12 Theoretical Distributions Binomial, Poisson, Normal 12.1 Random Variable and Probability Distribution Random Variable has been defined as a ‘function’ which assumes Teal values on the outcomes of a random experiment (See Section 11.19), In repeated performances of the random experiment, the random variable will in general assume different values with a definite Probability associated with cach value or interval of values Probability Distribution (or simply Distribution) of a random Variable is a statement specifying the sot of its possible values together with their respective probabilities, When a random experiment is theoretically assumed to serve as a model, the probabilities can be given as a function of the random variable, ‘The probability clistri- bution concerned is then generally known as theoretical distribution, We shall discuss here mainly four important theoretical distributions, viz. Binomial, Poisson, Hypergeometric and Normal, 12.2 Discrete Probability Distribution Let a discrete random variable Y assume the values x, %y. 0. %» with probabilities p,, Py. +... Px respectively, where Xp, = 1. The specification of the set of values x, together with their probabilities Pp, (= 1, 2, .... m) defines the probability distribution of the discrete random variable X, or in short, discrete probability distribution of X. Tas_E 12.1—Discrete Probability Distribution Values (x) Probabilities (p) | Ps Often, the probability (p) that the random variable X assumes a specified value (x) can be expressed in terms of a general mathematical expression f(x). f(x) = Probability that X assumes the value x = P(X =x) (12.2.1) x) is known as “probability mass function” (p. function’ of the ean ee varie & eSBINOMIAL POISSON NORMAL 93 If f(x) is known, it is to write the probabilities p,, P,, because p, may be obtained from f(x) on Putting pee S f(x). The discrete probability distribution is then defined simply Pp by stating the mathematical ex ion for fix), a possible values X,, Xyp ses Xe ao f(x), along with the set of -TaBLE 12.2—Theoretical Distribution (Discrete) _ The discrete random variable X may also assume countably in- finite number of possible values x,, x, ..., %» -.. with p.m. f(2). The probability mass function f(x) must satisfy the two conditions @ fe)>0, @ Zf@)=1 (12.2.2) where the summation is taken over all possible values x. Iustration 1. The following table shows the discrete. probability distribution of a random variable X: Value (x) 3 6 a u Total Probability (p) G2) tems 9s omy i. Tilustration 2. If X denotes the ‘number of points obtained when an unbiased die is thrown’, the probability distribution of X is : No. of points (x) fig T egw ps eG eeroed Probability (p) 1/6 116 16 1/6 1/6 1/6 1 This distribution may also be written as fx) = 1/6; (* = 1,2, ..., ) Illustration 3. If X denotes the ‘number of heads obtained in 3 tosses of an unbiased coin’, the probability distribution of X is: No. of heads (x) 0 1 2 2 Total Probability ‘p) 1/8 3/8 3/8 Ws 1 These probabilities may also be given by the p.m.f. (Example 12 : 8) Se) = *C2 G5 (X= 0, 1, 2, 3) Illustration 4. An unbiased coin is thrown lly until a head appears. If X denotes the ‘number of tails the first head” then its probability distribution is given below : No. of tails (x) 0 1 2 ac r eo Total Probability (p) + ()* @)? GS GT 1 This distributio w by the p.n ?94 STATISTICAL METHODS (Note: (i) A discrete random variable (f.v.) can assume either a ‘fnjtgs number of possible values (sce Illustrations 1, 2, 3) or a ‘countably infiniter number of values (Ilu: it finite probability (ie, a gon. negative real number) a: inct value of the variable, 4 that the total probability The ‘probability distribution’ of a discrete -y shows how the total probability 1 is distributed over the different POSSible values of the variable. The probability of an interval of values is then given, by the sum of the probabilities associated with those values which lie in that interval. Tt will be shown later that a continuous random variable assumes an ‘yn. countably infinite’ number of possible values. Probabilities are now associated with intervals of values, but not with each individual value of the variable, The Probability is obtained by integrating the ‘probability density function’ (paty Over the limits of the interval. (i) Generally, the capital letter X is used to denote the ‘random variable and the small letter x to denote ‘any specified value’ of the random variable, Af however no confusion aris.:, the small letter x is often used for both] Example 12:1 YX isa discrete random variate having probability mass function : xe 0 1 2 3 4 5 6 7 P(X=x): 0 k 2k 2k 3k k® = 2k® Tk*+k (i) Determine the constant k ; (ii) Find P(X < 6); (iii) What will be P(X > 6)? [W.B.HS. ’81) Solution : (i) For a discrete random variate X, the probability mass function f(x) = P(X = x) must satisfy the relations M fx) >0, UD sfx) =1 where the summation is taken over all possible values x. Using (II) here we have O+k + 2k + 2k + 3k + k* + 22 + (7kK*4k) = 1 or, 10k? + 9k = 1 or, 10k? + 9k -1=0 or, (kK+1) (0kK—1)=0; 2. k= 1, 1/10. virtue of (1), k = —1 is impossible (Note that th ybabilitic ioe to I, 2, 3, 4 become negative). Hence k o 0 ne (i) P(X <6) = KX =0)+ P(X = 1) 4... 4 X= 5) =OFK+ 242k 43k 4-42 = 8k + k* = 8.0.1) + (0.1)? = 0.81 (iii) P(X > 6) =1— P(X < 6), since the events ¥ <6 and X > 6 are complementary. =1-081 =0.19 Ans, 0.1, 0.81, 0.19BINOMIAL POISSON NORMAL 95 Expectation—Mean, Variance, Moments (discrete distribution) | Let a discrete random variable x assume the values x,, Xp)...» *n wi robabilities Pye Pz» +++» Pa Tespectively. Then the expectation, or C2 ted value’ of x—written E(x)—is defined as the sum of pro- quots of the different values of x and the corresponding probabilities. E(x) = 2pm (12.3.1) The expected value of x? is similarly defined as the sum of products of the squares of values and the corresponding probabilities. Ee?) = Spx? (12.3.2) In general, the expected value of any function g(x) is defined as E[a(x)] = pax) Hence, the expected value of a constant k is the constant k itself. E() =k, where k is a constant; because E(k) = Zpik = kip = k. If the p.m. f(x) is given, then the expectations are defined as follows : E@) © Exf() (12.3.3) E(x?) = 2x7f(@) (12.3.4) Mean of a probability distribution is the expected value of x. Mean (u) = E(x) (12.3.5) Variance is the expected value of (x—~)*, where « is the mean. Variance (0?) = E(x—p)* It may be shown that o? = E(x*)—p? (12.3.6) The positive square-root of variance gives the standard deviation (¢). Moments of a discrete distribution are defined as follows : r-th moment about A: uy’ = E(x—A)" = 3(x—A)” f(x) r-th raw moment : My = E(x’) = =x". f(x) (12.3.7) r-thcentral moment: py, = E(x—x)* = E(x—#)". f(x) where 4 = E(x) is the mean of the distribution. Note that Ho’ = Ho = 1: a = Ex) = 4 pa = E(x—p) = 2(x—p): fle) = Ex f(x)—HE f(x) =p-u=0 The central moments can be obtained from non-central moments using the relations (see 7.2.2) Bs = Hs’ — os® My = Me’ — 3ha' Hs’ + ms” (12.3.8) Mg = Hg! — its’ us! + 6uty'Hs? — 3,"*6 STATISTICAL METHODS Example 12:2 A random variable has the following Probability distribution : ie 4 5 6 8 Probability 0.1 0.3 04 0. 2 Find the expectation and the standard deviation of the random variable. ICU, B. Cm. (Hons) *80) Solution : (Note that the sum of the probabilities is 1 ; sp, = 1). Mean (x) = E(x) = =pyxq = (0.1x4) + (0.3x5) + (0.4x6) + (0.2x8) = 04415424416 =59 Bix) = Epyxe? = (0.1x4*) + (0.35%) + (04x64) + (0.28%) = 16 +75 + 1444 128 = Variance (o*) = E(x*) — 3, using (12.3.6) 36.3 = 36.3 — (5.9)® = 36.3 — 34.81 = 1.49 S. D. (¢) = J149 = 1.22 Ans. 5.9, 1.22 Example 12:3 For what value of a will the function f(x) = ax; x= 1,2, 3, ine be the probability mass function of a discrete random variable x ? Find the mean and variance of x. (W.B.H.S. °78, °82] Solution: The conditions for any function f(x) to be a p.m.f. are @ fix) >, Gi) =f) = 1 From (i), we have a>0, because none of the values of x is negative. From (ii), 1 = sf(x) = Sax = asx = atl) iad maa Mean=E(x) = =x f(x) = 3 xax = 2% n(n+1) ORS radease tie Sp eae = Again, Kat) = xx* f(x) =a Be = a(I*+2°4..n8) = men s. Variance = E(x*)—{E(x))* = for sai —(att)* ny a (n=1) (n+2) ee } ! 7 |BINOMIAL POISSON NORMAL Example 12:4 A random variable ¥ is defined as follows : Prob(X=1)=p, Prob (X = 0) = 1- i , =0)<1-p here O 0; because none of p and g is negative (i) E Ne) = va? + Pa + Pg? + Dat t+ om ow98 STATISTICAL METHODS The expectation of the number of tails preceding the first head ig EX) = Ex.f(x) =: xg = 0+ pa + 2pq* + 3pq* +». = pal 4- 29 + 3g? +o. oe ) = pg(l—9)-* = pg.p-* = 9p™* = alp. Ans. ap 12.4 Uniform Distribution (discrete) If a discrete random variable x assumes n possible yah Xi. Xe sey Xa with equal probabilities, then the probability that jt takes any particular value is a constant 1/m, The probability distri. bution defined by the probability mass function (p.m.f.) TG@y=ayns (x ='X,, X, .... Xe) (12.4.1) is known as Uniform Distribution. TABLE 12.3—Uniform Distribution (Discrete) The number of points obtained in a single throw of an unbiased die follows uniform distribution with p.mf. f(x) = 1/6, (« = 1, 2,3, No. of points x : 1 Dia Benin an SI 6 Total Probability f(x): 1/6 1/6 1/6 1/6 1/6 1/6 1 (Note: () The pm. of a discrete uniform distribution is a constant (independent of x), and this constant is one divided by the number of possible values of r. (i) In a discrete uniform distribution, the possible values of the random variable may be any real numbers, and the successive values may not have a common difference. However, there must be a finite number of possible values. Example 12:6 If a discrete random variable x follows uniform distribution and assumes only the values 8, 9, 11, 15, 18, 20, find the | probabilities (i) P(xr=9), (ii) Pix=12), (iii) PE15), (vi) P(|x—14 | <5), Solution: Since x has a discrete uniform distribution with 6 pos” sible values, the probability that it takes any particular value is ® constant ¢ and the probability that it takes a value other thanBINOMIAL POISSON NORMAL 9 given, is zero. Therefore, @) P(x = 9) = 1/6, Gi) P& = 12) =0 (ii) Now, x < 15 implies that x takes only 3 values, viz. 8, 9, U1. Hence, P(x < 15) = 3x} = 1/2 (iv) Again, x < 15 implies that the variable takes 4 values (15 included), viz. 8,9, 11, 15. Hence Px < 15) = 4x¥ = 2/3 (v) > 15 implies only 2 possible values of x viz. 18 and 20. P(x > 15) = 2x} = 1/3 (vi) | x—14 | <5 implies that the difference of x and 14 is to be less than 5 ; so x takes only the values 11, 15 and 18. Hence, P( | x—14] <5) = 3x} = 1/2 Ans. © OEE EE Example 12:7 Find the mean and the standard deviation of the uniform distribution f(x) = 1Jn; (x = 1, 2, ....m) Solution : Mean (u) = E(x) = 5x. f(x) = & x(Ifn) = (pn) ze = (I/nX1+2+...47) od, 2n 2 B(x?) = x". f) = = x*(Un) = (In? 428+... +4) = MOE D2n+1) _ (120-41 CSG mer = Using (12.3.6), : 2 SD.(e) = J@F=nz ¥ Ans, (n-+1)/2, 125 Binomial Distribution —100 STATISTICAL METHODS Suppose that we have a series of n independent trials in each rraisie foisbabilay ofeecurrence of an eveat is fxcd and'constanty a. Then the propebility that the event occurs exactly r times in p trials (Section 11.8) is *C,p’g"~", where g=1—p and r may assunie any of the values 0, 1, 2, n. Ip general, the occurrence of the event is called “success” and its non-occurrence is known as “failure”. The generalised version of the theory is as follows ; In a series of n independeng trials, if the probability of ‘success? in each trial is a constant p, and the probability of ‘failure’ is q, then the probability of x successes (and obviously »—x failures) is given by the Binomial distribution (12.5.1). TABLE 12.4—Binomial Distribution sep | Capt? | {Note: ()_In the expression for f(x), besides the factor *Cz, the power of p (ie. probability of success) is the number of successes x, and the power of q (ie. probability of failure) is the number of failures n—x. (ii) The sum of the powers of p and q is always n, whatever be the number of successes. Gil) There are (n-+1) possible values of x, viz. 0, 1, 2... (iv) The total probability for all the (n-+1) possible successes is 1.] The distribution is known as ‘binomial’, because the probabilities are given by the binomial series (gto) = + Cipgr*+°Captgr tt o. tp® (12.5.2) A series of independent trials of a-random experiment with a constant pepbebility ‘of success in each trial is called a “Bernoullian series”, and this distribution is known as “Bernoulli’s distribution”, after the name of its discoverer James Bernoulli. The two constants n and p appearing in the expression for f(x). are known as ‘parameters’ of the binomial distribution. Tf the values of pee are known, the distribution is completely known (q=1-p)- & It should be noted that the binomial distribution holds under the following conditions :— (1). The result of any trial can be classified only under two categories ; e.g. either head or tail in a throw of one coin; either “six” or a “non-six” in one throw of a die ; either a ¢ non spade when one card is drawn ; eithe iy ectiv when an article is examBINOMIAL POISSON NORMAL 101 (2) The probability of success in each trial remains a constant, and dees not change from one trial to another. For example, the probability of obtaining a head in successive throws of a coin is always }; the probability of obtaining a defective article from a batch Ross not ne in aes oe with replacement, and prac tically remains a constant even in drawings without 1 when the batch is large. ei en @) The trials are independent, so that the probability of success in any trial remains unaffected by the results of other trials. For example, in successive throws of a coin the occurrence of a head at any trial will in no way affect the probability of a head or a tail in any subsequent trial; or if several coins are thrown together, the occurrence of a head or a tail in any particular coin does not alter the probability of occurrence of a head in any other coin. Important properties : (1). Binomial distribution is a discrete probability distribution, where the random variable assumes a finite number of values 0, 1, 2, ... m The distribution is specified by two parameters n and p. (2) Mean = np, Variance = npq (12.5.3) Standard deviation (0) = /7pq (3) Binomial distribution may have either one or two modes, When (7+1)p is not an integer, Mode is the largest integer contained therein. However, when (n+1)p is itself an integer, there are two modes, viz. (n+1)p and (n+1)p—1. (4) Skewness (94) = es Kurtosis (74) = sea (12.5.4) When p=q=}, the distribution becomes symmetrical for all values of nie. f(x) = f(n—x) = "C.Q)". (5) If x follows binomial distribution with parameters (m,. p) and y follows binomial distribution with parameters (m,~p), and x and y are statistically independent, then (x+y) also follows binomial distribution with parameters @,+7,, p). The result can be to several independent binomial variates with a common p. (6) Binomial distribution may be obtained as a limiting case of Hypergeometric distribution. Example 12:8 Three coins are tosed. Find rchabaltes of (ahead heads? hei, 3 bed G mare than Oe head ee Solution: Here, the ‘random experiment’102 STATISTICAL METHODS If we denote the occurrence of head as “success”, and the coi, are assumed to be unbiased, then Fe A Probability of “success” in a single Pe estatsieyicd toad ath sicigio coin 4 q = 1-p=1-}=} : 1 = Number of fodehendent trials = 3 Since the value of p is constant for each coin and the trials any independent, the variable x follows binomial distribution with pané meters n=3, p=}. Therefore, the probability of x successes ig Six) = *Cyp*q*-* = *C,Q)4(9)?-* = *CaG)* (i) Putting the values of x = 0, 1, 2, 3 respectively, ) = Probability of 0 success ertcar cite S(1) = Probability of 1 success ="C,G) =3x5=e (2) = Probability of 2 successes esate (3) = Probability of 3 successes ="C,@) =1 x=} (ii) Prob (more than one success) = f(2)+/3) =tti=3 (iii) Prob (at least one success) = 1—Prob (0 success) =1-f0)=1-L=ay Example 12:9 Five coins are tossed 3200 times. Find the ex- pected frequencies of the distribution of heads and tails, and tabulate the result. Calculate the mean number of heads and standard devi tion, [C.A., May 77] Solution: We assume that the coins are unbiased. The bability of getting a head (or tail) is 4 for each coin at each oat Also the successive trials are independent. So, the conditions of binomial distibution {RE fulfiled. The probabilities of O head, 1 head, ... are given by the s ji i sion of the binomial expression y nocseeie er ta a G+D* =)" + "CLG + °C") + 8C, & *CG)G) + (4)! = 1/32 + 5/32 + 10/32 + 10/32 4 5/32 + 1/32 The probabilities £ the different numbers of heads and toss of 5 :oins, and the single expected frequencies inBINOMIAL POISSON NORMAL 103 Number of Probabil Beads Tal Ee eae 0 5 1/32 100 1 4 5/32 500 2 3 10/32 1000 3 2 10/32 1000 4 1 5/32 : 500 5 0 1/32 100 Total 1 3200 The mean number of heads, and the standard deviation are now cal- eee i the following frequency distribution (see Examples 5:57 Number of heads (x) 0 1 o 3: cathe nites Total Frequency (f) 100 500 1000 1000 S00 100 3200 Mean (=) = 2.5, Standard deviation (0) = /5/2 = 1.12 (Note that the mean and sd. calculated from the frequency distribution agree exactly with the formulae (/25.3), viz. np = 5x} = 2.5, 8d. = Jnpg= J5X¥X4 = V5/2) Example 12:10 The overall percentage of failures in a ae examination is 40. What is the probability that out of a group of candidates at least 4 passed the examinations? _[I.C.W.A. Dec. a Solution: Let us denote the event of 9 candidate the examination as “success”. Te ig tequited t9 Gnd the ‘probebiity of at least 4 successes. yg = Probability ity of failure of a candidate = 40% = ees ate at ingle tall p = Probability of success in a si = 1l-g= 1-2/5 =3/5 n = Number of candidates in the group = 6 ‘Using binomial distribution, the probability of x successes is fla) = "Cupar = CaN Since ‘at least 4’ in a of 6 implies ‘either 4, or 5, or 6’, the Rr. ty of at leant 4 successes is given by fhe sum of (4) +f(S)+H6). puting 2—8, Ses ee we have fA) = *C,)4()* = 15x a a ;104 STATISTICAL METHODS Thus, the required probability is = 4860+2916-+-729 _ 1701 MA)+IS) +/06) = 15625 3125 12:11 In 10 independent throws of a defective die, the Ree, that an even number will appear 5 times is twice the pro- bility that an even number will appear 4 times. Find the probability that an even number will not appear at all in 10 independent throws of the die. . Solution: Let the occurrence of an even number be called a “success”. We write p = probability of getting an cven number (i.e. success) ina single trial, % q = probability of not getting an even number (ie. failure) n = number of throws = 10 Using binomial distribution, the probability of x successes in 10 throws is Six) = *°C, prgre-* It is given that KS) = 2xf(4) or, *°Cyp®qh = 2x ?°Cyp4q* ee ana t— axgreiee Simplifying, we get 3p = 5q or, 3p = S(1—p) Solving we get p = 5/8, so that q = 1—p = 1-5/8 = 3/8 **. Probability that an even number will not appear at all (ic. 0 success) is (0) = 9*° = (%)*°. Ans, ($)*° 12:12 Find the mean and thestandard deviation of binomial ibution with parameters n and p. Solution: The p.m.f. of binomial distribution is Sx) ="Cep*get; (X= 0,1,2 Hence, by definition (/2.3.3), Mean = E(x) = 2 x fix) = Secure = 0X (*Cop°g*) + 1X(C,pg*-') + 2X(*Cptg*=*) + © 3K (Cap gh-*) + ow + .AXCQPMG). = 8, 3 ge = Oe eae) re on)BINOMIAL POISSON NORMAL 105 = npgr*en(n= pa" 1M GRD » PgR. emp = np [q"-?+(n—l)pq" 1002) pPgn-®e a pho] = np|qh-2-+*-C,pq* eee 5 oat 1 eae = np(q+p)*"*, replacing by n—1 in (12.5.2). = np(l)"-*, because pt+q = I. = np(l) = mp Therefore, Mean (#) = np. Again, by (12.3.6), o? = E(x*)—y" But, E(xt) = Ex4 f(x) = zlxxe— +210) = Ex(x—1) fix) + Ex fle) = sx(x—-1) fle) + 4 Now, =x(x—1) flx) =i x(e-1). "Cyptg*-* = %-1[*Cop°a} + 1x 0("C pg?) + 2x UIC ap2g® *t + 3X 2*Cap2gh?} + -. + m(n—1)(°C,p"Q"). a n(n—1) . = 040+ 2x1 (RD pre a + 3x2 ened erat} on + mln INLD") = n(n—1)p2qr-* + n(2—1) (n—2)p2gh-* + oe +(n—1 )p* = mn—1)p* (gr? + (n—2)pqh-? +. + Pm] = n(n—I)p? [g-* + *8Cypgh ®t wn + 8Cy gp 9] = n(n—1)p*(q+p)*"*, replacing n by n—2 in (12.5.2) = n(n 1)p%(1j""*, since p+q = 1 = n(n—1)p* 2. E(x*)= n(n—1)p*+p, and hence o® mm E(x*)—p* = n(n—1)p*+u—p* = n(n—1)p* +np—(np)* = ntp*—np*-+np—n*p? = —np*-+np = p\—p) = npg, since q = 1—p. ¢. Standard Deviation (7) = Jnpq das. raean = ap ad. = Jpg ” ial106 STATISTICAL METHODS Example 12: 13 Arithmetic mean and_standard deviation of g binomial distribution are respectively 4 and 8/3. Find the values of wand p. WBS. "8 Solution : Using the formulae for mean and 5.d. of the binomiay distribution np = 4, Jnpq = 8/3 uaring the second relation, npq = 8/3. Now putting »p = 4 and Solving we have q = 2/3. Hence, p = 1—q = 1/3. Putting the value of p in np = 4, we get n = 12. Ans. n= 12, p= 1/3, Example 12:14 Find the mode of binomial distribution with Parameters n and p. Solution: The ‘mode’ of a discrete distribution is the value of the variate which corresponds to the maximum probability. In the binomial distribution, the random variable X assumes only the values 0, 1, 2, ... and the probabilities are given by A(x) = P(X = x) = "C, p* ge-*; (x = 0, 1,2, ... n) The mode is that value (or values) among 0, 1, 2, ... n corresponding to which f(x) is the maximum. De) ee Ca-5 pe et! Kom rcp FO) 1 geen 1 < ga-« ie . xT G—ayt? ue erp Gs (1) {n—x+1)p (n+ 1)p—xp ‘ We find that : (a) fix—l) < fix) whenever xq <(n+1)p—xp or, x(p+9)<(n+1)p; ie. x < (nt)p o».(2) (b) Sx—1) = f(x) whenever “xq = (n+ 1)p—xp; ie. x =(n+1)p (3) () fix—1) > fix) whenever xq >(n+l)p—xp; ie. x >(nt1)p_ ...(4) We discuss the following cases : (Case 1) When (n+1)p is not an integer. Let r be the largest integer in (n+1)p. Si i a distribution, ‘5 assumes only integral a nen Pet —BINOMIAL POISSON NORMAL 107 From (2), flx—1) < f(x), when x = 1, 2,3, ..7 or, SO fix), when x = r+1, +2, ...0 or, f(r) > flr+ 1), fir +1) > fr +2), .. fal) > flr) or, fr) > fir +1) > fir+2) > ... > fin). (6) Combining the results at (5) and (6), $0) fr F1) > - > flr). We find that the probability f(r) corresponding to x = r is the maximum, and therefore the mode is r. Thus, when (+1)p is not an integer, the mode is the largest integer contained in (n+1)p. (Case 1) Whea (n+1)p is an integer, say (n+1)p =F. From (2), fx—1) < f(x), when x = 1,2, 9,0 aE or, f0) = fl) < .. fla), when x = rt, r+2, 0.0 or, fir) > fr t+) > «. > fir). (9) Combining the results at (7), (8) & (9), * $0) < f(D) << flr-)) = fr) > ft) > > f(n) We find that the probabilities f(r) and f(r—1) corresponding =r and x=r— i are equal, and have largest values. There are eee ? adi and the “Geiiaion is ‘bimodal (ie. have two modes). » Thus, when (n+1)p is an integer, say r, the binomial distribution has two modes, viz. r and r—1. Poisson distribution is a discrete probability distribution and is defined by the probability mass (p.mf) . fx) = 5 (220,10 oy OD (26.1) where im i i i constant and is tren oe Tesiguncimmepep nator sh108 STATISTICAL METHODS ‘The constant m appearing in (12.6.1) is known as ‘parameter’ of the Poisson distribution. It may be noted that the variable assumes an infinite number of values x = 0, 1, 2, 3, ... 2%. with probabilities as shown below: TABLE 12.5—Poisson Distribution This distribution is named after its discoverer S. D. Poisson. Important properties : (1) Poisson distribution is a discrete probability distribution where the random variable assumes a countably infinite number of values 0, 1, 2, ... o. The distribution is completely specified, when the parameter m (positive) is known. (2) Mean=™m, Variance = m (12.6.3) Standard deviation (c) = /m (3) Poisson distribution may have either one or two modes Ulike the binomial distribution). When m is not an interger, mode is the largest integer contained in m. However, when m is itself an integer, there are two modes, viz. m and m—1. (4) Skewness (7) = i Kurtosis (7) = 1 (12.6.4) Thus, Poisson distribution is positively skew and leptokurtic, (5) If x and y are independent Poisson variables with parameters m, and m, respectively then (x+y) also follows Poisson distribution with parameter (m,+,). (6) Poisson distribution may be used as an ximation to binomial distribution, when p is small, and n is large, np finite. Some examples of Poisson variable are: (@ Number of printing mistakes per page (or typographicalBINOMIAL POISSON NORMAL - 109 (X) Number of bacteria sent in a given liquid cults unit area Dbeervedl under is nd Number of radio-active atoms decaying in a given interval of time, (vi) (vii), Number of defects per unit area of sheet material (eg. paper, cloth, or metal sheet). Number of cars passing through a road crossing per unit time interval (e.g. 1 minute) during a busy period. (vill) Example 12:15 A random variable x follows Poisson distribution with parameter 3, Find the probabilities that the variable assumes the values (i) 0, 1, 2, 3; Gi) less than 3; (fii) at least 2, (Given that e>* = 0498), Solution: For the Poisson distribution with parameter m = 3, the probability of x successes is eee 1) ican xT (i) Putting x = 0, 1, 2,3 successively, the required probabilities are fo) = £ - “4 = e-* (since 0! = 1 and 3° = 1) fay = SF = SF et = 3x.0498 = .1494 n.3P 52.9) fl2) = © = = 45xe® = 4.5x.0498 = .2241 e79.27, FR SB) = = 4.5xe* 4.5x 0498 = .2241 (ii) Since ‘less than 3° implies ‘either 0, or 1, or . oe of less than 3 successes is given by the sum of the {O)+{(1) + f(2) = 0498 +.1494+4.2241 = 0.4233 iii) Since ‘at least 3’ implies ‘either 2, or 3, or 4, .. pati of at least 2 successes is given by the sum of the proba- ilities of the infinite number of terms. But since the sum of the probabilities for all values of x, viz. 0, 1, 2, ...0° is 1, ie LOAM +IQA)+ ow oh i a al aa i atiBINOMIAL POISSON NORMAL ul put it is given that P(x = 1) = P(x = 2) or, om -——* Dividing both sides by (e~".m), we have 1=F; ie. m=2 Therefore, the probability that x assumes the value 0 or | is P(x = 0 or 1) = P(x = 0)4+P(x = 1) = f0)+fl) = e-m™(1 +m) = e7*(142) = 3e* E(x) = m = 2, Ant. 30-*,2, ~ Example 12:18 Find the mean, and variance of Poisson distri- bution. [C.U., BSc{Econ) *81 ; M.Com. ‘75, 77) Solution: Poisson distribution is defined by the probability mass function (p.m.f.) » fix) = em (x =0,1,2,... ©) Mean = E(x) = = x fix) = 0,0) + 11) + 2.f(2) + 3,3) + ... to infinity. cs -m mm mm —=_m* HOF Lem + Rem + REG Ho -m mg oom mt =e eT ee é men(1 cE mem + 3) meen m E(x*)—{ E(x) }* E(x*)—m?, since E(x) = m™ q Elx(x—1)+-x)—m? } E(x(x—1))-+m—m* Now, Eix(x—1)) = 2, x1). fx) = Of(0) + 1xOfll) + oe Variance112 STATISTICAL METHODS =m? -m me -m m* = 2.e TELE 3x2.e Nias + 4x3.e ate =e" + on ap ont on ee m m* = miem(t+ m+ 2 +.) = m*e-" e"as m* ©. Variance = m*+m—m* = m Example 12:19 Find the mode of Poisson distribution. oo [C.U., B.Sc(Econ) '81 (New)} Solution: In the Poisson distribution with parameter m, the random variable X assumes the values 0, 1, 2, ... e and the probabilities are given by fis) =P (X= x)= em; (x= 0,1,2,~ ©) The mode is that value (or values) among 0, I, 2, 3, ... 0 correspond- ing to which the probability f(x) is the maximum. (x1 oe en ie fea) =e ‘gent /* br ( * 0) x == (1) We find that.: (a) f(x—1) < f(x), whenever x <_m eer (2) (b) f(x—1) = f.x), whenever x = m ee. (3) (c) f(x—1) > fx), whenever x > m w+ (4) (Case I) When mm is not an integer. Let r be the integral part of m. Since in a Poisson distribution esas only non-negative integral values, the result (3) is impossible From (2), flx—1) < f(x), when x = 1,2, 3, 7 2 fl) < fll) f(x), when x = r+1, r$2, 0... fy> fet) >fe+D> .. ae ae (O Combining the results (5) and (6), we have NO) fr+ > ... Thus, the mode of Poisson di: ion i: sd dae ms imation i'r. be. the Ingest integerBINOMIAL POISSON NORMAL is (Case 11) When m is an integer, say m =r. - From (2), f(x—1) < f(x), when-x = 1, 2,3... (r—1) A) < fl) < fQ) < ~ < fr—-1) we (7) From (3), f(x—1) = fx), when x = oe Srl) = fir) + () From (4), f{x—1) > f(x), when x = r+1,r+42, .. oA) > fie +1) > fir +2) > et) Combining the results (7), (8) & (9), FO) < JD << fll) = fl) > fir) > ewe We find that two terms f(r) and f(r—1) are equal and the largest of all. Thus the Poisson distribution is bimodal when m is an integer, the modes being m and m—1. (Note: This method is identical with that for binomial distribution > 12.7 Poisson Approximation to Binomial distribation Poisson distribution (12.6.1) may be obtained as a limiting case of Binomial distribution (/2.5.]) under the following conditions : (i) the number of trials 7 is infinitely large ; ie. n—> 00 ; (ii) the probability of success (p) is extremely small ; ie. p—> 0; (iii) the mean np = mm is finite. Under these conditions it can be shown that the probability of x successes in Binomial distribution can be closely approximated by the probability of x successes in Poisson distribution with parameter m= np. i.e. scape ee =) (12.7.") Proof: *Cyprges = MO=DOKD. A EED pac, pyee 1-384 sb Diah as 2 ¢apy*(1—"2)* = NB eay ‘As n->ee, each of the factors (1—"), (1-2), ... (1-*=") tends114 STATISTICAL METHODS *C,p*g-* = = 2) 1) 2 me aon x! Since the probabilities of Poisson distribution are easier to com- pute than those of Binomial distribution, for practical purposes the Poisson approximation is used when p is less than 0.1 and np is not very large (say, less than 10). Poisson distribution thus finds important applications in such phenomena where the probability of occurrence of an event (success) 4s extremely small, but the number of opportunities is infinitely large, so that the mean is something finite. Hence this distribution is some- times known as the “distribution of rare events”. Example 12:20 In tuming out certain toys in a manufacturing Process in a factory, the average number of defectives is 10%. What is the probability of getting exactly 3 defectives in a sample of 10 toys chosen at random, by using the Poisson approximation to the binomial distribution? (Take e = 2.72). 1.C.W.A., Dec. ’78} Solution: Let the occurrence of a defective toy be called a “success”. The number of defectives (x) follows binomial distribution with parameters n = 10 and p = 10% = 0.1. If this distribution is to be approximated by Poisson distribution = me flay =m we have m= np=10x0.1=1. Hence,, the probability of 3 defectives in the sample is wet, Boe t al 13) = Gee = ae =n = 1 [ Note: Using binomial distribution, the true probability is %€(0.1)4(0.9) = .057 (upto 3 decimals) Even for a small n=10, this agrees fairly well with the Poisson approximation 061 (the difference being only .004).] Example 12:21 2% of the items made by a machine are defec- tive. Find the probability that 3 or more items are defective in a sample of 100 items (Given e~' = 0.368, e~* = 0.135, e~* = .0498). [Dip. Management 77) _ Solution: The number of defectives (x) follows binomial distri- bution. Prep der care is small, and n= 100 is fairly large, making np = 02 =2 a finite quantity, we Pois i- mation to binomia) distribution with pa baba snes - ee138 STATISTICAL METHODS Since x is a continuous random variable, by (/2.11.5a), P(x = 45) =0 Ans. 0.3, 0.28, 0.2, 0. 12.13 Normal Distribution Normal distribution, or Gaussian distribution, is a continuous probability distribution and is defined by the density function (p.d.f.) _@-—? f@) = Ee Ww ( acca) (12.13.1) where ~=mean, o = standard deviation. [7 and e are two mathematical constants having the approximate values 22/7 and 2.718 (see 12.6.2.) respectively]. and o are known as ‘parameters’ of this distribution. Although at first glance the expression (/2.13.F) may look very complicated, this is the most useful distribution in theoretical statis- tics because of its many important characteristics. A random vari- able x is said to ‘follow normal distribution’, or to be ‘normally dis- tributed, if the probability density function is of the form (2.13.D. The probability curve of normal distribution is known as Normal Curve. The curve is symmetrical and bell-shaped (see Fig. 12.3) and the two tails extend to infinitely on either side. If a random variable x is normally distributed with mean « and standard deviation o, then z= (x—4)/o is called the “standard normal variable” or “normal deviate”. It bas the density function 2 we) = he PS ee Pa) (12.13.2) The continuous probability distribution defined by (12.132) is known as Standard Normal Distribution. This is in fact a special case of normal distribution with mean 0 and standard deviation 1. Area under Normal Curve : As in all continuous probability distributions, the total area under the normal curve is | ; and the probability that x lies between c and d, denoted by P(c Area between z = +1 is 68.27% ” ee aBINOMIAL POISSON NORMAL 141 ‘The last result is especially useful in the theory of Statistical Quality Control (S.Q.C.). It implies that if a random variable x is normally distributed with mean # and s.d. ¢, then almost all the values of x will lie between the limits u—3¢ to 4+3c, ie. Mean + 3 (S.D.). (x) If x and y are independent normal variates with means wy and #3, and variances o,* and o4* respectively, then (x+y) is also a normal variate with mean (u,+/3) and variance (7,*-+e,%). Importance of Normal Distribution : The normal distribution has many important algebraic properties for which it is used widely in statistical theories. In most cases of physical, biological and psychological measurements, the data are found to follow normal distribution. Statistical Quality Control. (S.Q.C.) methods in the manufacturing industry and the Theory of Errors of observations in physical measurements are also based on normal distribution. The normal’distribution is used (Section 12.14) to serve as approximations to the binomial and Poisson distributions. In the theory of sampling, it has been found that any statistic based on a large sample approximately follows normal distribution. The result considerably simplifies the work of testing statistical hypo- theses and is also useful to find the confidence limits of parameters. Example 12:36 The height distribution of a group of 10,000 men is normal with mean height 64.5” and s.d. 4.5”. Find the num- ber of men whose height is (a) less then 69” but greater than 55.5”, (b) less than 55.5”, and (c) more than 73.5”. : Solution: The mean (m ) and standard deviation (o) of the normal distribution are given to be « =64.5” and o=4.5”. (a) Percentage of men whose height lies between 55.5” and 69” =Area under standard normal curve between the vertical lines at the corresponding standardized values, viz. ae 5H 64S = 69-64.5 _ a 2 and z a5 i. From the percentage distribution of area under the standard normal curve (Fig. 12.3), it is found that the area between z= —2 andz=1 is (14% +34%+34%) = 82%. This theans that 82% of the total number of 10,000 men are expected to have heights between 55.5” and 69”. Hence, the required number of men is 82% of 10,000, i.e. Yoo 10,000 = 8,200. (b) Percentage of men whose height is less than 55.5” = Area under standard normal curve to the /eft of the Standardized value z = 35.5=64.5 _ —2. 4.5 = 2% (see Fig. 12.3) The number of men is therefore, 2% of 10,000, i.e. 200.142 STATISTICAL METHODS (©) Percentage of men whose height is more than 73.5” = Area under standard normal curve to the right of the Standardized value z = 735-645 — 2. = 2%. etd i Hence, the number of men whosé height is more than 73.5” is 29% of 10,000, ie. 200 Ans. (a) 8200; (b) 200; (©) 200. Example 12:37 The mean weight of 500 male students at a certain college js 151 Ibs. and the standard deviation is 15 Ibs. Assuming that the weights are normally distributed, find how many students weight () between 120 and 155 Ibs., (ii) more than 155 Ibs. Sed (0.27) = 0.6064 and © (2.07) = 0.9808, where ® (¢) denotes area under standard normal curve to the left of the ordinate at t.] Solution: The mean (#) and the standard deviation () are : # = 151 Ibs. 7 = 15 lbs. (@ Proportion of students whose weights lie between 120 & 155 Ibs. = Area under standard normal curve between the vertical lines at the standardized values, viz. z= (120—151)/15 = —2.07 and z= (155—151)/15 = 0.27. Using (/2.13.7), P(120 < x'< 155) = (0.27) — &(—2.07) = (0.27) — (1 — (2.07)}, by (2.13.5) = 0.6064 — 1 + 0.9808 = 0.5872 {see Fig. 12.4(a)]. The number of students whose weights lie between 120 and 155 Ibs. is therefore 0.5872 500 = 294 (approx.) (ii) Proportion of students who weigh more than 155 Ibs. is P(x > 155) = 1 — P(x < 155) = 1 — 60.27) = 1 — 0.6064 = 0.3936 The number of students who are expected to weigh more than 155 Ibs is 0.3936 x 500 = 197 (approx.) Ans. 294, 197. Example 12:38" A sanble of 100 dry batt t find the length of life produced the rollbiiog series cols ee o =3 hours. Assuming that the data are normally distributed, what percentage of battery cells are expected to ‘have life () more than 15 hours, (ii) less than 6 hours, and (iii) between 10 and 14 hours? [L.C.W.A., June 79) Given, z 2.5 2 1 0.67 Area 4938 47723413 2487 Solution: (Note: The t table shows the area sitBINOMIAL POISSON NORMAL 143 values of z. This is clear from the fact that the area increases with larger values of z, and the area upto z = 2.5 is almost 0.5. Remem- ber that the area under standard normal curve between z= —3 and z= +3 is almost 1, and hence the area from z = 0 upto z=3 may be taken to be 0.5] The percentage of battery cells with life between specified limits will be given by the proportionate area under the standard Sa between their standardized values. Here, mean = 12, sd. = 3. Standardized value of 15 hours = (15S—12)/3 = +1 Standardized value of 6 hours = ( 6—12)/3 = Standardized value of 10 hours = (10—12)/3 Standardized value of 14 hours = (14—12)/3 = +0.67 Ficures 12.4—Area under Standard Normal Curve (a) (b) Proportion ot battery cells with life more than 15 hours = Area under standard normal curve to the right of z=+1 = (Total area to the right of z = 0)—(Area between z=0 and z= +D = 0.5—0.3413 [see Fig. 12.4(b)] 1587 = 15.87' Proportion with life less than 6 hours = Area under the curve to the left of z= —2 = Area to the right of z= +2, (since the curve is symme- trical about z = 0) = 0.5—0.4772 = 0.0228 = 2.28% Proportion with life between 10 and 14 hours = Area under the curve between z = =2x(Area between z= 0 and 7= y a144 STATISTICAL METHODS Ficure 12.5—Axea under Standard Normal Curve @ ) 0°67 0 O67 ° 164 Hence the required percentage of battery cells with life (i) more than 15 hours is 15.87, (i) Iess than 6 hours is 2.28, and (iii) between 10 and 14 hours is 49.74. Ans: 15.87, 2.28, 49.74 Example 12:39 “he mean of a normal distribution is 50 and 5% of the values are greater than 60. Find the standard deviation of the distribution (Given that the area under standard normal curve between z= 0 and z = 1.64 is 0.45). [LC.W.A., Dec. 79] Solution: The probability that x takes a value greater than oO is 5%, ie. 05. This must be the area under standard normal curve to the right of the ordinate at the standardized value 60 — 50 _ 10 o ~ oe "Since the area to the right of z = 0 is 0.5 and the area between z= 0 and z= 1.64 is given to be 0.45, hence the area to the right of = 1.64 is,0.5—0.45 = .05 [see Fig. 12.5’ Thus 10 = 1.64; of, o = 10/1.64 = 6.1. Example 12:40 Assuming that the height distribution of a group of men is normal, find the mean and standard deviation, if 84% of the men have heights less than 65.2 inches and 68% have height lying between 65.2 and 62.8 inches. —_[1.C.W.A., June ’78] Solution: Let » denote the mean and o the standard deviation; and z, and z, represent the standardised values of 62.8” and 65.2”, ie. n= cee and z3 = $2=H er (Note that since 65.2 is greater than 62.8, z, will be greater than z,). Since 84% of the men have height below 65.2”, this must be the area under the standard normal curve to the /eft of the standar- dized value z, But from the percentage distribution of area under the curve (Fig. 12.3) we find that the area to the left of z= 1 is 84%, Hence, z, must be identical with 1, ie. z, = 1. sg Mai it 3s given that 8% aBINOMIAL POISSON NORMAL 145 normal curve between z, = 1 and a smaller value z, is 68%. From Fig. 12.3 it is found the area between z= 1 and z= —1 is 68%. Hence, z, must be identical with —1, ic. z, = —1. Thus, we find that z,= —1 and z,=1. or, S28" = — 1 and 2-8 21, by or, 628 —# = —o and 65.2—p=0 Adding the two equations, 128.0—2 4 =0, so that »= 64. Now, substituting this value of # in one of the equations, 1.2. Ans. mean = 64.0; s.d. = 1.2 (inches). Example: 12:41 For a certain normal distribution, the first moment about 10 is 40, and the fourth moment about 50 is 48. Find the arithmetic mean and the standard deviation of the distri- bution. [ Given: For a normal distribution 4, = 1.3.5 ... (n=l) o*] [1.C.W.A., Dec. 78} Solution: Mean = A+first moment about A, (see 7.1.10) = 10+40 = 50 Since the mean is 50, the fourth moment about 50 is the fourth central moment /,. Again, putting n = 1, 2 in the given formula, Hy = 97, ty = (1 X 3) ot = ot The first relation shows that o? is the variance of normal dis- tribution, i.e. the parameter is the s.d. From the second relation 30* = 48 (given); ie. o = Ans. Mean = 50, S.D. = 2. 12.14 Normal Approximation ‘o Binomial (Poisson) In the binomial distribution (12.5.1), if neither p nor q is very small, but n is large, then the binomial distribution can be closely approximated by the normal distribution having the same mean and standard deviation as those of the binomial, i.e. “#=np and o = Jnpq. _ In order to find the probability P(a < x < b) of the discrete binomial distribution by the continuous normal distribution, some corrections to the values of a and 6 are necessary. The number of successes is always a whole number, and hence the number of ‘Succcesses a corresponds to an interval a—jytoa-+ 4 inthe con- tinuous scale (¢.g. ‘47 successes’ refers to any value between 46.5 and 417.5). Therefore, the number of successes a actually starts from @ —4, and the number of successes 6 ends at 6 + % in the normal distribution. The probabilit 6) that the number of46 STATISTICAL METHODS 12.11,3), this probability is given by the area under standard cee vertical lines ute standardized values. : P(a < x <5) of binomial distribution = P(a—} < x < b+4) of the approximating normal distribution. = Area under standard normal curve between the vertical lines at a’ and 5’, (12.4.2) where a’ = CORR" and 0 -C+)-” npg The larger the value of n, the better is the approximation. But, for practical purposes, the approximation is generally used, if both p and q are larger than 0.1 and is not less than 30. When the parameter m is large, Poisson distribution can be approximated by the normal distribution with the same mean and standard deviation as those of Poisson, i.e. mean = m and s.d: = /m. In practice, the approximation is used, when m > 10. 12:42 A fair coin is tossed 400 times. Using normal approximation, find the probability of obtaining (i) exactly 200 heads, (ii) less than 210 heads, (iii) between 190 and 210 heads, both in- clusive. Given that the area under standard normal curve between z = 0 and z= 05 is .0199 » 2=Oandz 95. 3289 » 2 =0andz = 105 ,, 3531 Solution: Let us denote the occurrence of a head as “success”. The number of heads (x) follows binomial distribution with m = 400, and p =}. Since n is large, we use normal approximation to’the binomial with “= np = 400x} = 200 ; o = Jnpq = J400xFx} = 10 (i) Probability of obtaining exactly 200 heads = P(x = 200) of binomial distribution = P(199.5 < x < 200.5) of normal distribution = Area under standard normal standardized values zBINOMIAL POISSON NORMAL 147 (ii) Probability of obtaining less than 210 heads = P(x < 210) of binomial distribution = P(x < 209.5) of normal distribution = Area under the standard normal curve to the left of z = (209.5—200)/10 = .95 = (Area to the left of z = 0) + (Area between z = 0 and z = .95) = 0.5 + 0.3289 = 0.83 (approx.) (iii) Probability of obtaining between 190 and 210 heads = P(190 < x < 210) of binomial distribution = P(189.5 < x < 210.5) of normal distribution = Area under standard normal curve between z = 189.5=200 _ _1.95 and z= 210.5200 +1.05 10 = 2 x (Area between z = 0 and z = 1.05) = 2 x .3531 = 0.71 (approx.) Ans. 0°04, 0°83,071 Example 12: 43 A variable x follows Poisson distribution with mean 16. Find the probability P(x > 20). Given F(0.875) = .8092 where F(x) is the standard normal distribution function. Solution : Using normal approximation to Poisson distribution, Mean (4) = m= 16, S.D.(0) = Ym= Vl6=4 P(x > 20) of Poisson distribution . = P(x > 19.5) of normal distribution = Area under standard normal curve to the right of z = (19.5 — 16)/4= 0.875 = 1 — (Area to the left of z = 0.875) = 1 — F(0.875) = 1 — 0.8092 = 0.19 (approx.)APPENDIX TasLe I at z = O/and given values on Ol 02 - 03 04 05 -0000 .0040 .0080 .0120 .0160 | .0199 -0398 .0438 .0478 .0517 .0557 | .0596 .0793 .0832 .0871 .0910 .0948 | .0987 -1179 .1217 .1255 .1293 1331 | .1368 +1554 .1591 .1628 .1664 .1700 | .1736 -1915 .1950 .1985 .2019 .2054 | .2088 .2257 .2291 .2324 .2357 .2389 | .2422 -2580 .2611 .2642 .2673 .2704 | .2734 4 : és -3023 -3159 .3186 .3212 .3238 .3264 | 3289 3413 .3438 .3461 .3485 3508 | 3531 3643 .3665 .3686 .3708 .3729 | .3749 eee eee Se ee See main RWHHS CHIDH BLOHEO | N bb 2 2 2 So re) 3 oS S 8 & 3 g a Peeses Eeeve Serre g : g z g of z) 06 0239 0636 1026 -1406 1772 -2123 -2454 2764 3051 3315 3554 3770 +3849 .3869 .3888 .3907 .3925 } .3944 .. -4032 .4049 .4066 .4082 .4099 | 4115 « 4192 .4207 .4222 .4236 .4251 | .4265 .42 Area Under Standard Normal Curve @roportion of area under standard normal curve between the ordinates 07 0279 0675 -1064 1443 . -1808 . 2157 2486 - 2794 | -3078 .: -3340 . 3577 -3790 -: -3980 4147 « -4292 4418 « 4525 « 4616 « -4693 . 4756 « 4808 4850 « A884 A911. 4932 . 4949 4962 . 4972 \ 4979 4 4985 . + 4989 4992 « 08 .0319 0714 -1103 “4997 4997 4997 nny WNnNRENEDneenrerernere 09 0359 .0753 1141