Depar tment of Statistics & O.R.

Aligarh Muslim University Aligarh

BA/BSc I Semester

Descriptive Statistics (STB 151)

by

Dr. Haseeb Athar

 Unit -2
Moments
&
Measure of Skewness and Kurtosis

2 Lecture notes by Dr. Haseeb Athar, Department of Statistics & O.R., A.M.U., Aligarh, India
 Moments
The concept of moments has crept into the statistical literature from mechanics. In mechanics,
this concept refers to the turning or the rotating effect of a force whereas it is used to describe the
characteristic of a frequency distribution in statistics. Moments represent a convenient and unifying
method for summarizing many of the most commonly used statistical measures such as measures of
tendency, variation, skewness and kurtosis.
Types of Moments
There are two types of moments we calculate
 Moments about arbitrary point or raw moments
 Moments about mean or central moments

Moments about Arbitrary Point:

For Discrete Data

Let X1, X 2 ,..., X n are n observations, the r th moment about any arbitrary point A is given by
1 n
r   ( X i  A)r ; r  0,1, 2,...
n i 1
In particular
1 n
0   ( X i  A)0  1 :
n i 1
Zero order moment

1 n
1   ( X i  A)
n i 1
: First order moment

1 n
2  
n i 1
( X i  A)2 : Second order moment

1 n
3   ( X i  A)3
n i 1
: Third order moment

1 n
4   ( X i  A)4
n i 1
: Fourth order moment

For Ungrouped Frequency Distribution

Let X1, X 2 ,..., X n are n observations with their respective frequencies f1, f 2 ,..., f n , then the r th
moment about any arbitrary point A is given by

1 n
r  
N i 1
fi ( X i  A)r ; r  0,1,2,...,

n
where N   fi
i 1

Remark 1: In the case of grouped frequency distribution X i , i  1,2,..., n are the mid points of the
class intervals.
Remark 2: If we take an arbitrary point A  0 , then we get the moments about origin as

3 Lecture notes by Dr. Haseeb Athar, Department of Statistics & O.R., A.M.U., Aligarh, India
 1 n r
r   X i ; r  0,1, 2,...
n i 1
: For discrete data

1 n
r   fi X i r ; r  0,1, 2,...,
N i 1
: For ungrouped and grouped frequency distribution

n
where N   fi
i 1

2. Moments about Mean or Central Moments

When we take deviation from the actual mean (i.e A  X ) and calculate moments, then these are
called moments about mean or central moments.
For Discrete Data

Let X1, X 2 ,..., X n are n observations, then the r th moment about mean ( X ) or r th central moment
is given by

1 n
r   ( X i  X )r ; r  0,1, 2,...
n i 1

In particular
1 n
0  
n i 1
( X i  X )0  1 : Zero order central moment

1 n
1   ( Xi  X )  0 :
n i 1
First order central moment

1 n
2   ( X i  X )2
n i 1
: Second order central moment also called variance

1 n
3  
n i 1
( X i  X )3 : Third order central moment

1 n
4   ( X i  X )4
n i 1
: Fourth order central moment

For Ungrouped Frequency Distribution

Let X1, X 2 ,..., X n are n observations with their respective frequencies f1, f 2 ,..., f n , then the r th
moment about mean X or r th central moment is given by

1 n
r   fi ( X i  X )r ; r  0,1,2,...,
N i 1
n
where N   fi
i 1

Remark: In the case of grouped frequency distribution X i , i  1,2,..., n are the mid points of the
class intervals.

4 Lecture notes by Dr. Haseeb Athar, Department of Statistics & O.R., A.M.U., Aligarh, India
Relation between Central Moments and Raw Moments and conversely

We know that the r th moment about mean or central moment is

1 n
r  
N i 1
fi ( X i  X )r ; r  0,1,2,...,

1 n
r  
N i 1
fi ( X i  A  A  X ) r (1)

Let di  X i  A and also we have

1 n 1 n
1   i i
N i 1
f ( X  A)   di  d
n i 1

1 n
Also 1   fi ( X i  A)  X  A
N i 1

Therefore from (1), we have

1 n
r   fi (di  1 )r
N i 1
(2)

Expand (di  1 )r in (2) binomially, we get

1 n   r r  r 

r   fi dir    dir 11    dir 2 ( 1 )2    dir 3 ( 1 )3  ...  (1)r 1r 
N i 1   1   2  3 


1 n r 1 n r  n r n
  fi dir    1  fi dir 1   12  fi dir 2    13  fi dir 3  ...  (1)r 1r
N i 1 1  N i 1  2  i 1  3  i 1
In particular on putting r  1,2,3 and 4 in the above expression, we get

1  0
2  2  12
3  3  32 1  213
4  4  43 1  62 12  314
Converse

The r th moment about any point A is given by

1 n
r   fi ( X i  A)r ; r  0,1,2,...,
N i 1

1 n
r   fi ( X i  X  X  A)r
N i 1
(3)

Let zi  X i  X and X  A  d , then from (3), we have

5 Lecture notes by Dr. Haseeb Athar, Department of Statistics & O.R., A.M.U., Aligarh, India
 1 n
r  
N i 1
fi ( zi  d )r

1 n   r r  r 

  fi  zir    zir 1d    zir 2 d 2    zir 3d 3  ...  d r 
N i 1   1   2  3 


1 n r 1 n r  1 n r 1 n
  fi zir    d  fi zir 1    d 2  fi zir 2    d 3  fi zir 3  ...  d r
N i 1 1  N i 1  2  N i 1  3  N i 1

r r  r
 r    1r 1    12 r 2    13r 3  ...  1r (4)
1   2  3
In particular at r  2,3,4 with 0  1 and 1  0 , we get

2  2  12
3  3  32 1  13
4  4  431  62 1  14
Example 1: The first four moments of a distribution about the value 5 of a variable are 1, 10, 20 and
25. Find the central moments.
Solution: We have given that
1  1, 2  10, 3  20 and 4  25 .
Therefore,
2  2  12  10  1  9
3  3  32 1  213  20  3 10 1  2  (1)3  8
4  4  43 1  62 12  314
 25  4  20  1  6  10  12  3  14  2

Effect of Change in Origin and Scale on Moments

Xi  A
Let ui 
h
then X i  A  hui , which implies X  A  hu

Therefore

X i  X  h(ui  u )
Thus, the r th moment of X about arbitrary point X  A is given by
1 n
r ( x)  
N i 1
fi ( X i  A)r

1 n
 
N i 1
fi (hui )r

1 n
 hr  fiuir  hr r (u ) .
N i 1

6 Lecture notes by Dr. Haseeb Athar, Department of Statistics & O.R., A.M.U., Aligarh, India
Again, the r th moment of X about mean or central moment is given as
1 n
r ( x)   fi ( X i  X ) r
N i 1
1 n
 
N i 1
fi {h(ui  u )}r

1 n
r ( x)  h r 
N i 1
fi (ui  u )r  h r r (u ) .

Thus, the r th moment of the variable X is h r times the r th moment of the new variable u after
changing the origin and scale.
Example 2: Wages of workers are given in the following table:
Weekly 10-12 12-14 14-16 16-18 18-20 20-22 22-24
wages
No. of 1 3 7 20 12 4 3
workers

Find the first four raw moments by suitably changing origin and scale and then convert into central
moments.
Solution: Calculation of moments

Wages No. of Mid Point X i  17

Workers ( Xi )
ui  fui fui2 fui3 fui4
2
( fi )

10 - 12 1 11 3 3 9 27 81

12 - 14 3 13 2 6 12 24 48

14 - 16 7 15 1 7 7 7 7

16 - 18 20 17 0 0 0 0 0

18 - 20 12 19 1 12 12 12 12

20 - 22 4 21 2 8 16 32 64

22 - 24 3 23 3 9 27 81 243

13 27 67 455

The formula for r th raw moment is given by

1 n
r  r ( x)  hr 
N i 1
fi uir

Therefore,
1 n 13
1  h  
N i 1
fi ui  2   0.52
50

7 Lecture notes by Dr. Haseeb Athar, Department of Statistics & O.R., A.M.U., Aligarh, India
 1 n 27
2  h2  
N i 1
fi ui2  4 
50
 2.16

1 n 67
3  h3  
N i 1
fi ui3  8 
50
 10.72

1 n 455
4  h4  
N i 1
fi ui4  16 
50
 145.6

Now central moments are

1  0
2  2  12  2.16  0.2704  1.8896

3  3  32 1  213

10.72  3  2.16  0.52  (0.52)3  7.491

4  4  43 1  62 12  314

145.6  4  0.52  10.72  6  2.56  0.2704  3  0.07312  126.5874

Sheppard’s Correction for Grouping Errors

When moments are calculated for continuous frequency distributions or grouped data (i.e. data that
has been binned), it is assumed that the data is centred on the class interval mid-points. This erroneous
assumption introduces “grouping errors” in calculation of moments. The grouping errors occur
only in even moments (i.e. second and fourth moments).
The correction should only be made to data with the following characteristics:
 Frequencies should taper to zero in both the positive and negative direction. In other words,
frequencies should be symmetrical and gradually taper off (like the behaviour you would see in
a normal distribution).
 Variables should be continuous. The method is not suited to discrete variables.
 Class intervals should be equal in width.
 Class intervals should be more than 1/20th of the total range.
W. F. Sheppard suggested some corrections to be made to get rid of the so called “grouping errors”
that enter into the calculation of moments.

h2
2 (Corrected)  2 
12
3 (No correction needed)
1 7 4
4 (Corrected)  4  h2 2  h ,
2 240
where h is the width of the class interval.

8 Lecture notes by Dr. Haseeb Athar, Department of Statistics & O.R., A.M.U., Aligarh, India
Example 3: Compute the first four moments for the following distribution of marks after applying
Sheppard's correction.
Marks out of 20 5 6 7 8 9 10 11 12 13 14 15
No. of Students 1 2 5 10 20 51 22 11 5 3 1

Solution: Do it by yourself
Factorial Moments: The r th factorial moment of a variable X about the origin is given as
1 n n
(r )   i  fi ,
(r )
f x , N 
N i 1 i i 1
where
x(r )  x( x  1)( x  2)...( x  r  1) .

The r th factorial moment of a variable X about any point x  a is given by

1 n n
(r )   fi ( xi  a)(r ) , N   fi
N i 1 i 1
where
( x  a)(r )  ( x  a)( x  a  1)( x  a  2)...( x  a  r  1)
In particular
1 n
   fi xi  1
(1)
N i 1
1 n 1 n
 
(2)  i  fi xi ( xi  1)
(2)
f x 
N i 1 i N i 1
1 n 1 n
 
N i 1
fi xi2   fi xi
N i 1
   2  1
(2)
1 n 1 n
 
(3) 
N i 1
fi xi(3)   fi xi ( xi  1)( xi  2)
N i 1
 (3)
  3  32  21

1 n 1 n
 
(4) 
N i 1
fi xi(4)   fi xi ( xi  1)( xi  2)( xi  3)
N i 1
1 n
  fi xi ( xi3  6 xi2  11xi  6)
N i 1
 (4)
  4  63  112  61 .

Example 4: In Example 2, convert first four raw moments into factorial moments.
Solution: Do it by yourself.
Absolute Moment
The absolute moment of order r about the origin is given as
1 n n
r  
N i 1
fi | X i |r , N   fi
i 1

9 Lecture notes by Dr. Haseeb Athar, Department of Statistics & O.R., A.M.U., Aligarh, India
 Measure of Skewness
The term skewness refers to lack of symmetry or departure from symmetry. When a distribution is
not symmetrical (or is asymmetrical), it is called skewed distribution. The measures of skewness
indicate the difference between the manners in which the observations are distributed in particular
distribution compared with a symmetrical (or normal) distribution. The concept of skewness gains
importance from the fact that statistical theory is often based upon the assumption of the normal
distribution.
Nature of Skewness: Skewness can be positive or negative or zero.
 When the values of mean, median and mode are equal, there is no skewness.
 When mean > median > mode, skewness will be positive.
 When mean < median < mode, skewness will be negative.

There is different measure of skewness, which are discussed below.

1. Karl Pearson’s Coefficient of Skewness
This method is most frequently used for measuring skewness. The formula for measuring
coefficient based on mean, mode and standard deviation is given by
Mean  Mode
sk P  ,
SD
where sk P denotes the Pearsonian coefficient of skewness.

This formula can be used for fairly symmetric data. However, if the data is moderately skewed or
mode is ill defined, then the above formula can be modified by using the empirical relation as,
3( Mean  Median)
skP 
SD
 If skP  0 , the symmetrical distribution.
 If mean is greater than mode, implies sk p  0 , then positive skewness.
 If mean is less than mode, implies sk p  0 , then negative skewness.

The Karl Pearson's coefficient of skewness lies between 3 to 3 .

2. Bowley’s Coefficient of Skewness
This method is based on quartiles. The formula for calculating coefficient of skewness is

10 Lecture notes by Dr. Haseeb Athar, Department of Statistics & O.R., A.M.U., Aligarh, India
 (Q3  Q2 )  (Q2  Q1 ) Q1  2Q2  Q3
sk B  
Q3  Q1 Q3  Q1

 If skB  0 , then it is symmetrical distribution.

 If skB  0 , and then positively skewed distribution.
 If skB  0 , and then negatively skewed distribution.

The Bowley’s coefficient of skewness is used for moderately skewed distribution and distribution
having open end class. The Bowley’s coefficient of skewness lies between 1 to 1 .

Example 5: The IQ scores of 50 students of a class are given blow

IQ Score: 50 - 60 60 - 70 70 - 80 80 - 90 90 - 100

No. of Std: 5 8 16 12 9
Calculate
(i) Karl Pearson coefficient of skewness based on median and mode.
(ii) Bowley's coefficient of skewness.

Solution: To calculate all above measures, first we shall calculate mean, median, mode, standard
deviation, first quartile and third quartile.

X A
IQ Score No. of Std. ( f ) cf X d fd fd 2
h

50 - 60 5 5 55 -2 -10 20
60 - 70 8 13 65 -1 -8 8
70 - 80 16 29 75 0 0 0
80 - 90 12 41 85 1 12 12
90 - 100 9 50 95 2 18 36
f  50  fd  12  fd 2  76

X  A  hd  75  10  0.24  77.4
1
 d2 
N
 fd 2  d 2  1.52  0.0576  1.46

 x2  h2 d2  100 1.46  146

 x  12.08

h N  10
Q1  lq1    C pq1   60  (12.5  5)  69.38
f q1  4  8

h N  10
Q2  lq2    C pq2   70  (25  13)  83.75
f q2 2  16

11 Lecture notes by Dr. Haseeb Athar, Department of Statistics & O.R., A.M.U., Aligarh, India
 h N  10
Q3  lq3    C pq3   80  (37.5  29)  87.06
f q3  4  12

(i) The Karl Pearson coefficient of skewness based on mean, mode and standard deviation is given by

X  Mode 77.4  76.67

skP    0.06
x 12.08

The Karl Pearson coefficient of skewness based on mean, median and standard deviation is given
by

3( X  Md ) 3(77.4  83.75)
sk P    1.58
x 12.08

(ii) The Bowley's coefficient of skewness is

Q  2Q2  Q3 69.38  2  83.75  87.06
sk B  1   0.6256
Q3  Q1 87.06  69.38

3. Measure of Skewness based on Moments

Karl Pearson defined the following four coefficients, based upon the first four central moments.
The coefficients 1 and  1 are used to measure the skewness. These coefficients are defined as
below:

32 
1  , 1   1  3
23 3

For a symmetrical distribution 1  0 , but it does not tell the direction of skewness, that is
positive or negative. Because 32  0 and variance 2 is always positive. This drawback is
removed by calculating  1 coefficient. Thus, the sign of skewness would depend upon the value
of 3 whether it is positive or negative.

 If 1  0 , then positively skewed distribution

 If 1  0 , then negatively skewed distribution

 If 1  0 , then symmetrical distribution

12 Lecture notes by Dr. Haseeb Athar, Department of Statistics & O.R., A.M.U., Aligarh, India
 Measure of Kurtosis
The relative flatness of the top is called kurtosis or convexity of curve. The coefficients  2 and  2
are used to measure the kurtosis. These coefficients are defined as below:
4
2  ,  2  2  3
22

The quantity  2  2  3 is called excess of kurtosis.

 If  2  3 , then curve is called

Platykurtic.
 If  2  3 , then curve is called
Mesokustic.
 If  2  3 , then curve is called
Leptokurtic.

Example 6: Refer to example 2 and calculate 1, 1, 2 and  2 coefficients. Also study the nature
of the distribution.
Solution: In example 2, we have calculated the following first four central moments as
1  0, 2  1.8896, 3  7.491 and 4  126,5874
Therefore

32 (7.491)2 56.11508

1     8.317
23 (1.8896)3 6.7469

3 7.491
and 1   1    2.91
 3
(1.37)3

Since 1  0 , therefore distribution is positively skewed.

Further,
4 126.5874
2    35.4527
22 (1.8896)2

and  2  2  3  32.4527

Since  2  0 , therefore distribution is leptokurtic.

13 Lecture notes by Dr. Haseeb Athar, Department of Statistics & O.R., A.M.U., Aligarh, India
Practice Exercises
1. The first three moments of a distribution about the value 2 of the variable are 1, 16 and -40. Show
that the mean is 3, the variance is 15 and 3  86 . Also show that the first three moments about
x  0 are 3, 24 and 76.
2. The first four moments of a distribution about the value 5 of the variable are 2, 20, 40 and 50.
Show that the mean is 7, variance 16, 3  64, 4  162, 1  1 and 2  0.63 .

3. For a certain distribution, the mean is 10, variance is 16,  1 is 1 and  2 is 4. Find the first four
moments about the origin.
4. Calculate the first four moments about the mean for the following data. Also calculate 1 and  2 .

x: 1 2 3 4 5 6 7 8 9
f: 1 6 13 25 30 22 9 5 2
5. In a certain distribution, the first four moments about the point 4 are -1.5, 17, -30 and 108.
Calculate 1 and  2 and state whether the distribution is leptokurtic or platykurtic.

6. Show that for discrete distribution

(i) 2  1

(ii) 2  1 .

14 Lecture notes by Dr. Haseeb Athar, Department of Statistics & O.R., A.M.U., Aligarh, India