Biome 145% ic 5, 92 e- AF FITTING THE NEGATIVE BINOMIAL DISTRIBUTION + ‘To BIOLOGICAL DATA. ‘le Cat Art Beinn Sao ond Vie ae [Nore ox tm Emcee Frenvo op tae Neasnve Boyowie Det rie "Cole In stodying the covirénce of plants and animals autre, she ‘umber of fnviduals may be sounted la each af many eg so ‘pce or tine, The erg! count an be emanate! a's ogy distention, showing the number ef unite containing = = 01, 3, - individuals of a given spe If evry unl the mre Seg txposed equally to the chance of cntasing the oan, the teboton wold fallow th Poinwen sr, each nit hving the pope tion mean as its expected frequncy Its eay to txt whether the ‘aration in the number of india per tit neces mith this hyper hein Since the expected variance o'n Poiwon diction evel to its mean, the cbmrved variance mulipied by the dogo a freedom n, iy be divided y the ample naan #0 cian? = noe ‘More often than nat x? a wiguiastly larger than ite expectation ‘ot only in distebutons of plants and anima inate bu even oe Auber of distributions have buen devine for sein in which te ‘arian in iif larger than the nua 11, 1) requeny ‘ho basis of mare o lew complex Bisa! snd, fn the proeat ape his haractritic wil be ald “over dgeson” Perey te fit ofthese was the negative tinemlal, whieh ror in deste fe Poison sees from the pint binamial (27,2) aheugh it had bee formulated tn 1714)” Comparisons of expt and sbsrvo twibutions have ahova its wide splcblity to bilgi data he ‘etre ease with which the negative bic can be romped sd NEGATIVE mINOMIAL DIEERIBUTION m ber desirable propertin that have been denied by Fisher (12) end by Anscombe (2) have now ben sappemented by praciable ms mum Hiketiiood eatination of Its parameters deco in Ue ote ‘by Bi Rood Pier atthe end of the prem paper ere we il cailer some caratvatice of the neptive bin, elias at parameters and ther presen, the auton of the exported ie, ‘omnis for given ample, nd ea applablty totaal data, ‘The Negathe Binomial Diarition. The egaive toma ai. bution ie cmpleely dated by two parte, te arithmetic sn ns postive expnent kT called hy analogy withthe patie ‘binomial disbatice, (+ 9)", whore wb the number duals Jn group and q and par the expected proportions in to contesting wegncies with ¢ 4+ p= 1. Tn computing the matics ofthe pote ‘inonil trom dstcbutions of ear eo counted wth a hacmocy™ tometer, Student (27) bverved that tof hie suas gare negate alos for p and but verse Sted his chervties ver rl ‘Tine another cate (2) wer deuribl hy the negate inal, = 97%, whare p= mb and = I+ p. By expansion of ths ex ‘rin, the proeblity cat an chcrational unt wil contain 71,2)» indian po eB ® w/(8 4m). The probability fr agen =i mult otal suber of nla ceunod, to obtain the expected frequency (¢) of waite mith ndividsale The cary defaed bythe 8 (ri nimodal, 2 tat Sng the negative boil fo a olrred ditributon uy apparent bimodality (or multimodality) i ‘tibted to radar samp ‘The negative bina i the population mean m,the parameter of te Poon dation, ‘ot conan but varie oontinooal in» dateaton popartonl 0 ‘that of athe vasane a epativebinemial approaches the ene, ‘the over-lipersion decrease >= and'p 0 Under thew eo 6, for large vals of when > 13, and fr m in he ter redite toe wien (E+ m) (E+ 2)/m 3 16. Tn prt the pope ta values and are replied oss bythe statics 2 and & "hor eample the timation of fy leads by Bg. 8 to hy = (suey 2.27968 — 1.14067) ~ 110670. TE, ha been etimatd idk an efcioney of 90 perm or better, the inequity 2318 > ‘3407/L147 b 15 shoud bold, bt slace 69 > 1, ths method isnot al for ike preset sre {@) Analtrsative etna (2) i based pon the ratio ofthe total umber ef anita inthe mpl (9) tothe amber of nit widow frais (0). From Eq 1 the expected probability foe 2 = 0's = 17 an Pa roplaced by the propetion observed inte 70 slam mehave {UN = If Since q'= L'+ w/b, an erative aauion ls mmary, ‘Th rogiedertinate of By ie that Which alas tho auton Betog + af) = ioe 10 6 ‘The et sido of Ha, in computed wie, with diferent il values 1 ow giving larger andthe ther» smaller product than the coti i 19 BIOMETRICS TOE 88 stant term onthe right ofthe equality. Inerpltng between thee too product fer log) los to at approximaton off Wh ‘hich the prooas eas be continued unt ha te required ream, ‘Tocetinatokwith an eseny of 90 pest or more, a ar 13 ‘be unt st bem, ttf che mam ne han 10, enough re empty nll are gure omy th insult ( 4 0.17) (Pp ~ 0.23) 5020. The tomas inthe inequality re replaced coma wth fe = and Py = JN trom the ample ‘Since the numberof sro fequencis in Tab 1 erative are, smethod 2 i the more promising. Asa tet of te pected cine, I find (20:17) it ~ O82) » “108, which i very oe tothe Ievl fr 90 peront ficiency. From Tg 8 tho relied fy ete el ‘yale ¥ foe whlch lg(h “+ 4/8) = leg(80/20) = 380008. The “ttl al of — ta bated ce etna fa met end 5 cenptin erage lone YAN wag 1 awe 22075 9 same ane oma aaseea sores mas repented with H = 98 reversing the inoqulty. By interpol ig ='gn 402 (Saqp0a — sa074e)/(38i7e8 — S074) — 362, ‘Which Mighty underestimated th reqied value, By interpolation Tetwoon and bby = O02 1G) For many dition, cannot be deerme by ether of Ae above techniques with an aceptabeeficinay.Tn tne end all xt! cases, the computed by 9 8c 5 may be considered as the Art sap tomar» dative sation. This provided hy the method ff maxima Halood (17,2). By aulabe arrangement of hcl Into, a developed by Si Ronald ier ine sppends to the prevent ‘ape, the procedure pracGenbe and rapid wen te large oben fon doe uot rend 20 ar 3D. Bors) ae capt rm rial ae of Klcted vo that they back Ue rogied eliza F, fr which t= Olathe eqntion wn dB) — wali) ® Where Is Ssgntee anata logarithm. Ae ist tp in the comp ‘ation, the aceumtlatedfequeny 4, ln all nit omtaning ee than [NIGATIVE BINOMIAL. DISTRIBUTION a 4 roan in written opposite each z, ‘Tho rsprocale 1/(8, + =) fru Barlow's Table (to sven pce or more) remit in en by A, and the products nested to blasthe suramaten in he tenn. ‘The motnd term may be determined onthe ser plce common lth of (+ 2/4) mld by 23025851 > Toa fit mor (= 1) ono wih be it ae i ‘unaly bed upon the Erm Hy 8. Ta second eal vate depend ‘wpm thei a, Ir postive, K > ji agave 8 < the tp diflerig jt meugh to give oppouie wins to yaad 2, «The ‘hid tril value Kf, betwen and, obtaned by liner inter obion for += 0, but the emputad s,s rary exactly ero. For ‘rein, te peferble to compute a a witha gn opps to that tr by cloctng Bt about the sane dintance at byond ately Inttplated W for # =O. "This provides a masrower interval tan ich the fal may be interpolated and beter eetinate of Ste ‘atacand ‘Since the better of the two approximations of Fin ou example was tardy 80 percent ficient, the Ueno sletia would be prefered ‘Tha cumulative frequencies excoding each 2 wee Set Intel (Table), that for» 0, for example, = 190 — 70 = a) and for = = 1, [.= 80 ~ 38 = 42. Starting witha rl van off = 10 he ‘ioeal, 1/0 +2) foreach naa maltpliod by errepending 4, ‘ol the prods accurate in the ealelatar obtain the Se ter in he soe « (Ba. 8). This sm has bom lst patty at Plow 5h mond arm inte are, BAS87S X lag LAO Mnig 10s = 1006 = 10 ZealAlte +9) 1149089 1040s 12.5247 112.7961 Vind +27) =114s676- ~ni0Tm —u2s402 —119762 » eT 8 0185 a0 ‘Since ma pn te scond tit wae nee to = 1.05, leaing to n negative % Inerpalatng betwesn them for» ~ 0, = 10-4 (3067 x 05)/(3387 +3182) = 105, whic star, ave > O85. From lin interpolation betren and for» = 0, B= 10247, ‘Toinsure sponte mor'par sr, «new tl val wat sce off = 1028. Taerpaltion between sad gv thease frum leooleatinate ff = 102659 ‘The Variant of # and & "The sampling vainvos of he statics ‘the naptive binomial depend upon the parzters aad, tn Date the er replaced necemaily Uy the eoreponding sate12 IOMETHICS, JUNE ss and E, The mea x computed contin ll cam nd ite varince | afew ® “The err variance ofthe mean inthe empl from Table 1 somata wrth the mason Ualinood eins of mae Vi) = (1007 + 12820) /160 = 010, giving the wanda eer, = 1278 "The vasiance of F depends upon how it hasbeen estimated. Tn inn the variance of ie ethan Fi mal than ena (1) IE computed fom # by Ba, 5 ts largeanmple variance (2s ri = © nlved with B= £/(b +9. For hy = 110670 trom Ba. 8, B= 1.1407/281057 = 40007, aod V(b) = 6058/3888 = 13%, fm ‘hich this eine of hn a andard eroe 1872 ~ 270. (2) Hi determin from tha nomber of vr ait by a 5, largoaample variance (2) oem vii) * Non B= RI @ hee deal stadia at! pi ‘The x petite eg Teepe GO eal’ som eta ap = noe ‘id. the usd evr Oa wheat Wt sg a ea {Th vara oh mai alain he sea th eat fats SOAS aan de ep ave ‘emus ea ea ese a wy ine ‘Se oro lar ith wh ty ae hws mpd Sune vo HH 7 Hence the erotvasince of E = 1.0289 in | Beh) = (0m = 1.02 /¢2000 +0185) = 07614; tat the maxi Heino fad «standard err of =. 2780 [NBOATIVE BINOMIAL DISTRIBUTION 18 Tat for Agra with te Negative Binomial. Perhaps the moe cenvincog tat ofthe adequacy of the Degaive hao a ny geo ‘ase the greet between to fequencis (7) observed at ach = nd {heir expected valve () ae computed fram the tatistion ofthe spl ‘The drepancy betwee them easly etd by 2. “The expcad rune aro computed th B1, most ea in sscotsan and starting withthe somber expected at z= 0. This fist ‘xpectton i determined withthe ld of Place Inputs as aXe ay sd he enced ents for = = 1,2, 8, «+ a8 4 Gta, aa “Moc desi laces shoul! be rte in the ealeator at cach sags ‘han aed be recorded, 9 a2 to eras accumulating rounding ero. ‘The obwerved aod expect quale are thn compare by 2 whee re) « 2 has three fower dros of fem tn the umber of ration that fre summa, Aa unl te fequnces wth small expectations are Dove, preferably so that no expectation ls than 5. x some ood agroomest etree the ached france, no ober tt may ‘ensedd,epeialy I bth # and Fare efcentetimates, Ts the ample, the expected roquncy Tor 2 = 0 was determined vith Eq, 11 ath antlog of log (190) — 1.02459 log @2.11919) ot $y = C0870, vag the inal entry inthe fourth ela of Table 1. From p= LiN016 and y= 1 + p, B= LAL08/211915 = 528113 Ad ty Ba. i2, = 102460 % 39818 XCOASTO = 37.5000, oa = 220m4i9 x 526118 x 87.000/2 = 211011 aod ao on for sucrive alee of =. To avold rounding errors sc siguftant. figures were ‘red in tho cacao, although the were recoted ly to bo ‘lcnul place The Seal vale, for 2 ~ Sch, was obtained athe ‘ference between 160 and the wim of preeing The caesaton| Ud for the Sorepancy between te oberved and expected roqune (64.8) eshown in the le caluzn of Table 1, pong the requis {fre > So anto avid expotatiogs of lan thang = 3. The revuting P= 246 with thre depres ffeetom and P= 048 indicates prod Agrement withthe agave noni "The eomparizon of swerved sod expecta equeois by x" may be utred by chan iegulasios th individual eaten” Thos in| mf ‘MOMETRICS, JUNE 1089 testing agromment with Neyman's contagion distibtlon; Ball () eet! some is ee teen, tre erin campating 2%. This dliffealty can b aldo by towing the agement the cbaeved with tho expected socnd aad tind moment of the ‘peste bigoal.The relation of no tat to alter fre ‘dom, such athe logue, the dete log-normal ad the "ne ‘aia ous, haa ben cesiderod by Anveomta (2). The ve the forte advantage chat Chey take acon of the fe age vals which are mined by grouping total a oleerve dis. ‘sition i computing 2 ‘Two tes bave been doweibed by Anseombe (2), each having the form of Tetwoce an obneved and "erpected” monet, Although min each tv ie etinatod fcienty by fe vara has teen derive the ssumption lant an ficient etinate off snot salable for compating th expectation. A varance so derived may nt ‘poly when te expected moment cad by maximum oe, enc the terra and U are defined in tara oe etna for Thich the varices are knows.” Thos variance, however, shuld mays be computed withthe best avalaieetints a, url that ‘ervey sine Lelibod (0). "The dference T between tho third moment ofthe sumple snd it ‘alu peditod Iam Ui Sat to moment of the same sump of @ ‘ten binoeial in a i= Siste~ 9") = sue) — sesy2) +2842) ‘Te snieane ofthe dterence Tis determined by cmpaion with fig andar err, the gue rot of ta aan variance | ver = am + Dye + Oe) + MV 5) ‘The vavance of T shoud be compte wth extinaes of p, ¢ and ‘bated upon the mass Hethondk when tn knows, Alto the el tard me ye Steric me mri en he ‘Sine maximum ikeod ontimates ug + pm, the racine in By. 18 i then of dub applet ‘When the observed send momeat compared with ts eputation computed with iy, the diferace = ares a9) pecans mas poracoor = tare ges ase V0) = ant + pet = —acqEagmg)/ +n aD ¥ (hia defaed in Bq, 9 but computed with the maximum Heshoat tetiate thi anor, an ae the oe tem in Bx. 17. Here ‘en the expected scond morent, gm, can e estimated more exey ft te manu Mkthood vai for bat the appcaiity of the ‘asec By 17 i then in dob. "The ference betwen fhe cbeervd and expected moments a mush eter to compute than thir standard errs From (21) = TrRa7 (lq. 16) the prewent ample od an bmerved this! moment 15508 and an expected ale from the Bat bro memes and of 174, 0 tat T = L559 by Bg, 18. Since a varance by Ba. 18 seus4i272, th tandaed eer of T, 2.062, domed no dacepasey rom Tecate binomial, The corponding diSeence for the cod. ‘noment and ts errr has ben omputed by Ege 16 and 1, to obtain (Ganong 0.02, again greront wit the nagar bora “oil fr die Nee Binomat Wan the zune oil per ati paso tne in repeated counts eno be amu to have the ae expected value, Chey tay epremnt a miare of eveal homogenous Potaon disbutiane, The name im each walter ted to the iatges bu hit traf he expectations or means ns minture of Poiana, the sens represent.» poskive continuoas arate, ‘The ple fegueney dstsbuton which they might fallow {ide Balers dab rte Pots ype TI carve, an iin fact they reco dnd the obnevtonn wl conform tothe negative ‘iow ‘he expected frequency nay be known to wry within an ober lntuton. A cane in punt i tho dvsibution of bacterial camps (ream ln (AE lt to tring factors were valved In renaring fn O01 mle frie was laced co» mroncopic Hien spresd with a sede over an arn of ne square centimeter. ‘Baciera cnught by mutts tension the lower suiscn of the drop tenet th lin aon contact and thas increas the concent the of tactarin in tha aren of the fm, Seedy, the teh drop did rot hav the me thiknam ver the entre quae centimeter but doe tovrrace tension as thee i the senor than in the margins, 0 tal ore bactara mee depaed inthe center. Desi these to factors ak meeting ple bath standards ad o> low a hace cnt thet the dstsbton of bactral lamps per meronopie fd Sra seldom dings fram a Palmon.” However, when thewe ‘sowernucs, 10x. bred ston fatal pe (On) in in and ‘st arr me nt hasnt) a ape fees os ‘ed re seve Mo (Ban 9) ab om Seas eer) ‘Sonnet — | Rec ane ve . oer |e " 48 mre | = |e a ‘acter cout wa igh the dtcbain of cumpe per fd refed thin nom interop nd thn mas ena eet hizo ihe nm Taian of aia ogee ‘ola at low poston dei and wth the ep Bru af Sher ea sere i th i shal poplatos (5). Alas of rose in mrp outa ‘ea cbnrved by Seen i entng pest sla witha hnemocyt. ‘eer (Taal, ths se ot te nae Toe ted nih 3 eztv aia "Th gal ents te get Te 3 tat ‘ith the pected septive Bacmial equa at opal Sy ‘eam ton od how compute hy Nernst yp 1 contains dat, “The dtbton offset pe x so seldom efor tht mot sperimenta ise conta se adoneed and rept, Te Sid expernen ofthis 0 othe cor bor, ot ania wer ‘sranged in 15 randoniaed Moke (2) AU he endl the arn, ‘ict ilo on mre lad andr inet nd he aes ‘ecorde en enh “Thi experinent as ben repre ot in [SEIAIVE BINOMIAL DISTRIBUTION ws | HED HH tens ofthe total umber of borers per plot (25) and a frequency di ‘sition sowing foreach teen the oumber a ile with 2 = 1 2. bores (From aa salyas of variance o tho plot totals (i Toguithmie units) the lve of ore ingetation varied signieatiy fem block to blo (P 0). In cnanquene, the compte dis tebesine fom the 15 plots for each tretsent (Table 3) represen: ‘naga levels of infestation, Noptive binomials have Been Sad ‘epnmily to each of them. Since the expecta frequencies agreed well trib the sterved alas the data ar conaletent with the Hypothesis thst each represented the sum of woven! Poseon dstibationt of ‘una means "The ittons is Table might are from a difernt model {oc the ngutive nani in which the non-randomen x teed to "eatin, this ean,» rel ofthe iarvanhateleg (om es thit wee Ind a manos’ ‘Contagion, i fac, wat the bai for the ‘Neyman type 4 distribution developed originally for these ceevatons (26 a desrd In the next oeton, With «dierent mathomatl omtlation it lade ao tos egalive bso. "Contagion" wus one| [MBCATIVE BINOMIAL DIsTaIDUTION 0 of the explanations propomd by Biodent (8), who wrote, "I the ‘reonce of ene inieston! a divin ieencs the chance of other Indiv fing into that dvison, «nopatire bint! wl Shea, ‘but i it datas the ehanc, a paste boomin”. Ths explasation ‘fguredproanestly ithe sty of seident sate (10) ‘An erty eamplo othe dtibutog in Table of acsdente ex pevenood by 414 machinists i three months, whee the obeyed {quencies are atcha actor by chow computed witha ogative binomial. teach emchinit bad had the same iii probably of ‘bg involved i an acient but if thi probity were nord (or decreased) by ho having an accent, contagion woud be pnt and ‘negative nomial disribuion could rem Hlowever, tn oposite ‘inumption ads to exsclly the anne expected daiban, If Derencing at neident had no elect pon te risk of anther ace, but ifthe nvdual machin oe chet shape eterrle wit the ‘hres manth period diferd Inthe acide prec, egaive ‘inom would also rel Hene, the appearance of "sontagon not {nheret in mtare bot simply in car method of sampling” (11) ‘The relation ofthe to model mixed or compound Poss dition without contagion and contagion which changes Use ode of further vents is dicomed ably inthe rceat monograph by Artou and Kersh (Gy. Allenative dstzbtons have been developed from eter mth euticldeisitons of “contagion” an wil be considered io Ue next ‘The aepsve binomial may result fom « diferest but elated sola In sous of wll bacterin, Jones and Mollon (18) rocrdad far ack misreseopie Geld oth te numberof bacterial colar and ‘ho numberof baci in ach coloy. ‘Te numberof ealonee per fl ‘god wel withthe Paso expectation for a random Gstibtion and the sumber of bacteria pr olny inthe same conta with Faber’ Jogeithnic dtc." Under thee condition the expe dist ‘nt of buctaria per ald a negative binomial (28, ae ont ‘by he agreement of the expected sd absrved feeqaensie (Table) ‘Guna counts i plan eslgy which departed from Poisson have te ited tote osueaee af plate "clump Blackma (7) not that Prinulaewviula oprofuces vegetatively by shot sisomen 0 tat er individual are een suousdod by yosoger pants 19 sn ol-eld community (0) both Laas aera aad Laspars copia fended to oer im chinps The daribations of eamge per quads ade plans per lump have not ben pote mparataty, 90 that the ‘otal cannot be toed dei However, the dbuton of plats ‘ee quudiat in all tree came apo excellenly mith the negative ‘Soni (Tadle 8),[BEAMIVE BINOMIAL DIETRIBUTION mt jfte esese sees albeit av tn deal foie neve inal 2) and i no rhaon to mippme tnt the pean hare on ahaa ‘econ apg sem fr taper sin wha C2) ictoertndefeshetn doin of itt Ele hing th bet nthoppe @) sod the wroworm (I), teks oo Splane (62, whee Eguatin 8 gave fy = a0 and Pypation 6 rok = 208, Bren nuh ihe etna i ot ie peal, di the x tnt forthe agrument ofthe oberved end expected frequencies {20m th oo eatinaten gave P= O17 aad P = 25 respectively. Sally seo method fr ummariing ast of chervations wth two stains, oor of them the ian, the negative noma should be of Incasng Interest to bilge" An adaguate fof thie diatbaton {0 sbe data may serve to July freer station analysis sachs teen suming (2) or ranaormatin for sali the variance ‘rpaalry Lote analy lance. But sac weer quit diferent olla ight pom undeledata which conform tothe negative ‘Siena ne cannot Ui agrecment as th sla bal fr jotting ‘parca modal or eoslusios bard upan i, ‘Campari wih Other Dirt Jor Over Dioperion. Although the negative binomial I the eet to compute aad the most wily sppable of the dtbutins fr overdoprin, several others have ‘ee propoed. Some of thee have tro or more med, il th nee tive nomi ony n single mode." In Sting the egative hci) Aa sured dtutons ay “exes” modes sre amuned to representee RES TE 12 ‘BOMETRICS JU 083 don vt ns gh ee ote atin Top ratte ae ay ae Semis ee td pang Pe se nmpineoed iter, [he Se thm toa de a a et SS mayer toe ine gene Sa ethene Paden ease ‘mathematical modal of biological interest al Te Nefua nage Son Stn (2) ume i agri ia ap ye pela ea Scie rrr ane opal spec ae sa Teaneck he ca Seton nsdn oe Seat same Sane ds deeb ane shown in Table 3. For treatment 2 they reproduced the observations SEE hae gute cane og, ene ers Shas eee ies Salerro Sauce ree ee seins cones autciee Saat ye, ae? SA Pear “in Tomar dle Pinon dbtin (2 es pe slo ih pnts tht ar omen mete dons a Se Noman type enn nab 3 Sas lows ito ace tae ai Si nd hth ners ae Tater Tw niin pa Psa St "hich fequenty fall tothe gro with the sec sl he eared (Q. reauencio eumputd Ky mosats for this and other sven wth ‘he Thomas distribution are gen in Tables § and. They domly [NEOATIVE BINOMIAL DISTRIBUTION 19 ‘rumble the expectations forthe Neyman type A snd have snilar ‘vantage sd tations ‘der certain conditions, the Palys dtibtion (2) may have tro saador with meri lane rence a 2 than ia the nope ‘oad, which in other repeta fe claly pra Under eran endiins, the Palys distribution might repron! the number of In

individual, ithe exten of let Nght traps fot ‘caumpl (8). If ina negative binomial approaches ar and ot the aro clu, the obered fequencie can be considered sample ‘ran logartiae date. Wills (3) ba tnd the lgartimic Aiibsion 1 Banto'e data oa the nue of Hinda male penoess ‘ina oath Indian jl with 2 = 1,» oe per head, omitting the 2 indidule (we Ansorbe's eoreton, 2) nko were fee of le. ‘The cheered fruencios agreed for better with th logarithmic «x= pecatans than with the sempated frm the negative bina faethed I. Tn this intaneo, however, method I hasan ficiency of ae {han BD pecent aod sethod 2 and eficeney of meaty 08 Perce. ‘When evompated ith = 0.9057 6524 and fy ~ 144108 08105, the dvergene between the expetd and oberved faquancis Table 8) ‘vas wall within the eumplng sor ("= 3L.74,m = 2). Neverthe: Te the sbaeved scons third momenta were sintean'yIrger ‘tha ir expctationy, wilh U = 2409 2 889 and P= D048 208, Yat ws signify larger than ser, ite expected value for a lopeitnie dation. ‘eeminaton of the data showed that four of the 1,78 prisoners tad more than 200 hed ise. When fy was recmputed without these Individual the seen and hid moments ee no lenge dncrepnt, with U= 2288 4 2568and 7 = 575-27. ‘Thiserample indnstes {hedlapropatonate eet upon Pand U ofa very few urge vals of ‘Wh thew omiosows, the expected frequen aed tl more lair104 [BIOMETRICS ZONE 82 Samitpreenees ct remorse SEEEREEEL with th oberve vals, an evidenced by the exciton for 2 = 0 1 10in TableS. The apeement of Ue sepative nomial existateen ‘with tho obsrvd fequesle of oitee on individual rat Table 6 {id of tks pn sogp a ven by Faber (1) met greater fue fet the negtive binomial thin for he logan debutant dies on epacsion iting a Single to Saal Nepate Biscnial Dierbuions. The salyeis and interpretation of mast Blgial data re falta by Habiity i the vane, ao tha oly meats need fo be compared. A Sar #abiity in the cooficent F would bo laren te uly ofthe negative bison and increase our confidence it utbiiey for given problem. The asumption of stablty is eset in some sin th quent sumpling ef whieh for apoworm cat (2), ‘By dapllerunson of thn maximum Ihelhoed elon, combined gn be eapputed from slr of vtribations sod their hmogenity in pect tet by 2 "Th calyltion cows of computing the sor fo each expats Asbo rth the mane tril vale of Wand ang the sores for tach ¥ overall eoapocentdatstatons Til values ar lested wat {oo mums of the sores, (0), re obtained hich hy bracket Dy lnterpalaton ttre the, the requited eta‘ Ee that for whisk 'S(a)= 0. Tf See sus are deigaated a and fom carcepondng NBIATIVE mINopTAL DIETRIAUTION 195 ‘calves f Band Ei, the ror variance of may be computed senna 10 ‘he homogmnlty ofthe Pin the component distributions depends on the raed in each indo serie fr and Bt 'The tio Hi = 29/4 — 0) cy is computed frm ach compaoent itebution and from the totals over hin td es ed eo tin nt Tea pee er + | ata ca sos ager at nae Mien «4004 789 = 0/6 + 488 = HOR sll dnibtion. ‘The mm of he rato nm frm the 9 nda ‘strtions for Ky and, dni by te ratio compute from the axrspoding tata and, The diferenoe infor tenting the lomogusliy of E vith 9 ~ 1 ogre o from. sling this ‘expos, ie mmsteral which boundary Yai of Ue, ‘Sheth ee x bland with ther one “oe procltze can bo Masteatal with the for ditribatons of ‘cen bore in Table roprescting four dierent treatment in he mane196 moMETRICS, x8 8 fied. ‘The elelation requis foreach teem the mean # and the accumulated egies 4 thw inthe third eta Table Starting with ail male of = 15, te cleans a ‘Table 0 gave (0) = 012, that a snaler al rae = a, a sled next, giving 5) = 25872 By interpuation bemead Mand hi for S() 0, = 147 and fom the teuing S() by farther interpolation, Ki = 1.473. Interpolation betwen Bead K gre Ue ‘maximum iad ctinae E-Tetng the hemogenely of Foe ‘he fur tetment requred the rai in hela samt Tabs from whith x! = 1894 with tte degen of feed. We cred Wat the Hs Mor the four tretnente Gl net der sglBcaty sod ‘ould be reread by dag aie of fe 1.4718 1737 ‘Summary In analyzing boll count fr which the variance is igicantyIrger than the ma, the value ofthe megaive noma tribution in eae ty a vinplfed asin Hteliocd meted or extinatng the coatcist ‘Th cleiaton trad edt sith mimi! explo and camped wil ter erates the ‘ume parameter. The eer variance othe stat ofthe negating binomial and ents for fr agreement with swt of oberenins ae Mhstrated withthe time example Models underlying the negate ‘inom ae reviewed with reference to buervel dation oh Blunts and anal. compara with othr ditions fer over Aiperson sugrst thet the nepsive binomial the tel dey eaotable end generally wal f thw that have bron propre oo fag Oe maxim lool atination ete to ell ion of sgl coefficient # tom x sere of ilar dstrbuton ad the eting of his homegeety by Nove ox sm Eericore Frese oF me ‘Nreanvs Bivoxat ‘When iti dud to examine the reproventation of data having, touate of far vats of¢ Tram Oper, by mens of the negative lini ditt, a wich tha epocaton of av @te= Fa) NED Ga {scored in tems of wo parameters pan fyi well own that ‘he auntion of etination Bad on the tens phot efit. ‘cand equation, with ficiency varying with tho circumstances, ‘may be taken fm the cond mort or ratnee, vot Dea? ung other may, en dhe agua a eros Cpt = Ne 1s 10, the author gave a number of rules (12, p. 188) fr judging whe th rt of thee aden eceny, and in 180 (2), Annee Ina xmined mor fully the eonditons of ecency af both of these porches. Many, however, wl ish tom thee rethods only ar 8 {nt te toward fly econ Sting, andthe procure for lng ‘his, whatever meas aed fra ft entation, perhape wore setting ont. ‘ficient sering for Prom the pinay expectation POOR co eae SS ata ‘ve hive onng natural logarithm teoughout) slits aren) +P