UNIT 2

MEASURES OF DISPERSION, CORRELATION AND REGRESSION

MEASURES OF DISPERSION:
1. Range
2. Quartile deviation
3. Mean deviation.
4. Standard deviation.

Range:
Definition: The range is the difference between the largest value of a data set and the
smallest value of a set.
Formula: Range = Maximum – Minimum
Coefficient of Range = (Maximum – Minimum) / (Maximum + Minimum)
1) Find range and its coefficient for following data,
23,12,34,56,54,67,75,34
Solution: Maximum= 75 , Minimum= 12
Range = 75 – 12 = 63
Coeff of Range= (75-12)/ (75+12) = 63/87= 0.72
Quartile Deviation:
Formula: QD= (Q3-Q1)/ 2
Coeff of QD(inter quartile range)= (Q3-Q1) / (Q3+Q1)
1) Calculate the quartile deviation and it’s coefficient for the following distribution.

Solution:
Coeff of QD= (82.5-34.167) / (82.5+34.167)
= 48.333/ 116.667
= 0.414
 Mean Deviation
There are three types:

• Mean Deviation from Mean

• Mean Deviation from Median
• Mean Deviation from Mode

Mean Deviation from Mean (for ungrouped data):

1)

Mean Deviation from Mean (for grouped data):

1) Find mean deviation from mean for following data,

Mean deviation from Median (for ungrouped data):

1) Find mean deviation from Median for following data

6, 15, 4, 10, 12, 11, 5, 3, 16

Solution:
Mean deviation from Median (for grouped data):

1) Find mean deviation from median for following data

Solution:
Standard deviation (for ungrouped data):

Variance = (standard deviation) ^2

1) Find standard deviation
Standard deviation (for ungrouped data):

1) Find standard deviation

CORRELATION:
Definition: if there exists a relation between pair of variables x and y, such that change in one
variable is accompanied by change in the other variable then we say that variables x and y are
correlated and the relation between them is called correlation.

Types of correlation

1) Perfect positive correlation: if all data points lie on a straight line towards positive axis
of the graph, then it is perfect positive correlation and in such a case r=+1 2) Positive
correlation: If all data points cluster together and make a narrow band towards the
positive axis of graph then it is positive correlation and in such a case 0<r<1. 3) Perfect
negative correlation: if all data points lie on a straight line towards negative axis of the
graph, then it is perfect positive correlation and in such a case r=-1
4) Negative correlation: If all data points cluster together and make a narrow band towards
the negative axis of graph then it is positive correlation and in such a case, -1<r<0. 5) Zero
(No correlation): if all data points spread all over the graph paper then it is zero
correlation and in such a case r=0.

Methods of determining correlation

 1) Scatter diagram
2) Karl pearsons coefficient of correlation.
3) Spearmans Rank coefficient of correlation.

Karl Pearsons coefficient of correlation

Formula: r = nƩxy – ƩxƩy / sqrt((nƩx^2 – (Ʃx)^2)) * sqrt((nƩy^2 – (Ʃy)^2) ).

Illustration:

1) Find coefficient of correlation for following data and interpret your answer.
x 14 8 10 11 9 13 5

y 14 9 11 13 11 12 4

Solution:
x Y Xy x^2 y^2

14 14 196 196 196

8 9 72 64 81

10 11 110 100 121

11 13 143 121 169

9 11 99 81 121

13 12 156 169 144

5 4 20 25 16

70 74 796 756 848

Ʃx=70, Ʃy=74, Ʃxy=796, Ʃx^2=756, Ʃy^2=848, n=7

r = nƩxy – ƩxƩy / sqrt((nƩx^2 – (Ʃx)^2)) * sqrt((nƩy^2 – (Ʃy)^2) ).

= 7 x 796 – 70x74 / sqrt(7x756-(70)^2) * sqrt(7x848-(74)^2)

= 5572 – 5180 / sqrt(5292 – 4900) * sqrt(5936 – 5476)

= 392 / sqrt(392) * sqrt(460)

= 392 / 19.8 x 21.45

= 392 / 424.71

= 0.92

There is positive correlation between two variables x and y.

Spearman’s Rank coefficient of correlation (for non-repeated

ranks) Formula: R = 1 – (6Ʃd^2 / n(n^2-1)) where d=R1-R2

1) Find Rank coefficient of correlation for following data and interpret your answer.
x 15 32 25 30 35 20 19 22 27 31

y 50 70 65 72 90 58 53 57 68 74

Solution:
X Y R1 R2 D d^2

15 50 10 10 0 0

32 70 2 4 -2 4

25 65 6 6 0 0

30 72 4 3 1 1

35 90 1 1 0 0

20 58 8 7 1 1

19 53 9 9 0 0

22 57 7 8 -1 1

27 68 5 5 0 0

31 74 3 2 1 1

8
R = 1 – (6Ʃd^2 / n(n^2-1))

= 1 – (6x8/ 10(100-1))

= 1 – (48/990)

= 1 – 0.0485

= 0.9515

There is a positive correlation between the two variables.

2) The following data gives ranks of two subjects, find rank coefficient of correlation
between them.

R1 3 5 7 1 2 8 6 4

R2 2 1 4 5 7 6 3 8

Solution:
R1 R2 D d^2

3 2 1 1

5 1 4 16

7 4 3 9

1 5 -4 16

2 7 -5 25

8 6 2 4

6 3 3 9

4 8 -4 16

96

R = 1 – (6Ʃd^2 / n(n^2-1))
= 1- (6x96/8(64-1))

= 1 – (576/504)

= 1 – 1.14

= -0.14
Spearman’s Rank coefficient of correlation (for repeated ranks)

Formula: R = 1 – (6(Ʃd^2+(Ʃm(m^2-1)/12)) / n(n^2-1)) where d=R1-R2

1) Find Rank coefficient of correlation for following data

x 12 15 13 20 15 14 19 13 21 18

y 25 21 15 18 20 17 20 16 20 22

Solution:
x Y R1 R2 D d^2

12 25 10 1 9 81

15 21 5.5 3 2.5 6.25

13 15 8.5 10 -1.5 2.25

20 18 2 7 -5 25

15 20 5.5 5 0.5 0.25

14 17 7 8 -1 1

19 20 3 5 -2 4

13 16 8.5 9 -0.5 0.25

21 20 1 5 -4 16

18 22 4 2 2 4

140
 Value M m(m^2-1)/12

13 2 0.5

15 2 0.5

20 3 2

R = 1 – (6(Ʃd^2+(Ʃm(m^2-1)/12)) / n(n^2-1))

= 1- 6(140+3)/ 10(100-1))

= 1 – 6(143)/990)

= 1- (858/990)

= 1- 0.87

= 0.13

Multiple Correlation

Formula:

r = NƩfxy – ƩfxƩfy / sqrt((NƩfx^2 – (Ʃfx)^2)) * sqrt((NƩfy^2 – (Ʃfy)^2) ).

1) Find coefficient of correlation for the following data representing marks in test 1(x) and
test 2(y) for a group of twenty-five children.

x/y 7 8 9

3 2 1 --

4 1 3 1

5 4 4 2

6 2 3 2

Solution:
x/y 7 8 9 f fx fx^2 fxy

3 2(42) 1(24) ---- 3 9 27 666

4 1(28) 3(96) 1(36) 5 20 80 160

5 4(140) 4(160) 2(90) 10 50 250 390

6 2(84) 3(144) 2(108) 7 42 252 336

 f 9 11 5 25 121 609 952

fy 63 88 45 196

fy^2 441 704 405 1550

fxy 294 424 234 952

r = NƩfxy – ƩfxƩfy / sqrt((NƩfx^2 – (Ʃfx)^2)) * sqrt((NƩfy^2 – (Ʃfy)^2) ).

= 25x952 – 121x196 / sqrt(25x609 – (121)^2 * sqrt(25 x 1550 – (196)^2) =

23800 – 23716 /sqrt(15225-14641) * sqrt(38750 – 38416)

= 84 / sqrt(584) * sqrt(334)

= 84/ 24.17 * 18.28

= 84 / 441.83

= 0.19

Regression

1) Regression equation of y on x (to find y)

Y= a+bx

b= n*∑xy - ∑x ∑y / n∑x^2 – (∑x) ^2

a= mean y – b * mean x

2) Reg eqn of x on y (to find x)

x= a1+b1y

b1= n*∑xy - ∑x ∑y / n∑y^2 – (∑y) ^2

a1= mean x – b1 * mean y

1) Find x when y=18
x Y xy y^2

3 12 36 144

4 11 44 121

6 15 90 225

10 16 160 256

12 19 228 361
 13 17 221 289

48 90 779 1396

Mean x= 48/6=8 , Mean y= 90/6= 15

b1= n*∑xy - ∑x ∑y / n∑y^2 – (∑y) ^2

= 6x779 – 48x90 / 6x1396-(90)^2

= 4674-4320/ 8376-8100

= 354/276

= 1.28

a1= mean x – b1 * mean y

= 8 – 1.28x15

= 8 – 19.2

= - 11.2

X= a1+b1y

= - 11.2 +1.28y

When y=18

= - 11.2+1.28x18

= -11.2+23.04

= 11.84
Regression equation in terms of mean and std dev

1) Regression equation of y on x (to find y) Y=

a+bx

b= r (std dev of y/ std dev of x)

a= mean y – b * mean x

2) Reg eqn of x on y (to find x)

x= a1+b1y

b1= r (std dev of x/ std dev of y)

a1= mean x – b1 * mean y

1) Suppose income=x , Expenditure=y

r=0.8 , mean x= 3250, mean y= 2830

std dev x= 103, std dev y=93

find y, when x=3280

find x, when y=2785

b= 0.8 x (93/103)

= 0.8 x 0.9

= 0.72

a= 2830 – 0.72x3250

= 2830 – 2340

= 490

When x=3280

Y= a+bx

= 490+0.72 x 3280

= 490+2361.6
= 2851.6

To find coefficient of correlation using regression

coefficients. r = sqrt ( b x b1)

1) 2x-3y+14=0 ---------------1

3x-y-42=0 ------------------ 2

Find r, mean of x and y

Sol: multiplying eq 2 by 3 and then subtracting from eq

1, 2x-3y+14=0

9x-3y-126=0

2x-3y+14 – (9x-3y-126)=0
2x-3y+14-9x+3y+126=0

-7x+140=0

-7x=-140, 7x=140

x=140/7= 20, mean of x=20

Substitute x=20 in eq 1

2x-3y+14=0

2x20-3y+14=0

40-3y+14=0

54=3y

54/3=y

Y=18, mean of y=18

Consider eqn 1 as reg eqn of y on x

2x-3y+14=0

Comparing with General eqn y=a+bx

-3y= -14-2x

3y=14+2x

Y= 14/3 +2/3 x

b=2/3

consider eqn 2 as reg eqn of y on

x 3x-y-42=0

-y=-3x+42

Y=3x-42

b=3

b= 2/3 < b= 3

Hence, b=2/3

b1=1/3
r= sqrt(bxb1) = sqrt(2/3 x 1/3) = sqrt(2/9)

= sqrt(0.22) = 0.47