2 Linear Regression

In Module 1, we observed that the correlation coefficient is unable to express the cause and
effect of the relationship between two variables. Thus, we cannot predict the value of one
variable when the other variable is known. This limitation is addressed through regression
analysis. In regression analysis, the relationship between variables is expressed in the form of a
mathematical equation, wherein it may be assumed that one variable is the cause and the other is
the effect.

Regression is a statistical tool that facilitates the understanding of the relationship between
Yariables and predicts the unknown values of dependent variables from the known values
of independent variables.

Let Y be the dependent variable and X be the independent variable. If we know the relationship
between these two variables, we can predict any value of Y for a given value ofX.

2.1 Difference Between Correlation and Regression

Correlation Re ression
It is a relative measure of the relationship It is an absolute measure of the relationship
between variables. between variables.
It measures the degree and direction of the It measure the nature and extent of the
relationshi . relationshi .
Correlation coefiicient is independent of Regression coefficient is independent of
chan e of ori ·n and chan e of scale. chan e of ori in but not of scale
Correlation coefficient is independent of Regression coefficient is not independent of
the units of measurement. units of measurement.
There ma be zero correlation. There is nothin like zero re!ITession.

2.2 Regression Lines

If the two variables are highly correlated and are plotted on a graph, the points will lie within a
narrow strip. If this strip is nearly straight, we may draw a line such that all the points are close
to it from both sides. Such a line can be considered representative if the sum of the distances of
the points from the line is minimised. This is referred to as the line of regression or the line of
best fit. We measure the deviations of points from a line both i) vertically and ii) hori1ontally,
resulting in one line when the distances are minimised vertically, and a second line when the
distances are minimised horizontally. Thus, we obtain two lines of regression.

Regression equations are algebraic expressions of the regression Imes. Consider two variables, X
and Y. If Y is the dependent variable, then we have the line of regression of Y on X. This line
provides the most probable value of Y for a given value of X.
 f Yon X ed aIong the y-axis,. we obtai"~1a
f
ston o the line rneasur y === a + bX. [Y 1s dependent
h points frorn . · given by
• the deviations oft e X its equat10n is
tbinttnise. f egression of y on .
Wled thehne o r
able]
& Y - y- :::: byx cx -f)

. alled regression coefficient.

where byx is c fX
. f y for a given value o .
This line is used for estimatmg a value o

Line of Regression of X on y d I the x-axis we obtain a

. asure a ong
· ·
Ifwe minimise the deviat10ns o
fthe points from the 1me me . .
.
b X-
f X y its equation 1s given y -
a+ bY. [X 1s.
'

line referred to as the line of regress10n o on .

dependent variable]
or
X - X = hxy(Y - Y)

where hxy is called regression coefficient.

This line is used for estimating a value of X for a given value of Y.

Note: When there is either a perfect positive or perfect negative correlation between the two
variables, the two regression lines will coincide; that is, we will have only one line of regression.
If the variables are independent, r is zero, and the lines of regression are at right angles, that is,
parallel to the X-axis and Y-axis.
or
X=a+bY

Here, 'a' and 'b' (or hyx or hxy ) are two parameters of regression lines They nkn
constants having fixed numerical values which determine the position of the ·line. are u own

The parameter 'a' determines the level of the fitted line (i.e. the distance of th . .
above or below the origin). The parameter 'b' determines the slope of the line (i e 1
ethtne directly
Y for a unit change in X). · · e change in

Least Squares Method

method is based on the assumption that the sum of the squares of the differences be
ated values and the actual values of the observations is minimum.. This methtwe~n
find the best-fitting straight line or regression line for the given data. Od 1s
 sion line Y =a+ bX, we need to determine the values of the constants 'a' enil
~ equations for obtaining the values of constants •a' and 'b' are given by
£Y==Na+bLX
i) 2
£XY == aLX + b LX
ii)
To find the regression line X = a + bY , The n01mal equations are given by
iii) L X = Na + b L Y 2
iv) L XY = a L Y + b L Y
This method is also known as the direct method, which becomes quite cumbersome when the
values of X and Y are large. To simplify this, we will take deviations from their respective
means instead of the actual values. Thus, regression lines can be expressed as follows:

2.3 Regression Coefficients

The constant 'b' in the regression line is referred to as the regression coefficient. As there are
two lines ofregression: the line of regression ofY on X and the line of regression of X on Y, we
have two regression coefficients, denoted as byx and bxy respectively.

The following are the different methods to evaluate regression coefficients:

i) Direct method: N L XY - L X L y
regression coefficient Yon X i.e. byx = NLX2-(Lxi

NLXY-LXLY
regression coefficient X on Y i.e. bxy = N LY2-(L Y)2

ii) When deviations are taken from the arithmetic mean:

LXY
regression coefficient Y on X i.e. byx === L xz
LXY
regression coefficient X on Y i.e. bxy === L Y2

where x === X - X , y === Y - Y

iii) When standard deviations and correlation coefficient are known:
<ly
r-
regression coefficient Y on X i.e. byx === ax
<Jx
r-
regression coefficient Y on X i.e. bxy === ay
 ____!. .

--- _ _ ., _ _ rel' .....

. . . . . . . . .IJIP•--

• ,j; ../f,s ~ Jj;

!i tJx - r2
,.__ b ==r .r--
r"' ± ~ than one, the other 01ust be less daaii
• "Tl'" 'KY tJx tJy

,..._ion coeflldent1 ,s grea er

'Ille -81deDt of correlation cannot exceed t.
~r ~ t,r S 1
2
< 1
bi,, s t [Jbyx - -;-
xY

ftetwO,ifbyx < t ,then

regression bxy > 1. will share the same sign, either p ositive or n egath-e:
coefficients
b II aot possible for one regression coefficient to be positive while the othu ii

die regression coefficients by, and b,y are both po si ;j ,. e, r :s pas' :i ,· e. su '.:. ·i :0.=
aepdve.
If

~ ;~=--·~
gress,oncoe ,cienls byx and b,y are both neaat1 ve. then r j, -,a,.:.. ,
at of- ·the
re · m·
TIJe coefficient of correlation is of th e :ame . sign
4. ,oefficients. as yegre-ssior

the r is. ifth

That • coefficients
. e regress'.°n
egresswn coefficients b and b
':.: ,.,. J .,
b,, and bx-, are :Jc,:" -yo .... ..
: . _'.}<!Sl'.ffe. s=:'. :c
5. Regression coe&fic'1en ts areyxinde xy are both nega!l•:e.· r -- ,, _, •
___--,• ;;.
-- s:::. . -
Let u =~
• 'v11-- ky-b pendent of changes in oriain
b ' - b ut not of scale.

We know' ruv = r:xy,. (J.2u - -CJ

l 2 2 1
- .,.. au k
(l k2 V •/2
h 2 x , Clv =-r. •.

bU v - 'uv·-=
qv r:xy· 2h . -
k qy
- . . C7y = h-k . bxy
=h''Xy·- qx

Similarly' we can prove that b - h

vu - -b
k yx
1

huv = l~. bxy

k

buv == coefficient of x
coefflei entofy· bxy
6. The arithmetic
. mean of th e coefficients
coe c1ent of correlatio n, 1.e. 2 "! ~ r.
. b '" +b of regressl on Is greater than or equal to die
~
m
byx+bxy
2 > r
i.e. ~ (r <:!x +
2
<ry T <:!x) >
<:1y - T
<ry
1.e r -(1 + r 2.
Xvy
> 2r
(1

- -

.
1.e. ax2 + ay2 >
- zrr
v x rr
v
. 2
i.e ax + ay2 - 2aX ay >
- 0
y

i.e ( ax - ay) z > 0 which is true

7. The angle between the rmes of regression varies in accordanc .· th

coefficient. • "' the correlation

The equation of the lines of regression

. of y on X is

y- Y = hyx (X - X)
Its slope , m1 = r ay ( as hyx
ax
= r <:!y)
ax

The equation of the lines of regression of X on y is

X - X = hxy(Y - Y) i.e. y - y = ~
bxy
(X _ X)

Its slope m 2 = rax

ay (as b xy = r ~)
ay

Let 0 be the angle between the lines of regression.

ay ay
- -r-
0 = + 1+m1m2
m2-m1 == + r ax ax
tan - 1+ray _!Y,_ ax- r ax

Here, the positive sign gives the acute angle as r

2
~ 1 and o-xand D"y are positive.

If r = 0, we conclude that tan 0 c=

00
or 0 "'~ i.e. two regression lines are at right angles.
2
 ,
ncide
..,.....;.,!-..t MedlodJ
..,_.~ . two ret'
ot'CSS Ofl I~ cm
rB:;::; 0 i.e .
..i. ..t tanB:;::; O0
,.-tude i.u-
lf r= ±1, we"",._
m-t:'St!J,,
. , ,. c·....-., l'(!
,JO, ,

the lines of re'=" .

. ..11uations are
k the given ""1 ?
Example 1: Chee - ,, and x :;::; 2-; - .I ,, • - ~ --~ ,
2X + Y - .: ., • " i/ ,..,., ½ ,~,., ~- ;,.,,
~ === tI ' ,,
., --Y'
I
, ~"; ,. .,);~

'ne reg;essW-- ...

-2.X k 3 be t "'
-3 -:::,y==
Solution: let zx +Y -
. 1·tneXonY.
regression
. f regression
Then, coefficients o
2 ..
-=~= -r~~-n
4 , _ _
... ,--::-:,-.:-...,,,-

--2,bxy== . Jr..,.~':~ ..... ~,.._._ ......

byx- . 'theot1-,,,,,..i5 .../J-,1,. .. , - -
. "ble as one of them is negative anc. . .,. J • •

which ts not posst

than 1 numerically.
e lines to be in the other way:
Ifwe assume th
Let 2x + y == 3 be regression line X on Y.

2x = -y + 3

X
1
= - -y
2
+
3
-
2
hxy =- -21
x = 2y + 3 be regression line Y on X.
2y =x-3

Y
=!.2 x - ~b
2 yx
=!.which is again not possible as the two coefficiem::i are of d1ffer-:nt 5igns.
2

Therefore, the given lines are not lines of regression.

Example 2: State true or false with reasoning : "The regression coefficienb betv. een 3x and 3\
are the same as those between x and y". •

Solution: Ifu =ax+ hand v =by+ k

a
huv = b" hxy
From given data, u = 3x a = 3 , v = 3y b = 3
[Link]► '

'

-=--

.. r - l r +_ =0
r cannot be - _ :1:; >l
1
Example 5 lfur =u =u n, •.n_;I.. h~t\\.:>, n th~ 11,s 0fr~gn-,s1-.1 .~ .s :.rn- 3 find the
coe ilic1ent of c-tirrelauon

Solution The angle bcr.h.:-n lmr, of re~~.:-:-:-.,'· 1

.:- ; , .:-n ')

tan8 = l-

=t3--....!,__
r " +o
r 2 +6r-1 == O

r = - 6±"36:!: -3 +2/i.
2 -

Since r cannot be greater than I, r = - 3 + 2,/2 =-0 .17

Example 6: Find the two lines of regression for the following data:

X: 10 12 16 11 15 14 20 22

Y: 15 18 23 14 20 17 25 28

Solution: 7

yz l XY
I y xz I
X 225 150
10 15 100
144 324 216
12 18 368
23 256 529
16 154
11 14 121 196
20 225 400 300
15
17 196 289 238
I 14 500
I 20 25 400 625 -
I 22 28 484 784 616 -
I rx 120 1:Y 160
2
Lx 1926 LX 2 3372 IXY 2542 ~

N=8

Regression equation ofY on X is. y =a+ bX (*)

Nonnal equations are given by

LY=Na+brx
LXY = a LX + b Lxz
i.e. 160 = 8 a + 120 b (i)
~542 = 120 a + 1926 b (ii)
 •

Linear Regreffioo 41

Solving equations (i) and (ii), and multiplying equation (iJ with 15, we get

2400 = 120 a + 1800 b

2542 = 120 a + 1926 b
Subtracting

-142 = -126 b

b = 142
126
= 1.127
Substituting the value of b = 1.127 in equation (i)

160 = 8a +120(1.127)

8a = 160-135.24 = 24.76
24 76
a= ·
8
= 3.095
Putting values of a and b in equation (*),

The required regression equation ofY on Xis Y = 3.095 .L 1.127 X

Regression equation of X on Y is X = a + bY
Normal equations are given by

IX=Na+bIY

L XY = a L Y + b L Y 2
i.e. 120 = 8 a + 160 b (i)

2542 = 160 a + 3372 b (ii)

Solving equations (i) and (ii), and multiplying equation (i) with 20, we get

2400 = 160 a + 3200 b

2542 = 160 a + 3372 b

Subtracting, we get

-142=-172b

b=~
172
= 08256
'
 11
lub,dtutfn1 the value of b ,,.~2~71 " c<JW' ',n fiJ

120 sa •1r,o (O.H2~7 J

HM 120 112 12
,,
a 1I I

Puttinu valucn 1,I 11 :mil f, in equ:,11,,ri (11 J, the req1Jlft;d rq~rc;t,~,ic>n equation of X <m y ,s

X 1., t IJXJ,','/ Y

f(,umplc 7: f Jcfcrr11ine fb1; 1111, rnj'/!;'/,J<1n lmv, frc,rr1 th,;; f<>IJ1 1 11rn;~ <fat1.1:

X:7 k I
) 'J l;, ') )() l ,;

Y:4 1 I, (, ,, ,.)
'/ l J,

SoJufltm:

X '( x% /1, )Cf

7 /4 4') I~ 2x
H I'

I
) (,1 2, 'j 4()
~ 2 2'
, }
4
'J r, HJ
%1 1~
12 ')
144 t 54
'J l %1 I (J8

~
5 %1
I (J 25 45
7 I ()(J
+

15 4')
12 7()
rx 7:, LY SO LX%
225
7(/J ~
144
y% :~HO
18()

· cquati, r f y
RcgrCHHton
L XY = c·~c
.J
;:J ,
1
' '' ,,n X ·
y hy.,JX XJ
9.-~7S
'
y
-
';(J

i~ G2S
byx == ~53S-7sxso
8X769-(7S)2

-- ~B0-3750
- S'iO
61S2-S6is Si?= 1.0057

equation of y on X:

y - 6.25 :=: 1.0057

(X - 9,375)
 Linear Regression I 43

y == 6.25 + 1.00 57 X - 9.42 85

y == - 3 .1 7 84 + 1.00 57 X
Regression equa tion of X on y:
X - X == hxy (Y - y)
- 75
b _
where xy -
NIX Y-IX IY
NIY 2-(I Y)2 'X= -a == 9.37 5 y- so
== 8 == 6.25
b = 8x53 5-75 x5o
xy 8X38 0-(S0 ) 2

4280 -375 0 530

3040 -250 0 = 540 == 0.98 15

Regression equa tion of X on y:

X - 9.37 5 = 0.98 15 (Y - 6.25 )

X == 9.37 5 + 0.98 15Y - 6.13 44

X = 3.24 06 + 0.98 15 Y
Example 8: Foll owi ng is the data of sales of mob
iles in five cities and the no. of families
residing in thes e cities.
Delh i Mum bai Lucknow Mysore Pune
City 60
No. of families : 70 75 80 90
25.2 28.6 30.2 22.3 35.4
Sales('000s)
square meth od.
Fit a linear regr essi on equa tion of Y on X by the least

Solution·
City No. of families(X) Sales (Y) xz XY
25.2 4900 1764
Delhi 70
28.6 5625 2145
Mum bai 75
30.2 6400 2416
[Link] 80
22.3 3600 1338
Mysore 60
35.4 8100 3186
Pune 90
rx -
375 LY - 141 .7 Lx 2 - 286 25 LXY - 108 49

Regression equa tion Y on X is given by

Normal equa tion s are given by

LY =N a+ bLX

LXY = arx + hLX 2

 (i)

· 141•7 = 5 a+ 375 b
I.e. (ii)
10849 = 375 a + 28625 b
Solving equations (i) and (ii),
.) b 75 we get
Multiply equation (1 Y '
10627.5 == 375 a + 28125 b
10849 == 375 a + 28625 b
Subtracting,

_ 221.5 == -500 b

b == 0.443

Subsituting value of b = 0.443 in equation (i)

14 1, 7 = Sa+ 375(0.443)
14 1,7 = Sa+ 375(0.443)
141.7 = Sa+ 166.125

Sa = 166.125 - 141.7
a - - 24.425 = -4.885
- 5
The line of regression is
Y = -4.885 + 0.443 X
Example 9: From the following information find regression equations and estimate the
production when the capacity utilization is 70%.
Average (mean) Standard deviation

Production (in lakhs): 42 12.5

Capacity 88 8.5
Correlation coefficient: 0.72

Solution:
Let production and capacity utilization be denoted by variables X and y respectively. Regression
equation of pro~uction base_d on. ~ase? on capacity. utilization shall be gi vcn by x on y and
regression equation of capacity ut1hzatton of product10n shalJ be given by y on X which can be
computed as given below. '

: X = 42, Y = 88, ax = 12.5, ay = 8.5

' Regression line of Y on X:

where byx =r ~
y y

<1:l:

= 0.72 x ~I~ .!;

= 0.4896
Regression line is
y BB 0.1H'J(,( Y. 11.
y BB I() .·tr,
AU()(, X 1,(J ,. ( ·)
, 1 ,

y (,7.1n(,H I 0/l-WJI,·,/' ',;,

Regression line of X on y :
X- X hxy ( y Y)
here hxy =r <Ix
Uy

= 0.72 X 1 2.5
8.5
= 1.0588
Regression line is
X - 42 = 1 fJC:-Jf' ?r/ ,,,J • JJ ,,,-,)

X = 42 - 9?J. 174 / + -iJ,J.-1~0

r -, ,
. . X = -51.1 744 + 1.0588 y
. X y
Est1mat10n of the production wh en th e capacity
, . ut111zatJrJn
. . ~10,lr
1s , 1 ---,;--- ·,,
qua 10n on , where y = 70 - - - -=- - -- •
E t
Regression Equation of X on y When y =70
X - X = hxy (Y - y )
X = -51.1744 + 1.0588 / 7G
= -51.1744 - 74.:: ':i
= 22.9416
Therefore, the estimated production would be 22, 94,160 .1::-.i:s ·:.:1a:::-. ::.a:,a: :s ::
capacity utilization of 70%.

Example 10: Given data is as below:

X: 42 36 55 58 35 65 60 50 48 S1
Y: 98 93 110 85 105 108 82 102 118 99
Determine the regression equations using deYiations frox :::ieans.
Find the correlation coefficient also. Estimate the value ofY whe::. Xis.!:

-
'),, - x·•
'
Solution: x2 .

y X y
X =X- X . =Y-Y 64
.! 16
-8 -2 .19 98
42 98 196 50
I -7 l)O
93 -14 "~-
..!
36
5 I 10 ::5 -120
55 110
i -15 64 _-5
8 225 25
58 85
105 -15 IS
35
 120
180
15 8 0
108 -18 -36
82 IO
0
2 -1
102 18
118 -2 Ixy=-12g
-1 y == 120
I
99
Lx===O
Ly::::O
Ix= soo Ir=== 1000
_ I;Y ~:::: 100
X - ~_Lo_=
X
-500 = 50 I
y == -N == 10
- N 10 .
Regression coefficients b == fq == mo - -0.1143
-128 _

l:xy _ ~ == -0.1416 xy IY
byx = ~ xz - 904 X.
L, •
Regression equattot:
yon .
Y- y = byx(X-X) 50)
y _ 100 = -0.1416(X -
y = 100 - 0.1416X + 7.08)
y = 107_08 - 0.1416X

· X on Y·.
Regression equat101:
X - X = hxy (Y - Y)
X - 50 = -0.1143(Y - 100)
X = 50 - 0.1143 Y + 11.43 )
X = 61.43 - 0.1143 Y
Coefficient of correlation = r = ±Jhyx ~x_h_x=---y-.-----:;:::-:;-::;-:;-;::;-::;-
r = -J(0.1416) X (0.1143)
r = -0.1272
Estimation ofY when Xis 45,
Regression equation Yon Xis Y = 107.08 -0.1416 (45)
Y = 100.708

Example 11: The regression lines of a sample are x + 6y 6, and 3x + 2y =

10. Fmd i) =
sample means xand y. ii) coefficient of correlation between x and y. Also estimate y ,, hen x =12
Solution: i) Mean i and y are obtained by solving the two given equations.
3x + 18 y = 18
3x + 2y = 10
Subtracting, 16 y = 8 y = 2~ ,x = 3
ii) If the line x + 6y = 6 is the line of regression of y on x
then 6y = -x + 6 '
1
.
Y=--x+1 6
bYX -- - -l
.
:the line 3x + 2y = 10 ts the line of regression of x on 6
~~::::-2y+10i.e.x= -~y+~ Y,
i 3 3
 Linear Regression I 47

r = ±jbyx X hxy

r=± j(-~)(-¾)
r == -~ (negative sign since hxy and hyx both are negative)
iii) To estimate y when x = 12.
Line of regression of y on x is y = - ~6 x + 1
When x = 12,y == -2 + 1 = -1
Example 12: Given 6Y = SX + 90, 15X = 8Y + 130, CY} = 16.
Find i) sample means x and y. ii) coefficient of correlation between x and y. iii) Cl}.
Solution:
i) To find i and y,
Multiply the first equation by 3.
- 15 X + 18 Y = 270
15X - 8Y = 130
Adding, 10 Y = 40 Y= 40
X=30
ii) To find coefficient of correlation
Consider the first equation as line of regression of X on Y.
6 6
X = SY - 18, hxy = S
Consider the second equation represents the line of regression of Y on X.
15 130 15
Y=
8
X-
8 , byx=
8
r = ±jbyx X hxy

r= (~) c:) = ~ = 1.5 which is not possible as value of r can't exceed 1.

Now, let's take the first equation as line ofregression of Yon X,
5 5
Y = X + 15 byx =
6 6
Let the second equation be line of regression of X on Y.
8 13 0 8
X = 15 y + 15 bxy = 15
r = ± jbyx X bxy
r = =
~
- =
9
0.667

iii) For ay, byx -- r -<Jy

<Jx
S 2 <Jy
-=-X-
6 3 4
aY =5 .
 . are x:+ Jy-5 =0
ess1on
. s ofregf O 5
tes . Jf twO IUle ient is + · ·
. ·ustificatton. tion coeflic ndently
State true o
r false8 ==w1thJ
Othen
the correla fi nnances by indepe
and 4x + 3y - atic per o
[Ans: False] B graded dra111
. d es A and
2. A panel o~tw~;k! as follow;- 5 6 7 b .
awarding e No. I, 2, 3, ' ' 2, 32 34 oot 38 marks )
Performan~6 32 34 31
MarksbyA. 33 31, 30,
;4
32 36
h'chjudgeBc
Marks by B : 35, 'ce however, w I t how many mar
kswould e
h
ould not attend. be expected to

e? ,
B had 'also been presen
The eighth ~erfonnan
·udge A. IfJudge i hth perfonnanc .
liave awarded to thee ff_
33), 37] . make and annual
[Ans: y - 33 = 0.6 7 ( . " aoe of cars of a ce~am osts related to age.
3.
From the following tabl:t!~~:';e;.ession equati~n or
. tenance costs, o 4 6 '
°g
JO 12.
mam ( )· 2 ' ' 1 000
AgeofCars years. 1 700 2,100. -· .
Annual maintenance
) 600 1 500 1 800, 1,900, , ,f the same mal-..c
Cost (Rs. . 1, t of , ' . . ' a 3 years old car o
, maintammg
Find the approxunate cos y = 142 9. 98 + 5 2.86 X • h'
.
. and hence the equat10ns
4. Find the coefficients of regress10n ofth e lines of rL'~rL'~~JL)l1
- tor t c

following data, 5 82 90 62 65 3 9.
X 78, 36, 98, 25, 7 , , , , , 47 .
Y·. 84, 51 ' 1·91 , 60,
Of
62, 86, 58,
68,gression your. equa t'IOn son the boraph. Esttm~Hc tlh.' \ .ilnL'
from53, L1t

Draw the mes re fX h y = 90 from the graph

y when X = 50 and the value o w en . . , , •1
What is the significance of the point of intersection of the two lmcs
A chemical engineer is investigating the effect of ~rocc~s opcrn u ng tcmpcr~1ture
S. x on product yield Y. The study results in the following datu.
X: 100,110,120,130,140,150,160,170, 180, 190.
Y: 45, 51, 54, 61, 66, 70, 74, 78 85, 89.
Find the equation of the least square line which will enable to predict ) icld nn the
basisthe
and temperature. Find also the degree of relationship [r] between the k'111per,1ture,
ofyield.

6. Find the angle

LY =630, between the lines of regression using the fo II owing da t:,. N
O'x = 4 O'y = 5, rxy = 0.6
J O }: 270,

Discusstethe 2 h "The sum of the two coefficients of reg rcss •

th statement • l
grea r an r w ere r 1st . of correlation''.
· he coefficient · 1tn1sa
0 \\ays
<iy==Sx + 90, 15x= 8y + 130, u;=
16.
Y aud l,_, (ii ) r and (ii) uf
 I incnr Regression I 49

8. The equations of thr two l'l'gH·ssion lines Htl' Jx 12y 2<i and <ix I y ,I
Find: (i) thr means of' x and y,.
(ii) rncflkil'nt or rnndation lwtwl•e11 x and y,
(iii) Oy if 0~ J
9. The equations of the two n·gn·ssio11 li,ws n11· x I t,y ,, n11d h 1 2y IO
Find: (i) the means or x n11d y,
(ii) coeflicienl or rn1 rl'lnt1011 lwlwt·c11 x :111d y,
(iii) Also est1mall' y whrn x I> .
10. If the tangent and of'tlll' a11gk 111:tdt· hy the li11l' td ' rej'rl'ss io 11 ol y 011 x, rs O.<, and oy =
20'x . Find the rnrrl'lc11io11 rnl'lf1r1L't1I hl'IWl'l'II x :111d y.
1
11. If O'y - at o and the angle helwee11 lhl' lt11cs ol'rcgrl·ss io11 is ta n \, li11d the
corfficienl or condation .
12. The followin g datn n.:gardin g lh c hcrght s (y) a11cl wc rgltl s (x ) or 100
students arc given college
}:x- 15000, xL 22725 00 , LY <,X00, )~y 1 4<, HJ>..,, )~ \Y I 02225 0
Find the coeffici ent or com.: l;1t1011 bclwt:c11 hL·if hl and WL'ighl ;111d al so the equ ation
of regress ion or hei ght a11<l we igh!.
13. It is given that the mean s or x and y an.:::, a11 cl 10. l f't lw J111t· ol'rL·grcss 10 11 of' y on
X is parallel to the lin e 20 y 9 x 1 40, cslinia lc Ilic valuL' yor ro,
x rn .
14. The following tabl e gives th e age of' car or a ccrlain make and a11n11al 111a 1n lc1ia ncc
cost. Obtain the equation of lhc line or regression of' cost 011 age.
Age of a car 2 4 6 8
Maintenance 1 2 2.5 3
15. Obtain the equation of the line of regression of yon x from lhc following data and
Estimate y for x = 73
x: 70 72 74 76, 78, 80
y: 163, 170 179 188 196 220.
16. The heights in ems of fathers (x) and of the eldest sons (y) are given below.
x: 165 160 170 163 173 158 178 168 173 170 175 180
y: 173 168 173 165 175 168 173 165 180 170 173 178
Estimate the height of the eldest son if the height of the father is 172 ems . and the
height of the father if the height of the eldest son is 173 cm. Also find the coefficient
of correlation between the heights of fathers and sons.
[Ans: (i) y= 1.016x- 5.123, (ii) x = 0 .476 y + 98.98, (iii) 169.97, 173.45, (iv) r = 0.696]
17. Find (i) the lines ofregression, (i) coefficient of correlation for the following data.
x: 65 66 67 67 68 69 70 72
y: 67 68 65 66 72 72 69 71
[Ans.:(i) y = 19.79 + 0.72 x, (ii) x = 33.29 + 0.5 y; r = 0.6]
18. Find the line ofregression for the following data and estimate y corresponding to x = 15.5.
X: 10 12 13 16 17 20 25
Y:19 22 24 27 29 33 37
[Ans.: y= 0.8 x + 13.23, 25.63]

tiMtri II:
r· If"

SO I S1atistical Methods ing data o 20 1~

the follow 20 22 4 15 1
ssion for 26 35 16 21. . 19 and
19. Find the two linedo(~fzs 22 ~~ 17 22 t~~ age of wife is
Age ofh~sban . . 18 15 sband when is 30.
Age ofw1fe(yrs): the age ofhu- [husband 4
Hence estimate (1~fi when the age~ 385 x + 7.3
th ge of w1 e 6 & y ::::: .
(ii) e ~ 2 23Y - 12.7 d ===30 years
[Ans: x - . e ofhusban rs]
Required the ag f wifo :::: 19 yea obtained: 1 .
. d the age o ults were 4 f corre at10n.
Require . > I following res 54 ~xy == 36 h coefficient o
,auons tic
0 From 8 obsef\ ., --,4 '("y :::: 4 ,~ O ~yz== 2 '~
·on ofx on y and t e
2 . 9 t'_r-=== .,_ • '- f regress1 d
1:x= 5 · '- . of the line o . of X on y an
Find the equat1~1 5 r = 0.981] b a line of regression
Ans: x== I .5 y - . , . b low which can e
· s given e
f th two equation 3y- 8 == 0.
21. Outo e , ,_ 6 ~ oand2x+ h following
Why x -i- - ) 8 = OJ . elation data t e
[Ans: 2x + 3y - d of analysis of corr sion
artially destroyed labora;"':';";~uations of the lines of regres .
O
22. In a P ible. Variance x ' . •') coefficient
results are leg O 20x - 9y - 107 = 0. d rd deviat10n of y, (u
4x -Sy+ 33 = ' f x and y, (i) the stan a
Find (i) the mean values o
of correlation betwe~Ox6a]nd y d
13 1 7
[Ans: , , 4' r- . . s of regress10n. are 5x- y =22 an
23. Given: var (x) = 25 · The equations of the two 1me
.
64x - 45y = 24.
Find i) the means of x and y,. dy
• of corre1at1· on
(ii) coefficient - between x an ·
[Ans: byx =3.5, hxy = 0. 249, r - 0.93]

2.5 Multiple Regression

In simple regression between two variables, X an~ Y, we ascerta11:1

· th ere1ative
· move ment of
the dependent variable for a unit movement in the independent vanable.

Multiple regression analysis represents an extension from two variables to three or more
variables. In multiple regression, instead of one independent variable, two or more
independent variables are utilised to estimate the values of the dependent variable. Thus, the
effect of two or more independent variables on one dependent variable is investigated.
2.6 Regression Equations

fit,ession equation is an equation for estimatin .

gent variables. It is referred to as the regr g ~ depende_nt vanable, say Y, from the
ess1on equation of y X
on .
 Linear Regression I Sl

The ~r~ced_urc for studyi_ng multi~lc regression ii,; analogous to that of simple regression, with
the d1stmct1~n that adchtional variables an; incorporated into the regression equation. If there
arc three vanabk 5 X1, X z a nd X.1 the multiple regression takes the following form:
X, a1.:-!:1 I h,'J..:1X1, + b.1 3.2X3
In this equation, th(.,'fC are three c;onstanls a 1 .z:i , b %.:i and h, . The subscript after the dot
1 32
signifies the variables that arc held com;tant. h12 .:i is the slop·~ of regression line of X on X
when X3 is held constant. b, :i.2 is the slope of regression line of X on X when X is1held 2
constant. 1 3 2

a 1.23 is the intercept of the regression plane. It gives the value of the dependent variable
when all the independent variables are zero.

2.7 Normal Equations for Multiple Regression Equations

As in simple regression , there exists a regression line approximating a set ofN points X and
Yin a two-dimensional scatter diagram. Similarly, there exists a least squares regression
plane fitting a set of N poi ntsX1 , X 2 and X3 in a three-dimensional scatter diagram.

The regression equation of X1 on X2 and X3 is

X, = a,. 23 + b12 .3 X2 + b13.2X3

In this equation, the values of b12 .3 and b'J3.Z arc determined by solving simultaneously the
following three normal equations .
.Z:X1 = Na1.23 + b12 .3 IX2 + b13.2 LX3
LX1X2 = aJ .23 LX2 + b12 .3 rx/
+ b13.2 LX2X3
I:X1X3 = a1 .23LX3 +b12.3LX2X3 +b13.2LX3 2

The regression equation of X2 on X1 and X3 is

X 2 = a 2 .13 + h2uX2 + b23.lX3

In this equation, the values of b2 u and b23 ., arc determined by solving simultaneously the
following three normal equations.
I:X2 = Na2.13 + h2u LX1 + b23z1 LX3
L X1 X2 = a 2 .13 I X1 +- b'l.u L X1 + b23.1 L X1X~
L X2X3 = a2.13 L X.3 + b'l.1.3 L X1X3 + b23.1 L X3

The regression equation of X3 on X1 and X2 is

X3 = a3.12 + b31.2X1 + b.32.1X2 . .

In th1s
. equat·10n, th e va Jues of b 31.2 and b 32.1 arc determmed by solvmg smmltaneously the
following three normal equations. ~
LX3 = Na3.12 + b31.2LX1 + b32/L..X2
~x1X3 -- a 3.12 L..~ X1 + b312LX1
L.. · + b32,1LX1X22
L X2X3 = a3.12 L X2 + b31.2 L X1X2 + b32.1 L X2
 111
,lilsMultlpl& Repslon l!llfd*doJJS \V)lell J)el'fadons are Taken fro

Actual Means
Let X1 , X. ,Y," be the actual 2
ans of the tJ,ree variables X1 , X and X3 respectively. The
1110th••• regression equ~tions are_considerably simplified ifth
~al~~tions invol""'! in solving
eviattons of the vanous yariables are taken frO!ll thetr respective means- e
:x, = (X -XJ,x, = (X, -x,J,x3 = (X, -X,).The multiple regression equation tak""'
e following
1 shape: ,,
To x, = b,z_,x, + b13.2X3
find the values of b . and b _ , nor;n•l equations are given below.
13 2
12 3
L X1X2 == b12.3 L X2 + b13.2 L X2X3
L X1X3 == b12.3 L X2X3 + b13.2 L xi

The values of b12.3 and b13.2 can also be obtained as follows:

17

b12.3 = r,12.3 X -0'2.13

0'1.23 , b13 .2 == r13.2 x -0'31.3,122

3
, , -anX, . -2.7
E!ample 1 : Given r12 = 0.28
FmdtheregressionlineofX d;'X-Od.49,r31 = 0.5 1,cr1 = 2.7 ,2
cr = 2 .,<T
4 -

Solution:
x 3 -onX
The regression equation ofX .
. given
b 1 andX2 is by
31.2X1 + b 32.1X2 (i)
3 -

b 31.2 = -0'3 x r 31 - r12r23

27 0'1 1- r:2
b312 = 2:7 x 0.51- (0.28)(0'.i9)
l - (0.28)2 = 0.405
b32 ·1 = 0'3 X r 23 - r 12 r 13
0'2 1 - r:2
12

. = 2-7 0.49-
Puttmg these values in . 2.4 x (0.28) (0.51)
equation (i) 1 - (0.28)2 = 0.424

X3 = 0.405 X1 + 0.424x2

'~lVQA the follow.

t!j,2 :;::; 0 mg data
.8,r13 = 0.7 = 0.6,O"i - 10
I T23

™'tion of Xt an X2and X - ' a2 == 8 , 0"3 == 5

3•
 Linear Regression I 53
Solution:
The regression equation of X1 on X2 and x · •
3 1s given by
X -
1 -
b12_3Xz + b13.2X3 (i)

h 1 2.3 = 0'1 x r12 - r13rz3

CTz 2
1 - r:23
1
b3 1 2 -_ - 0 0.8 - (0.7)(0.6)
. 8
X ---=---.::....:__:_
1 - (0.6)2 = 0.742

_ 10 0.7 - (0.8)(0 .6)

-5 X 1 _ (0. 6) 2 = 0.6875
Putting these values in equation (i)

x1 = 0.742 x 2 + 0.6875x3

Miscellaneous Exercises

1. What is regression? How does it differ from correlation?

2. Define regression lines. Derive the equations of the regression lines of yy on xx and xx
onyy.
3. Explain the method of least squares for estimating regression coefficients.
4. State and prove the properties of regression coefficients.
5. What are the uses and limitations of regression analysis in real life?
6. Show that the correlation coefficient is the geometric mean of the regression coefficients.
7. Discuss the relationship between regression lines and the coefficient of correlation.
8. Under what conditions do the two lines of regression coincide?