FurdamentaJ.. otMatbematlcal Statlat1e.

= _f!E[(Y -E(Y)V1
a -
=_.f!.
a
ayl

E(X-E(X)J2 =~E[(Y _E(Y)}2J = ~. ayl

_ £. . a y2 _ £. a y2
r= a =_-=a:.....-_
-fayl" ~. ay2 I ~ :1 ay2

={+ I, if b and a 'are of opposjte signs.

-I, if b and a are of same sign-.
Example 10·5: (a) If Z =aX + bY arid r is the correlation coefficiem
between X and Y, show that
ai! =ala;' + b2aYJ. + 2abrax ay
(b) Show that the correlation coefficient r between two random variables X
and Y is 8iven by
r =(ar + a.r -ax -.r) /2axay .
where ax, ay and aX_yare the standard deviations of X. Y and X - Y
respectively.
[Calcutta Univ. B.Se., 199~; M.S. ~qroda Univ. B.Se. 1992]
Soluti~D.•. Taking expectation of both sides of Z =aX + bY, we get
E(Z) =aE(X) + bE(Y)
Z -E(Z) =a(·X -E(X)} ,.. b.(Y ...,E(y)}
Squaring and taking expectation of both sides, we g~t
a'; =alar + bla.r + 'lab Cov (X, Y)
=ala;' + 1J2aYJ. + 2abrax ay
(b) Taking a =I, b = -1 in the above case, we have
Z=X-Y and ax_YJ.=ax2+aYJ.-2rajcay
ax 2 + ayl - ax _ yl
.. r= 2ax 01
Remark. In the above example, we have obtained
1 • , ,
V(aX + bY) = a2 V(XJ + b2 V(Y) + 2ab ~ov <)f;
-
n
Similarly, we could obtain the result_
V(aX - bY) =0 2 V(X) +·b2 V(Y) ~ 2ab Cov'(X: n.
The above results are useful in solving theoretical problems.
Example 10·6. X aM Yare two random variables witll variances aJ-
and ay2 respectively and r is the, coefficient of correlation between them. If
U = X + kY and V =X + ~
qy
Y" find the value of k so that U and V are
-
UllCorreialed. [Delhi Univ. B.Sc;. 1992; Andhra Univ. B.Se. 1998]
Conoelationand~ion 10·11

Sqlution. Taking expect~tions of U =X + kf and V =X + ~ f, we get

E(U) = E(X) + kE(Y) and E(V) = E(X) + ,CJx E(Y)

, CJy
U -E(U) = (X - E(X)) + k(f - ~(f») aQd

V - E(V) =(X -'E(X») + CJy

CJx (f - E(Y) )

COy (U, V) =E[(U -E(U)) (V -F,(V»)]

= E[(X -E(X)) + k(f -E(y»)1.x

. [(X -E(X)) + CJy
CJx (f -E(Y»)]

= CJx2 +~x COy (X. f) + k COy tX. Y) + k CJ.\ • CJ~

CJy ~y

=[~x2
. "
,+ kCJXCJY] + ['CJX ... k.] COy (X. Y)
CJy' ",a.

rCJx'+ kCJY] Cov ("

= CJx I.\CJx + kqy)- + [CJy X...V\.,

.=.(CJx + [cov(x.n]·,
CJxkCJy)
- CJy
= (CJx +
(+ kay) (1 + r)CJx
,-
U and V will be uncorrelated if
r(U, V) = 0 => COy (U, V) = 0
i.e., if (c;sx + k~y) (I- + r) CJx = 0
=> CJx+kCJy=O (.:CJx~O,r~-I)
CJx
=> k =--
CJy
Example 10·7·. The "randOm variables. X ,and Y are jointly normally
distributed and U and V.are·dejine(!by ,"
.. U <;: X cos· a + f sin a,
V = y.' cos a - X sin a
Show thai V and V will be pncorre,lated if
2rCJxCJy
tan2a
." = CJx2 r- CJy. .2',
where r =. COTTo (X, fJ, CJ~ = VaT (XJ and CJ; =- Var (V). Are U and V then
independent? .
[Delhi Uraiv. B.Se. (Stat. Bora••) 1989; (Math •• JI01&ll.), 1990]
Solution. We have
COy (lI, V) =E[(U - E(C!» [V - E(V)}]
=E[[(X -E (X)) cos a + (f -'E(y») sin a]
x [[f - E(Y») cos a - (X - P(X») sin a]]
 Fundamentals otMa~ti~ Statistics

= cos2 a COy (X, Y) - sin a cos a.aJl

+ sin ~ cos a.a,z - sin2 a (Cov (X. Y)
= ·(cos2 a - sin2 a) COy (X. f) - sin a cos a (a~ - ay2)
= cos2a COy (X, Y) - sin a cos a (a~ - a,z)
U and V will be uncoitelated if and only if
r(U, V) = O. i;e., iff <;ov (U. V) = 0
i.e.. if cos 2a Cov (X. Y) - sin a cos a (ax 2 - ay2) 0 =
sin 2a
or if cos 2a r axCJy =-2-. - . (ax 2 - a,z)

2r axCJy
or if ·tan 2a = .2 . 2
ax - ay-
However, r(U. V) = 0 does ~ot imply that the variables.U and V are
independenL [For detailed discussion,see Theorerpl0·2, page 104.].
Example 10·S. [IX, Yare standardized random variables. and
1.+ 2ab
r(aX + bY, bX + aY) = a2 + b 2 " .(*)
find r(X. Y), the coefficient 01 correlation between X and Y.
[Sardar PaI,1 UIIi". B.Sc., 1993; Delhi UIIi". B.Sc. lStat. Hons.), 1989]

.
Solution. Since X and Y are standardised random variables, we have .
E(X) = E(Y) =0 . j'
and Var (X) = Var (Y) ='1'~ E(X2) = E(f2) = 1
and Cov(X, Y) = E (X Y) ~ E(XY) = r(X,y).axCJt'
,.:(**)
= r(X,Y)
Also we have
r(aX + bY, bX + aY)
_ E[(aX + bY)(bX + aY)] - E(aX + bn E(bX +'an
- [Var·(aX + bY) .. Var{IfX + aY)]m
= E[abX2,+ a 2 XY + b 2 YX + aby2] - 0
{[a 2 Var (X) + b 2 Var (Y) + 2ab COY (X.y)]
x [b 2 Var (X) + a 2 Var Y T 2ba Coy (X,y)])·l/2
. ab.l + a 2 rex, Y) + ·b 2 rex. Y) + ab.l
= ([a 2 + b 2 .+ 2ab r(X, f)][bl + a 2 + 2ba r(X, Y)]) 1(1.

[Using (**)]
_ 2ab + (a 2 + b 2). reX. n
- a 2 + b 2 + 2ab. r(X, Y)
From (*) and (**). we get
1 + 2ab _ (a 2 + b 2). r(X, -y) + 2ab
a2 + b 2 - a 2 + b 2 + 2ab: r(X, Y).
.
Cross multiplying, we get
 10·13

(aZ + IJ2) (1 + lab) + lab. r(X, y) (1 +"2ab) = (a2 + 1J2)2. r(X, y) + lab (a2 + IP)
:::> (a4 + !J4 + 202b 2 - 'tab - 4 a21JZ). r (X. Y) =(a2 + IJZ)
[(a2 _1JZ)2 - lab] r(X; Y) = a 2 + b2
a2 + b2
r (X, Y) =(a2 ~ b2)2 _ 'tab
Example 10·9. If X a¢ Y are uncorrelated random variables with means
zero andvariancesCJI 2anda,.2 respectively, show thai
U =X cos a + Y sin a, V =X sin a - Y cos a
haVe a correlation coefficient p given by
CJI2 - CJ22
p = [(CJ12 - CJ22)2 + 4CJ12cJ2~ cosec2 2a]1I2
Solution. We are given that
= =
r(X, Y) 0 ~ Coy (X, Y) 0, CJll CJ? and = CJ22CJr = •.. (1)
We have
CJU2 = V(X cos a + Y sin a)
= cosla V(X) + sin2a V(y) + 2 sin a cos a Cov (X. Y)
=cosla CJI2 + ~in2a CJ22 [Using (1)]
Similarly, L.
CJ'; =V(X sin u - Y cos a) =sin2a.CJ12 + cos2a.CJ22
Cov (U, V) =E[(U -E(U)} [V -E(V}}]
=E[ {(X -E(X» Cos a + {(Y -E(Y) sin a}
x [(X -E(X» sin a - (Y -E(Y» cos a}]
=sin a cos a V(X) - cos2a Cov (X, Y)
+ sin2 a 'Cov (X, Y) - sin a cos a V(Y)
=(CJ12 - CJ22) sin a cos a [Using (1)]
2 _ [Cov (U, V)]2
Now p - CJU2CJv2
= (cos2a CJI2 + sinZa CJ22) (siJl2a CJ12 + cosZa CJ/)
== sinZa cos2a(CJ14 + CJ24) + CJl2a22 (cos4 a + sin4a)
= sinZa COS2a(CJI4 + CJ24) + CJl2a~2[(sinZa+ cos2a)2- 2 sin2a cos2a)
=sinZa cosla (CJ1.4 + CJ24 - 2CJl2a22) + CJI2a22
=sinla cos2a (CJ12 - CJ22)2 + CJ12a·i
rr= CJl2a22 + sip2a COSla(CJ1 2 - CJ?)2
~(CJ12 - CJ22)2 sin 2 2a
= CJ12CJ22 + sin2 2a. ~ (CJ12 - CJ22)2
 ~ qfMatbematical StatWtio.

(a1 2 - (22)2 ,
= 40'12cr22 COS~~ta + (0'1 2 -'-.0'22)2
_ 0'1 2 -0'22
=> P- 2 2 20
[(0'1 - 0'2 )2 +'40'1 22 cosec22a]lf2

Example 10-10.11 V = aX + bY and V = eX + dr, where X and Yare

m({lSured Irom, their respeetivtf means and if r is the eo."elation cOl'ffi.eient
between X and Y, and if V and V are unCo"elated, sho.w that
, =(ad - be) O'xay (1 - r2)1/2
auav
[Poont;J Univ. B.Sc., 1990; Delhi Unii1. B.Sc. (Stat. Hons.), 1986]
Sqlution. We have
_ Cov (X, 1 1 [Cov (X,
r - ax ay Y) =>
2_
- r - ,- Gx 2 arY)J2

=> (1 - r 2) axl O'r =O'x ar - [Cov (X. Y)]2

2 _..(*)
[This step is suggested by the answer]
V =aX + bY , V =eX + dY
Since X. Yare measured from their means,

E(X) = 0 = E(Y) => E(V) = 0 = E(V) }

",,(**)
=
au 2 E(V2); a~ E(Vl) =
Also aX + bY - V = 0 and eX + dY -v = 0

X . Y 1
-bY +' dV= -eV + aV= ad -be

X = ad:...
1 be (dV -bV). }
.. ,(***)
Y =ad ~ be (-e V + a V)
Var (X) =(ad _1be)2 [til O'el + b20''; - 2 bd Cov (V, V)
=(ad ~ bc)2JiPael + b2 6';]
[Since p. V are uncorrelated ~ =
Cov (V. V) 0]
Similarly, we have

Var (Y) =(ad ~ be)2 (e~ a.cl. + Q2 aif).

Cov (X:y) =E(XY) -E(X) E(Y) =-E(XY) ['.' E(X) = 0 = E(Y)]

= (ad ~ be)2 E[(dV - bV) (~V + aV)] [From (***)]

~tionandReer-aion 1~.21

Su~mation of product of deviations of X and Y series from ~ respective

arithmetic means = 122.
ADS. , (X. Y) = 0·891
(b) Coefficient of correlation between two variables X and Y is 0·32. i'iteir
co\'llriance is 7·86. The variance of X is 10. Find the standard deviation of)'
'es.
set! (c) In t~o sets of variables X and Y with 50 observations each. the
following data were observed :
X=10. C1x =3. Y= ~. C1y =2 and ,(X. y) =0·3
But on subsequent verification it was found that one valu~6f X (= 10) and
one value of Y (= 6) were inaccurate and hence weeded out With the remainin&,
49 pairs of values. how is the origipal value of , affected? '
(}Iagpur Uni(). B.Sc., 1990)
Hint. IX = nX = 500. ry = nY = 300 ~
IX2 =n(C1~ + X"Z) =5450. ryz =50(4 + 36) = 2000
, C1x C1y
LXY -)(
= Cov (X. y) = - n-
-u-
I

=> 0·3 x 3 x 2 =lfoY - ~O x 6

=> rXY = 5q(I·8 t 60) = 3090
After weeding out the incorrect pair of observatioJ;l, viz .• (X = 10, Y = 6),
lite corrected val~es of IX, rY,~, rf2 and IXY for the remaining 50 -1 =
49 pairs of observations are given below :
Corrected Values ;
LX = 500 - 10 = 490 ; ry =300 - 6 = 294
L XY = 3090 - 10 x 6 = 3090 - 60 =3030
L XZ =5450 - 102 = 5350. ryz =2000 - 62 = 1964
." _ Corrected Cov (X. Y) =' 90/49 = 0.3
'.. , = (Corrected C1X)x (Correcte~) C1y '450 -200
~49)( 49
Hence the correlation coeffiCient is invariant in this case,
(d) A prognostic test in Mathematics was given- to 10 students who were
about to begin a course in Statistics. The scrores '(X~ in their test were-
examined in relations. to score~ (Y) in the final examination in Statistics. The
following results were obtained :-
r X = 71, r Y =70,'l: XZ = 555, r yz =526 and r XY = 527
Find the coefficient of correlation between X and Y.
(Kerola Uni(). BoSc., 1990)
7. (a) Xl andXz are independent variables with means Sand 10 and standard
deviations 2 and 3 respectively, Obtain ,(U, V) where
U = 3XI +4Xz and V = 3XI -Xz
Ans.O (Delhi Uni(). B.&. ,1988)
10.22 Fundamentals 'of Mathematical Sta~

(b) If X and Y src ~ormal ~d im,ependent with zero means and standard
deviations 9 and 12 respectively, and if X + 2Yand leX - Y are non-"correlated.
find k.
(c) X. Y, Z ~e random variables each with expectation 10 and variances I.
4 and 9 respectively. The correlation coeffiCients are
r(X, y) =0, r(Y. 2) =r (X, y) =1/4
Obtain the numerical values of :
(;) E(X + Y - 22), (il) Cov (X + 3, Y + 3), (iii) V(X - 2 Z) and
(iv) Cov (3X, 52)
Ans. (i) =0, (il) 0, (iii) 34, and (ill) 45/4.
(d) XaDd Y an. discrete random variables. If Var (X) = Var(Y) = CJ2,
2
Cov (X, Y) = ~ , find (i) Var (2X - 3y), (il) Corr (2X + 3, '2Y - 3).
8. (0) Prove that:
V(aX ±by) =a2V(X) + blV(Y) ± 2ab Cov (X. Y)
Hence deduce that if X and Y.are independent
V(X ± Y) =V(X) + V(Y)
(b) Prove that correlation coefficient between X and Y is positive or
negative according as
CJx + Y > or < CJ x _ y
9. Show that if X and Y are·two ranc;lom variables each assuming only twO
values and the correlation co-efficient between them is zero, then they are
independent Indicate with justification whether the result is true in general.
Find the correlation coeffident between X and. a - X, where X is any
randOm variable and a is constant.
10. (a) Xi (i =1, 2, "3) are uncorrelated variables each having the same
standard deviation. Obtain the correlation between Xl + X~ and X; + X3'
Ans. 1/2
(b) If Xi (i =1, 2, ~) are three uncorrelated variables having stimdard
deviations O'r, CJ2 and CJ3 respectively, obtain the coefficient of correlation
between (Xl + Xv and (X2 + X3).
Ans. /..J (CJ12 + CJ22) (CJ22 + CJ32)
CJ22
(c) Two random variables X and Y have zero means, the same variance (J2
and zero correlation. Show thlilt
V =X co~ (l + Y sin a and V =X sirta - Y cos a
have the same variance 0'2 and zero correlation. )
Ulangalore Uni". B.Sc., 1991
(d) Let X and Y be iincorrelated random variabl~. If U =X + Y ~d
V =X - Y. prove that the coefficient of correlation between U and V IS
(O:x 2 - O'y2)/(CJX2 + CJy2). where CJx2 and .q,z
are v~ances of X and r
respectively. . ,
(e) Two independent random variables X and Y have the following
variances: O'xZ =36, O'yZ =16. Calculate the coefficient of correlation betWeen
U=X+YandV=X-Y