50 ± 2, z=20 ± 1. The three variables are
independent and random. What you would give for the
values of q=x+y-z and r=xy/z with their uncertainties?
• Question 2: Suppose you measure three number as

x=200±2, y=50 ± 2, z=40 ±21. The three variables are
independent and random. Use step by step propagation to
find the quantity q=x/(y-z) with its uncertainty. [First
find the uncertainty is the difference y-z and then the
quotient x/(y-z).]

CE1602 - 2015 Statistics

Hochgeladen von

Dokumentinformationen

Copyright

Verfügbare Formate

Dieses Dokument teilen

Dokument teilen oder einbetten

Freigabeoptionen

Stufen Sie dieses Dokument als nützlich ein?

Sind diese Inhalte unangemessen?

Copyright:

Verfügbare Formate

CE1602 - 2015 Statistics

Hochgeladen von

Copyright:

Verfügbare Formate

Statistical Analysis of Experimental Data

Statistical Analysis of Experimental

Past Exam Questions Past Exam Questions

Section 1: Least-Squares Curve Fitting Problem 1: Simple Linear Problem

Wind Speed Probability

• According to thermodynamic principles, a relationship 7.8 0.57

density function f(u) using the least

square curve fitting method.

Linear Models Curve Fitting and the Method of Least Square

Linear Function: Draw a line of best fit Least Squares: Why?

Least Squares: Why? y = ax + b Least Squares: Why? y = ax + b

Optimisation and Derivative Optimisation and Partial Derivatives

Least Squares: Why? y = ax + b Least Squares for Curve Fitting: Concept

Least Squares: Optimisation Optimisation and Partial Derivatives

Least Squares Solution Least Squares Solution: Quality of Fit

Least Squares Solution (b=0) Least Squares y = ax + b

Simple Least Square Example Example: Solution

Example Solution Example: Least Square Solution for Nonlinear Model

b) Write the equation connecting P and V.

log P + γ log V = log C 40

log P = −γ log V + log C 30

Define: x = log V y = log P

and C can be calculated.

Example Solution Example Solution

Example Solution Nonlinear functional relationship and

ln(y) = dx + ln(c) Define: y1 = ln(y); a=d; b = ln(c)

Functions/Curves Can be Transformed into Linear Wind Speed

y = a0 + a1 (log x) + a2 (log x)2 13.9

Section 2: Correlation of Experiment Data Model Examples

• nth order curve: y = a0 + a1 x + a2 x 2 + " + an x n

Linear Model: The Simplest and the Linear Correlation Coefficient

Values of Correlation Coefficient

Verify Linear Relationship Critical Value of Correlation Coefficient

– We can presume that y does depend on x in a non- function of the 15

random manner and can expect that a linear number of samples 17

functional relationship. significance, α. 30

Table: Minimum Values of the Correlation

variables (with 95% 64

confidence)? 62 6 × 23925.4 − 358 × 400.5

Example: Solution Example

Mass (m, kg) 2.0 7.6 13.2 18.8 24.4 30

Example Example x (m) y (f) xy x2 y2

6 96 23.5 272.12 2084.8 119.87

• Since rxy. is less than rt, a linear relationship • A linear

Example x ( m1 ) y (f) xy x2 y2 Example

Correlation Coefficient: Comments Correlation Coefficient: Comments

What type of problems do we want to solve?

– A production line produces millions of

What type of problems do we want to solve? Inevitability of Randomness (Uncertainty)

Inevitability of Randomness (Uncertainty) 3.1: Dealing with randomness (uncertainty):

Dealing with randomness (uncertainty): Measure of central tendency: Mean (Average)

component length, room temperature etc), it is n n

The Weighted Mean (Average) Example

Data Spread & Standard Deviation Standard Deviation: Definition

Frequency Frequency distribution

Frequency Distribution (Temperature

1100 ≤ T <1105 21 1085 1087.5 1090

1100 ≤ T <1105 1102.5 21 1105 ≤ T <1110 14 4

1105 ≤ T <1110 1107.5 14 1110 ≤ T <1115 7

• A histogram is a tabulation of the frequency-

• Example results of 21 students: 38, 39, 42, 44, 48,

Class Rel. Fre. • The normalised relative frequency is the relative

1095 ≤ T <1100 20% 20%

Frequency Polygon Mean and Standard Deviation of Grouped