As COVARIANCE Tf (X% Y) is a two-dimensional random variable with E(X) and E(Y) fini, then the covariance of (X, Y) denoted by Cov(X,Y) is defined as Covi, Y) = {ELX ~ BOQyy — EW)]} = EY) ~ EWE) Note: (i) IfX and ¥ are indepc not be true. Git) Cov(x + Y,X-Y)= Var(X) — Var(Y) (iii) Cov(aX + bY, eX + d¥) = ac Var(X) + bd Var(¥) + (ad + be) Cov (X,Y) endent, then Cov(X, Y) = 0 but the converse need‘Two-dimensional Random Variables © 387 onREL ATION AND REGRESSION Cc cate distribution, we may be interested to find out if there is any varia iance between the two variables under study. If the chan 1 Oe covariance x b e change w enor affects a change in the other variable, the oN al : is a chang v te variables are said to oe ai If the two variables deviate in the same direction, i.e. if the later on . in on@ variable results in coi ing, increase ee (or decrease) nding increase ase ositive. But if they ise) in the other, correlation is said to be direct or p er deviate in the opposite directions, i.e. if the increase (or decrease) constantly ¢ ible results in corresponding decrease (or increase) in the other, in one a said to be inverse or negative, calalioh spe noted thal covariance is positive for positive correlation and ci for negative correlation. . |The closeness of relationship between two variables is not proportional Me correlation coeicient. 5, Ws Pearson Coefficient of Correlation Iris a numerical measure of intensity or degree of linear relationship between the wo variables. Correlation coefficient between two random variables X and ¥, denoted by yy (or rxy) and is defined as Cov(X,¥)___ Cov(X,¥) Oxyd, [Var(X) Var(¥) E(XY)— E(X)E(Y) . Ox Oy AX Y) = txy = (XY) = Pyy = ItX and ¥ are discrete random variables taking values x1, x2, ... , x and y1, Jy oy Yn Fespectively, then Covx, N= Yx-H-F)- iat Var(x)= Bl(r~ 3) 1=— YY ial Ae Var(1)= Ey WI=~ Diy - a Abonerties 71 The correlation coefficient lies between -I and 1, ic psi = -1spsl388 Probability and Random Processes y that Proof We know Z [™): [Eany bis o% |p 20 7 po py-E) [X = EQOIY - Eq) x50 EP] 2 ele = Alo Oy 2) -pe(X)P , £07) LEO? EQ) -TECOT ON 2p0¥,r)39 => Ox Oy e 1+1#20% Hs) 2 SAK Hs) Hence the proof. Pie veto coetient i independent of change of Sl digg Xd yoY=® where adh, k>0 Proof Let u=-5" ra Then, xen +a,Yokv + UM) = hE) + a, EY)= KE) + 8 o} = E(IX- E(X)P)} = BUC - EW)P = hoy Oy = hoy of = E(Y- BOP) = eV — EWP) = Kor = hk Cov U, 1) _ Covex.Y) _AkCovlU.V) _ CoWW.Y) _ gy, ¥) PED EG, hoykoy may Hence the proof. A Two independent random variables are uncorrelated if Cov(%, 1) = 0 But the converse need not be true. ACEAXIY = y) is known as the regression curve X on Y. SCE(IX = x) is known as the regression curve Y on X. A The correlation coefficient is also denoted by r (X, Y) = ryy- EXAMPLE 5.53 Let X and Y be any two random variables and a, b be constants, Prove that Cov(a¥, BY) = ab Cov(X, ). — [AU December '08]Two-dimensional Random Variables ¢ 389 | -gefiitions 1 ga vet 1)= BOE) — EC) EON) s covtak,b1) = Blak bY) - E(@EGN = abB(HT) ~ aE COBE(H) = ablEQY) ~ ECOEC = ab Cov(X, ¥) poo® rool, vr on By definition, ee covtk, N= EAN ~ EXE) = EIXC2X + 3)] - EQN) B2X + 3) = EC-2X? + 3X) - EX) [(2) EQN + 3] =-2E(X?) + 3E(X) + 2[EQ)) - 3EQ) E(x?) + 2[EQOP = -2(600) ~ [EOP =—2 Var(X) AMPLE 35S ‘Two random variables X and Y have joint PDF IY =-2X +3, find the Cov(X, 1). [AU May *08) x =, O0 fondle] e I Ban) = j j ay f(a, yaedy 54 = [Jo geo 19 Eee, ys Pe 64 (22) ~ 3x96, 3} 3x96\ 3 x 124, 64 _ 248 “3” 28827 FOX + 3¥) = 260) + 30) no = f2 EY?) = ie soar Zo 4 “Ty deg “42 | ag ~D=13 Var(X) = E(X?) — (EG)? 2 --() 28 3) 9Two-dimensional Random Varisbeg e391 128 4g 2 Ox 9 7 Ka 31Y 92 2_ WH 2 243-(31) _92 a Vat) = 02 =13 (2) a 7 = Covlt, N= EXY)- EXE) = eb 3S y 7 3° 9 Cont, 1) = 0, X and Y are uncorrelated ow 7 ave ft Two random variables X and ¥ have the following joint nF a 2-x-y, OSxS1LO = oO 0 ~y view eas Jrcam o,= V52 = E(UV)- EUU)E(V) Oy Oy — EQX*)= EO?) ~1E(X) + EQYLE(X) = EW) 52 _ E(X?)~[E(X)P - BW?) - LEP 52 Ww‘Two-dimensional Random Variables € 407 var(X)—Var(Y) _ 36-16 _ 29 Yat a OL Cs 52 3232] Blo 13 eh ttand Fare 69 independent random va y@bE Me yy =X cosa + Y sina and V = gd te" A jom variables with zero X sina ¥ cosa are also pit) = EO) cos a+ EY) sin a wn A= EO sina - E(Y) cosa we = EU)=0 gb 6 (X) EQ) = 0 (eX and Y are independent) sp tat gun) = EW) BU) ; ; jit) = EE cos + ¥ sin.) (X sin cr ~ ¥ cos 0] = F(A?) cos @ sin a — E(XY) costar + E(XY) sin? E(Y?) sin cosa = F(A) cos @ sin a ~ E(¥) sina cos a [- EQY) = 0) = cos @ sin a [E(X?) - E(Y)| SUE) = ELK cos @ +Y sin @) ELK sina - ¥ cos a) =[F(X) cos a+ EY) sin aq [E(X) sin a - EC) cos al 0 Iv EQ) = EQ) = 0) AUN) = E(U) E(V) = 0 only when E(x?) - = BN?)= EY?) + Uand V’are independent only when X and ¥ have equal variance. DUNPLE $67 Two random variables X and Y are related as Y = 4X +9. fad he correlation coefficient between Y and Y. [AU 2007] Soition Given: Y=ax+9 E(Y) = EX + 9) = 4E(X) +9 EY) = EIX(AX + 9)] = E(4X? + 9X) = 4E(X?) + SEX) Contk, Y) = BOM) ~ ECDEC) = EQ?) + 9E()] — EVD [4EQ) + 9] = 4E(X?) + 9E(X) — 4[EQOP - 9EX) = 4{EQ?) ~ [EQOP} = 40%408 & Probability and Random Processes ox Po Oxo OxOy Oy of = EO?) ~ EOP ‘ But, , = E[(AX + 9)"] — (EGX + IP = E(IGX? + 81+ 72X) ~ [4E(X) + 9)? = 16E(X2) + 81 + 72E(X) — 16[EQX)? ~ 81 — rey = 16(E(X?) -[E(X)P?} = 160% > oO; = 4oy Substituting in Eq. (i), we get ox =42e 21 Px = 475 EXAMPLE 5, Y and Z are uncorrelated random variables with 0m ‘and SD 5, 12 and 9 respectively. If U = X + ¥ and V = ¥ + Z, ingy correlation coefficient between U and V. TAU Ape Solution Given: — E(X) = E(Y) = E(Z) = 0 E(U) = EX) + EQ) = 0 EW) = EQ) + EZ) = 0 E(UV) = EU(X + Y) (Y + ZI =E (XY +.XZ + ¥? + YZ) = E(XY) + E(XZ) + B(¥*) + EZ) X, Y and Z are uncorrelated Cov(X, ¥) = Cov(¥, Z) = Cov(X, Z) = 0 and So, E(QXY) = EX) EY) = 0 and E(XZ) = E(¥Z) = 0 P ts E(UV) = EW?) Given: Var(X) = SD? = 5? = 25, Var(Y) = 127 = 144, Var(Z) = 9? = 81 CovU,V) —_ E(UV)- EWU)EW) @ 0° Wa WarV) Gy Ov aye Var(X) = UX) — [ROD = 52 = OM) 0 = BO? Similarly, and From Eq, (i). E(UV) = 144Two-dimensional Random Variables > 409 144-0 144 @ Cu Oy Oy -a, o% = BU?) - (EDP =El(X + YP] -0 = E(X?) + E(¥?) + 2E(XY) =25 + 144 = 169 o,=13 oF = EY + Z| = BY?) + E(2) + 2E(YZ) = 144 + 81 = 225 o,= 15 : . iuting in Eq. Gi), we get gsttting Lia ae os, P= 3x15 195 65