(t)1—A dt) + P,.,()adt + O(dt)? ty FG20-F0 apy sp.+ 2G Proceeding to the limit as dt 9, we get lim Reteat)-F) AP.) +4P, (8) Therefore, P()=-AP,( +P, (0), x21 where P/(t) denotes'the differentiation w.r.t.’t’. (ay For x=0," POs P_(t)=P (C1) calls in time t}=0. Hence from (11.4.2), weiget P(t ¥ dt) =P,(t){1-a dt) +0(at)* which on taking the limit dt 0, gives RADE =A BIO = ae a Integrating watt, we get Jog P(t}=-2t+C . ‘where C-is an arbitrary constant to be aetemined from the condition PO) = 1 i, Hence, log1=C = C= 6. Therefore, log P()=-at => P(t) =e* Substituting this value of P,(t) in (11.4.3), weget, with x=1 e P)=+AP (t+ 2078? POE) +AP(t)= Ae, This is an ordinary linear differential equation whose integrating fet ise“. Hence its solution is eM PA)=af eMeMdtsC, =at+C,ie Discrete Distributions G isan arbitrary constant to be deterind ——_.__ 419 W 20 ined from p eon 100)= 0 which vo, Pt)sen At. jn substituting this in (11.4.3) with x= ine oid We get PL) +APR()=Ae™ at : ting factor of this equation is and its solution i ion is pjet=[ teMe*atac, =O “x J +c, $+C procent- Hence » F{O)=0 which P= OY ‘ pnceeding similarly step by step, we shall get P= os x=0,1,2,.. tlisisthe probability function of a Poisson distribution with Parameter 2t meerder cy ee Poisson probability functions are given below for D1 2s a's o1 25 4.8 Fig. 11.4.1. Poisson probability functions. Theorem 11.4. If X is a Poisson variate with parameter?, then mean =,'=2 and Variance ==) Poof, The probability function of Poisson variate with parameter 35 as eee420 Fundamentals of Probability & Probability Distributions fog = SE, x=0,1,2 By definition . . ae mean =E[X]= E x = aye ' xt (X12)! j . nates “ogee \ AZ Go = ag x when y=x-=1 Again, var [X= EDX" —(EIX])* Now, E[X’] = E[X(X - 1) +X] soo seaete a2. > where y.=x 42 aMerer+r= +d. | Thus, var[X] = W42-a? = 2. Hence mean and variance of a Poisson distribution are equal. , ‘Theorem 11.4.2. If X is a Poisson variate with parameter A then, Bi=d and Br=3+¢ where Bi and Bz are the measuires of skewriess and kurtosis. Proof. By definiton, ei i sf 2 B= 4s .vand-Be= i ' as Pa where pg, Hy and py are central second, third and fourth moments of distribution. ; 7 natise cDiscrete Distributions 21 Wehave, from theorem 11.4.1 py’=A, py’= A442, and w= 2.Now, wi=EDC] + 2EIX(K = 1)(K 2) + 3X(K+1) +X] =EIX(K-1)(K-2) + 3[X(K- 1)] + EX] i Le gen a es ea or +30 +A aioe +a +d =v Lae! w=EDC] : ; = E[X(X- 1)(X - 2)(K —3) + 6X(X - 1)(K -2) + XX-1) +X] = a x(x 1)(x - 2)( = 3). oe +6E[(X — 1K = 2)] + EX - 1)] + EX] oMe? § ear +O 4 PHA OP LORE EA HMO + TED. Hence. Hs= Hy’ - 3Hy'by’ + 2(4')° = +3N+A-3A A+A) +20 =A and Hy = Mal ~ Abie’ uy’ + Oya! (ay)? 3(uy')6 = AS + 6034707 +A = AN(A +A? 4A) + ORAZ HA) SAF , HSN +2. : gle. abe a BAP +A 1 Therefore, B= Bs = 35> x and f= by Bessy. Thus the coefficients of skewness and kurtosis are = vB = Gandy, = fr-3= 1. Remarks : ” () Poisson distribution:is always positively skewed and leptokurtic. (ii) When A — ©, fy tends to zero and B2 tends to 3 which shows that a Poisson distribution tends to anormal distribution if A =.4225 —_~ Fundamentals ofProbability & Probability Dstibuons (i) Fist thre ceral moments are equal to in magnitude, variate is measured in some unit, the variance will be sq By iy unit and ps will be cubie of the unit although they ne : magnitude, 1142 Recmee Relation fore momentsof be Peso Digg Theorem 1143.1X isa Poisson vasite with parameter), then A Host = TAH * ie where 41 Hy and p,., are the (r + 1)th, rth and (r — Vth ecg moments of the distribution. Proof. By definition, the rth central moment of the distribution is w= EIK-EOO} ee “. SEIK-al‘ eye c =Ee- a uy Differentiating (11.4.4) with respect to A, we have: a, tet, nent. ged a & =f) yea § CM eta 1 vel ea, : eority § E (esa) Bark t poet: Ap,» Hence pyr= 1 Ab. ae The relation (114.5) is known as the recurrence relation for the mores ofa Poisson distribution, * Now, ifwe put r=1,2,3 in (114.3), we get us) women dpe, since == and he pyr een dan and jute ways B= Reh.crt Discrete Distributions 2 Theorem 11.4.4. Let X be a Poisson variate with probability function fig MSE; x= 0,1,2, then My(t) =e", y(t) = MY and Pfs] =e"? and where My(t), x(t) and P[s] are the mgf, cf and pgf of X. Proof. @) By definition, the mgf of X is e tye Mi)= He] = £ e* SE = EOE tne adit, : & (ii) The characteristic function of X is et) = Ble} = & (ii) The probability generating function of X is af. PIB = Este net EPL thet, Wecan find different moments of the distribution from mfg or cor pg. We have, from theorem (11.4.4) M(t) ==, By definition, rth raw moment of the distributionis Ms 0 dt w= (BQO) = freee] = femciho BA, since [My]. =1. =| i a ol reM,(+ net SMD x] Ate uf | = [herent MLO. IM 5 mg} to dt at de SALEM AQHA EAL IEDa Fundamentals of Probability & Probability Distributions 1 [MAb Te 24 .dM (t) 1dMy(t) wy (4. [perenne IM 4 299! MD oo oe ne PMO ne EM ae une BASEL IAEA) FA (AE M +h Orgy aie 647+). Now)‘one'car find different central moments of ihe distribution fy, the raw moments 1’, :’, Hy’ and 14’. ‘Theorem 1145. IFX is a Poisson variate with parameters 2, then where k; is the rth curnulant of the distribution. Paes We have megf of the distribution from the theorem 11.4, Aas Myp=e 4-9, Hence the cgf of X is . Kx(t) = log My(t) = AC"? Bt! Now, Ket) =a +t+ atartatt: ; e ae Thus, = Coefficient of 4.in Ky(t)= a Hence all cumulants of Poisson distribution are equal to A. Thetis, ki=k=k=ki=. ke=h. Additive property of Poisson variates Theorem 11.4.6. Sum of k independent Poisson variates is also a Poisxn variate. Proof. Let X:, Xz, parameters 2,1 . Xe be k independent Poisson variates wit Thatis, * eas to, A= qh . X= 0, 1, 2s Then; we have to prove that EX; = X_ is also a Poisson variate wih parameter EAj= A. ‘om Discrete Distributions 25 ‘The mgf of Xjis Mx(t) = EL es) Dne=1) for allj Now, the mgf of EX)=X is ' My(t) = Efe * em] = Efe] E [e™] ate, ae! E[e**] exe =1) 2: : alte to etet=0 | ee which is the mgf of a Poisson distribution with parameter = 24;, since gf uniquely determines a distribution. Therefore, the probability function of X is ay foc y= Shere aa da, ana x= fx, This completes the proof of the theorems. Theorem 11.4.7. If X; and. Xa are two independent Poisson variates with parameters Ai and Az, then X = X:—X2 is not a Poisson variate. Proof. By definition, the mgf of X = X1—Xzis Mx(t) = Efe} Ele] Ee] = My, (0). My) a edit“ dete” which cannot be put in the form e*“'~”, Therefore, X = X1 ~ Xz is not a Poisson variate, although X = X: + X2 is a Poisson variate. It is to be noted that X = X: ~ Xz cannot be a Poisson variate because here X can take both positive'and negative values but Poisson variate can take only positive values.”Foor ah eee a i € 426 Fundamentals of Probability & Probability Distributions 11.43. Recurrence Relation of Poisson Distribution Let X be a Poisson variate with parameter 2, then oa fx; 4)=P fe] = which is the probability that X takes the value x. ‘The probability that X takes the value x + 1 is “yee P+ 1]=f(+1, 2) = z (+i) Th Pix+1) | a yt Pale ot th Hence P[x+1]= 4 P[x] +1 (1144 The relation (11.4.6) is known as recurrence relation of Poisson distribution, This relation is very important for graduating Poisson distribution from an observed distribution. First, we have to find the value of P [0] which is equal to €*. If'the value of 2. é not given, it is estimated from the observed data by the method of morhents as” ‘ Lox where X is the mean of the observed distribution. The other probabilities for different values of X can be obiained as follows: = re(24), pr , Pal= (4), Pa nd so on. i 11.44: Fitting a Poisson Distribution to an observed distribution The first step in fitting a theoretical distribution is to estimate the parameter of the distribution A by the method of moments form the observed data. t The mean of the Poisson distribution is 4, while the mean of tht observed distribution is x.Discrete Distributions 47 Hence the estimate of 2 if it is not known is 3 Rs ‘The expected frequencies corresponding to the observed frequencies will be obtained as Ue, x! nf Aen 0,1, 2, -. 147) where n is the total number of observations. Remarks : If it is asked to fit a suitable distribution from an observed distribution, one can varify whether the sample mean and variance are approximately equal or not, since the mean and variance of a Poisson distribution are equal. Example 11.4.3. The following data gives the number of printing mistakes ina book of three hundred pages. Fit a Poisson distribution to the data. Number ofprintng mistakes | 0 | 1 | 2 [ 3 | 4 | 5 | Gandabove ‘Number of pages 130 | 72 [40 [35 | 15 | 6 2 Solution. In order to fit a Poisson distribution to the data, we have to estimate the parameter A from the relation § Sa Eee. 359 2 =f. 3G = 1.197. ‘The probability function of Poisson distribution is eT fx; y= SF x =0,1,2, Here, =1.197, then P[0]=e7 Thatis, log [P(O)] = - 1.197 log,e 1.197(0.43429) 0.519845 = 1.480155. Thus, — P[0] = 0.302099. Now, by using recurrence relation [11.4.4] and relation [11.4.5] we find the probabilities and expected frequencies corresponding to the different values of the variable X which are shown in the table 11.4.1.& i i 2 Fundamentals of Probability & Probability Distributions Table-11.4.1 Number of printing Number of fe B mistakes x-~ pages f £4) iene 0 130 302099 Eyer 1 n 361613 108.48 144 i 2 40 216425 6893p 3 35 ‘086358 B.D 4 6 025841 77565 5 6 ‘006186 18604 6 2 “001244 Pe ‘Since’ frequencies are always integers, therefore: by converting them jo nearest integer, we get [Observed frequency | 130 | 72 | 40 | 35 | 15 | 6] 2 *, [Expected frequency [91 | 108 | 65.[ 26 | 8 [2 fo Itis to be noted here that the stim of the expected frequencies is equal “the sum of the observed frequency.‘ ‘Theorem: 11.4.8. If X is a Poisson variate with parameters A, then the distribution function of X is. F@)= TweaT fretted; x=0,1, Proof. The probability fuinction of a Poisson variate with parameter his fs = x=0,1,2, ss 3 Now, consider the incomplete gamma integral k= J, [fete at positive integer) Catt a Ls [revert ee ae a3 (uss which isa reduction formula for Ix, ae applications ee eat et ~ But p= fetdtDiscrete Distributions tg = PIX=0] + PIX = 1] +PIX=2) +. PIX sx] = F(x). Here F (x) is the distribution function of the random variable X. Hence F (x) "ete de Ten ki fletrat= since I(x+1)=x!. This result is of great practical utility. It enables us to represent the cumulative Poisson probability in terms of incomplete gamma integral, the values of which are tabulated for different values of 4 by Karl Pearson in his table of incomplete I’ - Functions. Mode of the Poisson Distribution e oe i ‘The mode of the distribution is clearly that value of r for which is ‘greater than the term that precedes it and the term that follows it. Thats eat ett ign ei * * wy These gives r SA and r2A-1 ie, (Q-I) =. a Proof: Let X’ be a Poisson variate with parameter 2. Then the mean and variance of X are both 2. The standard Poisson variate Z is defined by X=2 z=X=n ve Moment generating function of Z is M,(t) = Ele] of]452 "Fundamentals of Probability & Probability Distibujog, ea, =e Now proceeding to limit as 2. >», we get . limM,() = ee Tiss the mgf of standard normal variate. By uniqueness p igi, standard Poisson variate tends to a standard normal x, on 4-4 Hence Poison distribution tends to normal distribution ley values of the parameter 1. by Theorem 1.4.12. If is a Poisson variate with parameter 1 ang another discrete variable whose conditional distribution for 4 4° value of X isa binomial istrbution, then the marginal dstuing ¥ is Poisson. Proof. Let X bea Poisson variate with parametr A, then say ‘ 7 PKexd =f yoo, £2012 oncg Suppose the conditional distribution of ¥ for a given value of X= 5, binomial with parameters xand p, then PY=y| X=x]=(*)ptt-p)% O 29) antps small such that 2=np<7, then: binomial distribution a t approximated by a Poisson distribution. Example 1145, Suppose life insurance company insures telnet 5,000 men of aged 42. If,actuarial studies show the probability th ‘A2-year old man will die in a given year to be 0001, find tee probability thatthe company will have to pay 4 claims during given)Discrete Distributions 435 S = ‘The exact probability is given by the binomial distribution as jon ot PIX = 4] = (4; 5000, 0.001) 5000: F9561 (0-001)(0.999)" ch binomial tables are not available. To compute P[X=4] without ny of a computer would be very time-consuming, but the Poisson wa tion can be used to provide a good approximation to P[X=4]. ng 2=np=(5000)(0.001)=5<7 and substituting into the an for the Poisson probability distribution, we have 348 PIX= x1 = 4; 8) = 8) (= (625)(0.006738) _ Sn 0178- ple 11.4.6. A car hire firm has two cars which it hires out day by The number of demands for a car on each day is distributed as 4 Jn variate with mean 1.5. Calculate the probability of days on Miah neither car is used and (ii) some demand is refused. saution. Let X be the number of cars demand per day. It is.a Poisson ‘arale with parameter 2.= 1.5. Then the probability function of X is Ba te ess fer A) = fq 15)= 05" @) PIX=0] = £0, 1.5) =e? = 0.2931 - Gi PIX> 2)=1- PIX <2] =1-[P(X =0) + P(X =1) + PX =2)) visas = 1 - 0.2231x3.625 = 0.19126... Eample 11.4.7. A manufacturer. of power lawn mowers buys 1-horse Power, two-cycle engines in lots of 1000 from a supplier. She then uips each of the mowers produced by her plant with one of the ‘mgines, Past history shows that the probability of any one engine from that supplier providing unsatisfactory is 0.001. In a shipment of 1000 tgines, what is the probability that (i) none is defective? (ii) one is ‘elective? Two are? Three are? Four are?Sere gurad Solution. This isa binomial ex periment with n= Te expected number of defectives in a shipment ae sat p = np=(1000)(001)=1. Since this is 2 bino Bl oe Oe np < 7, the probabiliy of x defective engines in the shi , approximated by the Poisson distribution given by vPement may, PIX @ FiX=0)= Gi PIK=1) = se that the number of telephone calls com ME AM. my. actin Example 11.48. Suppo: a telephone exchange between in disteibution with parameter 2. Similarly ty variable with Poissor e ae'between 11 A.M. and 12 noon éay X, fase number of calls arriving seg istribation with parameter 6% and X,_ are independe, what is the probability that a particular day more than § calls coe: between 10 A.M. and 12 noon? Solution. Let X=%,+% By the additive property of Poisson distribuin X isa Poisson variate with parameter pe2t6-8.) ‘The probability function of x calls in between 10 A.M, and 12 noon o*(8)* pix =x) =: 8)= ae, x=0,1,2,-« more than 5 calls come in between 1AM Hence the probability thal and 12noon is x je1-Pxs5] & Bh vo x! PIK>S.Discrete Distributions ‘37 some examples where Poisson distribution may be successively applied @ The number of cars passing through a certain street in time t. (i) Number of suicides reported in a particular day. (i) Number of faulty blades in a packet of 100. . iv) Number of printing mistakes at each page of a book. (v) Number of air accidents in some unit of time. (v) Number of deaths from a disease such as heart attack or cancer or due to snake bite. , (ii) Number of telephone calls recieved at a particular telephone exchange in some unit of time. i (vii)The number of fragment received by a surface area t from fragmentation bomb. (x) The number of défective materials in a packing manufactured by a _ good concern. 6) The number of letters lost in a mail on a given day in a certain big city. i) The number of fishes caught in a day in a certain city. ii) The number of robbers caught on a given day ina certain city. Example 11.4.9. Suppdse that the number of emergency patients in a _ given day at a’certain hospital is a Poisson variable X with parameter 4 220. What is the probability that in a given day there will be (i) 15 emergency patients, (ii) at least 3 emergency patients and (iii) more that abut less than 25 patients. Solution. (i) Here we havé 4 = 20, x = 15. ‘The random variable X follows Poisson distribution with 2 = 20, therefore (From the table of Poisson distribution) Gi) Plat least 3 patients ] = PIX = 3] = 1-PIxX <2] = PIX = O}- PIX = 1]-PIX=2] ~ ,0000 - .0000 - .0000 = 1. Gil) PRO Q=1-PIXs2] = 1-P[X =0]-PIX = 1]-PIX =2] =1- [2 313] -5.=0323. 5 =0323 Example 114.11. A manufacture of cotter pins knows that 5% of hs | product is defective. If he sells cotter pins in boxes of 100 and guarantes: that ‘not more, than 10 pins will be defective, what is the approxinae probability that a box will fail to meet the guaranteed quality Solution. We are given n = 100, Let p be the probability of defective pin = 5% = .05. <. A= mean number of defective pins in a box of 100 = np = 100% 05=5 Since p is small, we may use Poisson distribution. Probability of x defective pins in a box of 100 is eM Sor, PIX=x)4 LGN 2012, Probability that a box va fail to meet the guaranteed quality is “p[x> 10} =1-P[Xs 10] =1= per “Gy 1 nets. Example 1.4.12. If X is a Poisson variate such that PIX = 2] = 9 PIX = 4] +90 P[X = 6] Find (i) A, the mean of X, (ii) B1 , the measure of skewness. el)Discrete Distributions ep ee ation. If X is a Poisson variate with parameter 2, then 90! P[IX=x= 2%, 5 h>0. x! weare given, PIX = 2] =9 P[X = 4] + 90 P[X =6] aya 6 oes. SFr +90. $7 [94 90n8]_ af nat ae oem) oats 7 § SE [an ea] > 34M 4 oy A +30?-4 20. solving as a quadratic in 47, we get qin r3tV9FI6 945 2 2 since 4>0, we get 27=1, wheal Hence Mean = A = 1, variance = A=1 and jis=1. Bee pel. Example 11.4.13. If X and Y are independent Poisson variates, show that the conditional distribution of X given X+Y is binomial. Solution. Let X and Y be two independent Poisson variates ‘with Parameters 1 and p, respectively. ‘Then X'+ Y is also a Poisson'Vatiate with parameter (A. + 1). PIX enpe PXsraXx+ Yen). Pix=roy=n=y Keath X+¥ =n] PIX+Y=n} Pix+¥=n) = P&X=rP[Y=n-y P[X+Y=n] [since X and Y are independent]FW =(‘)p'a-p" Hence the conditional distribution of X given X.+ Y = distribution with Parameters n and p = ~~. 15. binom peor ‘ ° 11.5. Hypergeometric distribution 11.5.1. Introduction. If a sample is drawn without replacement from g finite population which contains two types of things, we ge hypergeometric distribution. eRt Con Suppose an um contains N number of balls among which M balls are white. If a random sample of size n is drawn from this urn without replacement, then the probability of getting k (k < n) white balls is ; rout { OCS2/(9) k=0,1,2,. a, ae otherwise Definition. 11.5.1. A discrete random variable X is said to have * hypergeometric distribution if its probability function is defined a5 f(:NMn)= 4 (") ley) f (")° x 0, 1,2, “qi 0; otherwise, , @ ; is at Om where N is a positive integer, M.is,a.non-negative integer that ar tion N and n isa positive integer that is at most M. Any distribution defined by the probability function given in equation (11.50) hypergeometric distribution. x called= | os Discrete Distributions i pesivation of Hypergeometric distribution o an urn contains N balls among which M balls are white and iisare black. A sample of n balls are drawn at random from the ithout replacement. Let X be the number of white balls in n balls, a take values 0, 1, 2, n: Now, n balls can be drawn from ls in (3) ways and x white balls and (n - x) blacks balls can be \ n ‘ e = vf N- nx sn from M white and N~M blacks balls in (“)( ) ways. Then, i ability of x white balls among the selected n balls is fig N/Myn) = MELV/O): X=0,1,2, isis the probability function of the hypergeometric distribution. The jane of the distribution comes from the fact that the distribution can be fined from the hypergeometric function, N, M.and n are the three paneters of the distribution. A. iseasy 10 varify that : @) f% N,M,n) 2 0;, (ii) & fx; N,M,n)=1- Nehave.f; N, Mn) = (M) rem /() ( Fete iil “E()(2) = (722), we get Eioonaine & (MYM \(N) mex CC) Theorem 11.5.1. If X is a hypergeometric variate with parameters N, M, ‘ndn, then “ = npg =442 Fundamentals of Probability & Probability Distributions a oF Proof. By definition, not $0) )/(n) iy M! (N-M)! ni(N-n)! <7 I (Rox(N-M-n4xy aM & (M-DMN-M)! 2 (@=J)(N-n)" N 3 &-1M—x)! “x KN-M—n xo RM & (M-)(NEM) ,(N-T =P ECV): Let y=x-1, then y=0,whenx=1 and y=n~1,whenx=n, n-l-y = A (ME) (9) + By using the result, & (M)(N.) = (M*") Now, EKX-D]= Ex 6- oye) = HOD 8 (WNC Me ") Let us put x-2=y, then FIX) = n(n=1)M(M~1) ety syse0tey, (2) (N=1 = Mac IMIM=1) (NY (NY. 4 2B MMED, grocers ‘Thus, y var [X] = EPC] - [EQ]? =EX(X-1) + X]-[EQ)]? =EIX(X-1) +E XI - B ool n(n~1)M(M~1) nM? 2 oNDiscrete Distributions ete as nse nN+N?—N-nMNsnM4N \ gt PE NOE BN] - i aan aoa NET aoe maa 1-2 a1- P- ah ‘ peorem 1252.18 X isa hypergeometric variate with probability function soa NM) =(“Y(NMI/(N); x=01,2,. n, on? van ae where j1'1, is the rth factorial moment of X rook, 5 Har ERC] xf0(M) (N- Ne N’ kOe): i M! Cy ni(N=n)! H nx) NT (2) nn-r)(N-r-(n-9)}! =T- HE A nx Nea! Ua(x-1) =}, then j=0 when x =r and j=n-rwhenx=n. t= Mons (M=n! (NA (M-1)) Mae “Ne & faa ( ey (Ne) ME ECORI) SE mae Hence, yt, — M®.n® (115.2) nN444. Fundamentals of Probability & Probability Distibuting We can exsily find the different factorial moments from 4 (11.52). Thus. he ey we M(M~1)n(n-1) v= %, was —NN=1 Mo ip es and M(M~1)(M=2)M~ 3)n(n~1(n~ 20 Ko" NN= N= 210-3) : (On simplification, we find the third and fourth central mo distribution as « + 2 BPN =n)(N=2n) b= =a an npg (N=9) nee 1y-6nav-. aaa Me A ayn-aN= ay NN U-6n(N-n)+ 3pq (0-2) — nn Gv men ay Remarks: (i) Hypergeometric distribution is symmetrical if p= 3, is patng FAVA skewed if p<}and is negatively skewed if p>. (il) If we sample n balls from the N balls among which M ate g with replacement and denote, Y the number of white balling sample, then Y is a binomial variate with parameters n and Pt Thus HiYJ=np=nM=e[X]and 3 var(YJ=n¥t NM) =npq > vac IX] ‘equality in variance hold only when n= 1. (ii) The mean of the’ hypergeometric’ distribution_concides wit! mean of the binomial distribution if 44 = p, but the varianolé binomial distribution is greater than the variance of hypergeometric distribution unless n = 1.Discrete Distributions 445 = The figures 11.5.1. gives the probability. function of hypergeometric distribution for different values of the parameters N, Mand n. N=10;M=4;n=4 N=10;K=4;n=5 7 Seed oft, 2 3 4 o 12 3 45. Fig. 11.5.1. Hypergeometric Probability Functions. i) ‘Theorem 11.5.3. Let X be a hypergeometric variate with parameters N, Mand n if N tends to infinity then hypergeometric distribution reduces tobinomial distribution. Proof. The probability function of hypergeometric. distribution with parameters N, M and nis foc N/M, n) = Ge “y(): x=0,12 ty 'S) when N tends to infinity, then M= p and N. ‘ ; T ptqek Thus fox NM, n) = ed eG) a3 M — 1 ~p-=q such that (N-M)! Moa RNR 09 Now, f(x; N,/M,n) = ah: “ore (N= My {N-M)- (n= )}!(N-n)! IN=M=(@—x)}!NO(N—n)t i since n! nn — -x)! ’ - =”) M(M=)).....(M —x + 1).(N~ M)(N-M ~1). M-=n¢4: en: iH “EO NAADN-2) NS46 Fundamentals of Probability & Probability Distributions since M = Np and N~¥4 =Ny Now, Jim fs; N,M, n)=(Q)p'a'* = F082 9, P) X=0-1,2) mmo, ‘This is the probability function of binomial distribution with parameieg nandp. = 11.5.2. Comparison of Binomial Distribution ‘with Hypergeomeni, Distribution 4 : Binomial Distribution Hypergeometric Distribution i) Inbinomial distribution, the) i). In hypergoemetric distribution probability of success p remains _ probability of success varies the same from trial to trial. from trial to trial. ff) The sampling is done with ii) The sampling is done with replacement. out replacement. ifi) The distribution is hypothetical. iii) It is.not hypothetical. It is clear from the above comparison, if the sampling is done with replacement, then hypergeometric distribution turns into binomial distribution, or if the populations size’ from which the sample drawnis infinite, then the hypergeometric distribution turns into binomial distribution. Example 11.5.1. A bag contains 5 white and 5 red balls. 4 balls are dravin from this bag without replacement (i) What is the probability that te selected balls are 2 white and 2 red, (ii) Also find the probability that the selected balls contains at least one white ball. Solution. Here we have N = 10,M =5andn=4. Let X be the number of white balls, then x follows hypergeometric distribution with parameters, N=10, M=5 and n=4 and the probability function of X is f(x; 10, 5,4) = (2): forx = 0,1,2,3,4.Discrete Distributions 47 robability that the selected balls will contain 2 white balls is am? 5) (3) (10) _ 10 ed 10,5,4)= ()GV(C?)= x ropability that. the selected balls will contains at least one i THeball is “ox? yet -PIX=01=1~£0; 10,5,4) 5\ (3 0) -1-1.4 21-()QVC2) =1-ae ation Hypergeometric distribution is used to. estimate the wee offishes in a lake: ga ake contains N fishes which is unknown. We have to Soph his number. First a catch of m fishes is made at a time and nt them alive immediately after making with red spot. After a shut period, during which these marked fish are assumed to have nab Femselves at random in the Jake. Another catch of n fish is se at a time. Here M and n are regarded as fixed pede ened constants. Among this'n fish caught there will be say X ee yrere X is a random variable following discrete Pw fishy vr om given by hypergeometric model. ted poy N-M) ,(N fo; N/ Min) = C nex i) = f(N) say. aenvis estimated by the principle of maximum likihond. That the value a & . Had te value A =51 (0) of which maximises f(). Sine Nisa random variable, the principle of maxima and minima in cS Tapnot be used here. Here, we proceed as follows : ie gS), _(N=MyN-n) 2)= GED ~ NIN=M=n4) ° NM estore, MND=1 Ne leo rats AN)? ND ifN> At (1133) ane MB or MN) <1 if N< (1153) ig Mn matis «)< AN-D ta= } 48 Fundamentals of Probability & Probability Distributions From (11.5.3) and (11.5.4) we see that f(N)=f(x|N) reaches the maximum value when N is APPrOtnag equal to ME. Here the maximunit likelihood estimate of Nis given hte, x Thatis Noy= ME, ‘Theorem 11.544, Let X be a binomial variate with parameters n, ad, and Y be another binomial variate with parameters. ny and pI iq Y are independent, then the conditional distribution of X, given Xe, isa hypergeometric distribution. " Proof. Here X ~ B(n, p) and Y ~ B(ny, p) are two independent binons, variates with common probability of success p. Then X+Y=Z is ako, binomial variate with parameters ny+n,=n andp. : By definition, © PiX=rax+Y=2) PIX+Y=z] _PiX=r1P¥=2-11 5 ; DXF Yat since X and Y are independ - Ce pre 0%, yr per FE ro (72) pa-pree? “OCP rea which is the probability function of a hypergeometric distribution, PIX | XY =z] Therefore, P[IX=r | X+¥Discrete Distributions is 11.6, Negative hypergeometri¢ distribution Anurn contains N°” balls among which M_ balls are white and N-M balls are black. A sample of -n balls are drawn at random from the urn without- replacement. Let X denote the member of white balls. We saw that X has a hypergeometric distribution with probability mass function (11.5.1). If, on the other hand, the sample is drawn with replacement, then X isa binomial variate X ~ B(n, p) where P=M/N. Let X denote the number of draws needed to draw the rth white ball. If the draws are made with replacement, then X has the negative binomial distribution given in (11.7.1) with p = M/N. What happens if the draws are made without replacement? In that case in order that kth draw (k2r) be the rth white ball drawn, the kth draw must produce a white ball, where as the previous k-l draws must produce (r - 1) white balls. It follows that ? M\(N-M ees for k=r,e+1,. PIX=K)= Rewriting we see that roca ()4/(8) ose Arandom variable X with probability mass function (11.5.5) is said to have a negative hypergeometric distribution. Itis easy to see that } E[X] = fe BRO = MDS a _ H(N-My(N+1D(M+1-1) wa) (M+1)(M+2) Also, if. :/N 0 and k/N+0 as N-ve, then rt) Mdm) Met RC. (eve) = This.is the pmf of negative binomial distribution defined in (11.7.1).450 Fundamentals of Probability & Probability Distributions 11.7. Multivariate Hypergeometric Distribution Consider an um containing N items divided into K catogoy containing n,, ny, .. sample, without replacement, of size n is taken from the‘um. Let number of.items in sample of type i. Then Xe xed = T (eaue) (sy min (n, n) and Eayen i ‘ k 4 Ty, item, respectively, where z TW=N. A tan, cn Xs PIX: = Xp Xp = Xp where x)=0,1, We say that (Xj, X; X,.2) has ‘multi-vatiate hypergeomets distribution if its joint pmf is given by (11.5.6). It is obvious that each has a marginal hypergeometric distribution. Moreover, the condition distributions are also hypergeometric. ‘ (nas Thus, PIX, =x, | X=) = and so on. It is therefore easy to write down the margirtal and conditional means and variances. The means, variances‘and co-variances are EDX) = aq; var [X]=n. and cov [X, X= Application. Hypergeometric distribution frequently replaces a binomial distribution when it is desirable to make allowance for finiteness of sample size. An interesting application of this distribution is in the estimation of the size of animal populations from capture-recapture data. The distribution also finds its application to linguistic problems. A further application of the hypergeometric distribution is in a model for the number of children attached by an infections disease, when fixed number (N) are exposed to it.Discrete Di: 451 1.8. Geometric distribution Definition. 11.6.1. A discrete random variable X is said to have a geometric distribution if its probability function is defined as fOy p)=pq; x=0,1,2,. where p is the only parameter such that 0SpS1 and p+q=1 “ (1181) ‘The probability for the different values of the random variable X are the yarious terms of the geometric progression, hence the name geometric distribution. 11.8.1. Derivation of Geometric Distribution suppose, we have a series of independent Bernoulli trials with constant probability, of success p such that p + q = 1. If (x +1) trials are required to get the first success, then X can take the values 0, 1, 2, ..., with corresponding probabilities p, pq, pq’ Hence the probability of first success in (x + 1) trials is f(x; p) =pq°s X50, 1,2, 00 This is the probability function of geometric distribution. Actually pq" is the probability that there are x failures proceeding the first success. Itis easy to varify that () £6 p) 20 (ii) E tox: p)=pd+q+q+.. P= 1-4 The probability functions for different values of p of geometric distribution are given in figure 11.8.1. & Itis clear from the figure (11.8.1) that the mode of geometric distribution isat X=0.452 Fundamentals of Probability & Probability Distributions 1 > : ’ | tJ _¢ _¢ 9 4 Oe de Se es Og I 1 1 = pet 4 , 4 ' I J__¢ 9 9 2 « ,, “8 0 SERS 3 eh SE Fig. 11.8.1. Geometric probability functions. Theorem 11.8.1. If X is a geometric variate with parameter p, then “ Mx) = 2, 1-¢e' 2 * 2 ERI=4, w=, yj=242 and w= at +d Pp P Pig where Mx(t) is the mgf of X: Proof. By definition, i M,(t) = Ele" = be pq =pz (ae ney =p(1-¢e'y" ; =plinafisrefietetts = [rnd {eg tet Hence.the cumulant generating function of X is Kt) =~tog [= Ser E ret # ~)Discrete Distributions 453, * ef ie eB “asi of ae tat os kis the coefficient off in K,(t). Hence ; “Coefficient of + 7 ink) = . = mean - pecoetlcient of £ ink0= de nb Spe be at y,, yconficient OF $7 Pin Kit) = 438 oe 6g! coefficient of 5 t ink(t) = a3 - Sf ES 5 P q+ Sp’g? + 4q'p" +12q"p + 6g" -4 ip +)" +4qp* +9pq? +54") = eae a 9 +9pq" + 5q'] “2 [+ 4pq (p + 4) 459° (Pp +.9)] = 4 [i+ 4pq +5q'I= [+ 4g +4). 3, Hence, y= ky + Sky? = giltaq+ a?) , 24° a+ 7q+q') P P P and f, measure the skewness and kurtosis of a distribution : 2 Hince f= 4. a Hse asa. (+g? nen 2 q B= Hee asZg4a).. A47q+q) a q q . Remarks : The mean of geometric distribution is smaller than its "atiance, since > Pp P454 i ‘undamentals of Probability & Probability Distributions —_—_S seem — Theorem 11.8.2. If X isa geometric vi with probability func \etric variate with pi Ly fancticg, foc p) = pq: X=0,1, 2, nner The 2p 3 iy en x(t) ae) pile)", W(t) =p —4ey" and Pls)=p(1-4s)" where x(t), W(t) and P [s] are the cf, fmgt and pgf of X respective Proof. By definition, . ’ t) = Fle] = Fe pqr=p § , ox(t) = Fle] =e pa‘ =p & (gel) = pal - 4 i By definition of X w(t) = El T= E pa" =p E (ta)" () = Ef = & tpa'=p & (tay =p(l-qty’. By definition, paf of Xis Pls} = Els") = 3,5*Pa'= Ey = 72 = ptt-sa 11.8.2. The property of lack memory of Geometric Distribution Geometric distribution is sometimes known as the lack memory Suppose an event A can occur in any time t= 0,1,2— certain sense. Then the occurrence waiting time X has a ime t= 0, 1 probability function ne geometric distribution} fx; p)= PT? x=0,1,2, f If it is known that the occurence of the event A has ha| | ; i 7 7 Tet atone ie XK Lat Y= Kk, Thus ¥ willbe SP time needed for A to occur. It can be shown that P[Y=t|X2kl=pq' he additional time to wait has the same distri? wt which shows that the - ows Eo wait. In this sense geometric distribution s Wi as initial S the lack of memory: We can cite the above statement as a theorem Theorem 41.8.3. Let X bea geometric variate! with’ probability fund] defined a5 fag 1 fog p)= Pas 20,1, 2p sry 2: 1p Y=X-* where k isan integer, then pret | X2kK=piX=t.< Discrete Distributions PIX2r]= F pokapeata aeet T= 2 pqt=pqitgit+ A Now, PIYet | Xakj= PI¥2toxek}_ PRC Ke tOXeH PIX2k) PIXzh : = Pkkek+q_ gg PXSH ge 4 Hence P[Y=t | X2k]=P[Yet | X2k]-PIX2t41.| X2K] | =q'q? =qp = PIX=#). ‘This completes the proof of the theorem? ‘Theorem 11.8.4. Let X, and X, bé two independent geometric distribution with the same parameter p, then the conditional distribution of X, for given value of. X,+.X2 =n is uniform: Proof. Weare given PIX, = k] = P[X, =k] = pq‘ k=0,1,2, By definition, the conditional distribution.of X, = r_ for given value of XtX=n is ? PIX, = X,+X2 =, PIX [XyiXye n= PRG Se okie kan) : Pern X=en-1) = PIR +X: =n) _ PR =r X,=n-4) r Spi =snx%=n-s)-. = 7 Etre, =s) Po Since X, and Xz are independent. X= r1X + Xp=n] = PLPe Therefore, PIM = 11% +X Ete pg"456 Fundamentals of Probability & Probability Distributions aah eRe zp'd" & 26 = pe Biase G@rppg Bet THOL2. Hence the conditional distribution of X; for given value of Xi#X,<5 discrete uniform distribution. ; Application. The geometric distribution is used in Motkov chitin mo4s particularly meteorological modes of weather cycles and precisiays amounts. The geometric distribution is often referred to as the fui. ume distribution, The geometric distribution would be used to dean the n. mber of interviews that have to be conducted by a selection boar to appoint the first acceptable candidate. Example 11.8.1. A couple decides that they will take children until ao, Suppose the ptobability of a son’in that community ¥ 1/3. ()) What the probability that the fourth child is a'son, (ii) Also find the expect number of children to get a son. Solution. Let X be a random variable which is defined by the numbero, female children required to get a son. Then the probability function of (H-QVG cone where 2 is the probability of getting a female children in that commuriy, (i) Probability that the fourth child will be a son v 1) _(1)/2)_ 8 Pix=3)=¢(3 3)-G)G) are (ii) The expected ‘number of female children ‘to get a first son is te mean of the distribution which is..q/p = 453 =2. Hence, the numiber of children will be (2 + 1) = 3 to get a son. Remarks : The random variable X defined'in geometric distributionis the number of Bernoulli trials required before the first success. Moreover, a random variable X has a geometric’ distribution is ofte referred to as a discrete waiting time random variable. It represents hov long one has to wait for a success. [eine Se epee ee eeDiscrete Distributions 457 eg ee ag ae ne ee We note that some authors defined the geometric distribution by jssuming 1 instead of 0 as the smallest mass point. The probability junction then has the form f(x p)=pll-pt; x=0,1,2, (1182) and the mean is 3 the variance is 4 and the mgt is B(1-ge)". 419. Negative binomial distribution Introduction. The important characteristic of Poisson distribution is that the mean and variance are equal. Whereas mean is greater than variance for binomial distribution. But in practice, there are situations, where the mean is smaller than the variance. In such cases, the model of negative binomial is more appropriate. For example, death of insects, number of insects bites lead to the negative binomial distribution. Negative binomial distribution is sometimes known as Pascal distribution after the name of French mathematician Blaise Pascal (1623-1663). The negative binomial distribution can be derived:in many-ways. We shall derive it here from Bernoulli trials. First, we shall define the distribution. [Definition 11.9.1. A discrete random variable’ X is said to have a negative binomial distribution if its probability function is given by * (xttl) or x. ra can { CA) 06 xe OBR o otherwise “ a1g9a) lwhere r> 0 and 0 Sp s 1 are the parameters of the distribution such that lp + q = 1. Actually, f(x; r, p) is the probability of getting exactly r successes in x +r independent Bernoulli trials. 11.9.1. Derivation of Negative Binomial Distribution Suppose we have a succession of n Bernoullittrials. We assume that the () trials are independent and (i) probability of success p in a trial remains the same from trial to trial. totes : Let f(x; 1, p) denote the probability that there are x failures preceding the mth success in x + r trials. .458 Fundamentals of Probability & Probability Distributions Now, the last trial must be a success, whose Probability js remaining (e + r—1) trials we must have (¢~ 1) successes ang whose probability is given by lung : fctrel) eta (a )e ae : Therefore by compound probability theorem, f(x ; rp) is given product of these two probabilities. That is, by fx; ne p)= (45) Pp os (257) pq: 50/1; 2)02.5, 0, Moreover : (cn!) = tr=1)! : (Dit rel = *Cr- DEr-2). =o). Hence {(x; 1, p) = (pre: x20,1,2, (=r =x +2)(=r—-x41) (usa (11.92) is the (x + 1)th term in the expansion of p*(1 - q) a binonial ‘expansion with a negative index. Hence the distribution is knowns, ‘Régative binomial distribution. ! Tt is easy to varify that (i) f(x; x, p) 2 0 and (ii) E.A(x; r, p) = 1. Since, Eft; 1 p)= E(Z)pca ‘ ep k(t on =F El) =p'(l-q)'=1. 5 That is f(x; x, p) isa probability function. Remarks: @ lfr=1, in (11.9.1), we get geometric distribution with parametesp In this sense, negative binomial distribution is considered #8? generalisation of geometric distribution.A *% Discrete Distnbutions 9 {i The negative binomial distribution of the form (11.9.2) is also known as Pascal’s distribution with parameters r and p. ‘Theorem 11.9.1. If X is a discrete random variable with probability function for ep) = (257) pats x=0,1.2, then mgf of X is Ma0= (qa) and EDI = FL, var bd = pe ota, 1+ a va iq? 2 6._9+ ee and ne Per, Proof. By definition, Mx(t) = Ele"] = ow (tr x . = Ee) ra = PEC) er -p Ecar(Z) er Now, ‘ve shall find the moments, B, and f, with the help of cumulants. Bydefin tion, cgf of X is K() =r log rl ws rlog[2+4=a¢"] . ao -Det- = log] eee! »] ° of fee =-rlog{ See Sates ] Sat see faa. besetSE 460 Fundamentals of Probability & Probability Distributions Bot. pe tt o Sfp} ee atta ds = ff heal at 3a F| B [eEfes§ ae + + (gy Thus, ki = Coefficient of tt in K,(t) = a =E[X]. ka Coefficient of 4 in Kx(t) = © afi +a]=S | > p-i-¥: Hence, Jum (1- J “6 2) (G +2) ‘Discrete Distributions a 1 x=0,1,2,.. which is the probability function of a Poisson distribution with jarameter A. Hence the proof. 4 | some characteristics of Negative Binomial Distribution 4. If = 1, negative binomial distribution reduces to geometric distribution with parameters p. 2. Ihr, q > 0, and rq — A, then negative binomial distribution tums into Poisson distribution. ty | 3. The variance of this distribution is always greater than mean. |. The distribution is positively skewed and leptokurtic. 5. Under certain conditions, negative binomial distribution follows additive property. That is, if Xi, Xz) +, Xk are k independent negative binomial variates with parameters (r, p), (rz P) + Ce PD. Ir then X= Xi + Xa + sue +X. is also a negative binomual variate with parameters Coniparison of Binomial Distribution with negative Binomial ds:cibution () Both the distributions are constructed on the basis of Bernoulli trials with the constant probability of success p from tnal to trial. ()) Bin mual distribu'ion is the general term of the binormal expansion (p +.q)" where a= negative binomial distribution is the general term 9 the binomial e pansion with negative index, ie., p"(1 - q)*. (i) Mean is always greater than variance for b:nomial distribution, where as varance is always greater than mean for negative binomial distribution. 0) The number of trials ‘n’ in + nomial distribution is finite and « prédetermined, where. 5 the n: mber of trials in negative binomial distribution. depends on the n- mber' of successes r which is fixed nd it may be infinite,