SAMPLING THEORY Onin || WITHAPPLICATIONS 1 «PAL SUKHATNE “BY. SUKHATAE #5. SUKHATINESAMPLING THEORY OF SURVEYS WITH APPLICATIONS SHASHIKALA SUKHATME, Pn.0, G.ASOK, Pro. IOWA STATE UNIVERSITY PRESS AMES (OWA USA) ano INDIAN SOCIETY OF AGRICULTURAL STATISTICS NEW DELM-1100%2 tha){© 1954by he nan Gocay of Ages Satis, Now Dat rate lows Stato Univers ron, Aros fo, U3, © s270by Pendurana V. Sukh and BaleetnaV.Sukatne [Ak hts served Prada the USA © 1964 by Pandurng V. Suthstmo, Balrghna V. kha ‘Shabtlatesukhoineand © Asck AU ign Besorvea ‘rst Printed in 196 bythe Indian Socey of Acute! Stats, ‘ict Bat nc wd te fom Stat Uivray Press, 8 3. ‘cand Reprinted in 197 by Inn Solty of Agricul Statist, ‘Second Pine, 1076 Serena Prag, 197 Laseropesat and pete: PREFACE TO THE THIRD EDITION 1 nat hink | would be abo to wie ary more wh my hand, but ae is Teller wil show, 4 can type and | am now euticlenty Fecoveed to fel conden that wil beable to complta tha revision ‘ol our book In anoter fw mons” ‘Ths is how Bal (LV, Sulssma wrotto me on Ap 23,1070 fal passed anay within a wook ator wing tis. ACualy, | had fushod from india to 0 him at Mayo Cini, Rochester. He fecognised my voce end was 20 ll of emetons that tho doctor ircharge of ie Emergoney Room was forced to warn him vest Imovornon twas ny afr I etured to Ina that 1 got he eter oor 23. ‘That Bal was working hard on the tid Eon was known 10 his students and 10 his coloeguss, parteulaly Dr. HA. David Director, Statateal Laboratory, lowa Slato Unversity, Am, Iowa. al had spent a month with Prot VP. Godanbe, dscussing wan tim the foundations of the Sarging Theory of Suveys ond thng to tink how fo go abou In vellocing tam in tho new eon fad yet mamntaining estontaly the same charactor as In He old Edifons. Ba oven hegaa Tasching modied course In samping thoory of surveys and as he taught, bo oompleed the maruscrDt Colts rat ve Chagters, Ho was Busy aling tho txt of Ne St fnd 7h Chapters, whan ne was takon i in December, 1078 Normaly, 1 woud have had ie culty in completing the revslon of tho book, bit | have boon cut of euch withthe subject {oo long, Bats dosh also reed for Dr. HAA David the queston ‘ol making temporary srrangomodt to gude Ph.D. studots and fing hs course of leurs, pandng new appointment in his jBaoe Iwas agreed that shoud depute Dr. AGO, Bas stent, {ote in these tas. Howover, the loa of work proved too heavy Tor De. Asok to complete tho revision of the book. Meare, Shashkala was fing hae way in her onoly tf. ‘as sho Dogan puting together Bas les, {found Ree gradually Interested in ho subject of sampling surveys. | suggested to er that she night ty her hand fo complet the jo which Bal latSAMPLING THEORY OF SURVEYS WITH APPLICATIONS| SAMPLING THEORY OF SURVEYS | WITH APPLICATIONS PANDURANG V. SUKHATME, Ph, D.Se. (Late) BALKRISHIA V, SUKHATME, Ph. 'SHASHIKALA SUKHATME, P.O, ©. A80K, eno. IOWA STATE UNIVERSITY PRESS ES, OWA USA) AND. INDIAN SOCIETY OF AGRICULTURAL STATISTICS NEW DELM-110012 ta)© 1970 by Pending V. Suhatms and Balirishna V, Suknatne AV Fights Reserved. Pines USA © s904byPanaurang V. Sutetns, Balsa V.Sukhstmo, st Pint in 1998 the ncn Society of Agricultura States, [Now Bah naa andthe low Stat Univraty Press fon U.S.A ecard Reprinted in 1897 by Indian Soy o Agricul Sates, The, 1950 Sedona Raed Eton, 1970 PREFACE TO THE THIRD EDITION “Lo not hink 1 would be ae to write any more wth my hand but as is Tater wil show, (can Type and | am now sulin ‘covered fo fool confident that wil bo able to compte tha raven tt ur book In ananer few mon “This show Bal (V. Suknatme) wrote to me on Api 23, 197 Bal passed away within & wook ator wring tis, Away, | had fused trom Inga to, see tum at Mayo. lino, chester He Focognised my voice and was so full of emotions thatthe docor Incharge ofthe Emerganey Room was forced o warn hr to vest movement. was ony afr I retured to Ind tna got he leer oh Apt 25 ‘That Gal was working had on the tid Eon was Keown 10 his students and 40 he coleaquos, partoully Or HLA. Dad, Director, Statseal Laboratory, 1owa Slate University, Ams, Iowa Bal had epant @ mont wih Pro. V.P. Godamba, dscusing wit Rim the foundations of te 'Samping Theory of Surveys and trying to tink how fo go about in eloctng thom inthe new elton and yet mamntaning essentialy the same characor 23 In he ott Eaiiors.'Ba oven began tosching # modtied coures In sampling ‘hoor of suneys and as he Ugh. No compet the manvacret oftho fret ve Chaptrs. Ho was busy tinalong tho txt ot ho 9 fad 7 Chapt, when ho wa token i in December, 1978 Normal, | would have had ite ditiosy in complain the revision ofthe book, but have bean outa ouch wih be Sect to fang. Ga"e doa also raed for Or. HA. Davi the quaston 0} making temporary anangoment to gukle PhD. students. and ‘completing fis cous of lectures, panding new appotment In Nis pace. twas agreed that shoud dopula Dt C.Aaoh, Bas ten, {orp in thes tacks. Howover, ho load ol werk proved foo heavy for Or. Acok to complete to toiion of the book Meanwhile, Shashikala was fang her way inher lonely We, As she began puting togathr Bal note, | fond hot gras) intrested in tho sibjstof sarong sues. | euagosted to her ‘that she might ty er hand to complete th Job wich Bal lt iheAnd to my delight: he suggestion worked. Nght and day she worked en the manusorpt aa competed tne revison by May 198 tr Aol on ha part wert trough te Chapters and sated eel that the revision wa well dona A is was going on, | was tying wit the Kea to rer the manuscript somo weltxnownsuthorty onto subject. | fed feta to gat anyone sullen senior to spare tho be 10 lock trough tne manuocept. | wanted re book to be pubsshed in tho yobt names of 8st Shashi and Aco, but De. David ad not the ine to orop ol. Fina, i was dociod that weal should bo shown co-adhors, Azo assosiaton ae coauthor has the sditona fevantage of hong assurance tet he weuld bo able f0 help fora rovisons ae Sod whan nocoseay ‘The one satiation | hav is at this rovison hapa Shashi to rehabataharelfeoonar than | had thought. Se Isa lady of tremendous wil power an once sho takes thing in hard, she ‘wil never leave ntl she compat “To Or, HA, David! am most grate for al he asistance ae encouragement he gave. |oo| gabled to he sacretaral stat for tho enormous ok they pa in fyplng. | would oko 10 ox0r055 my tharke to Ne Nort. Baley, Distr of the fowa Stato Universty Press for ne cortnuod hen and ercauragemont inthe pubilaton of te book Iam also tote to Bas students, parolary Shean Biya, Ccrand tha, James Drow, ick Aver and Ese Boo), or roeing pats of the manuscipt 1 am lo grat to Or. Prom Narn, Secretary and Ors BAPE. Goal and Shivar Singh, doin Sevetares of He Indian Secity of Agreviural Sates ov thee conus inert in the revision of the book and or thar hap in reading tough poo. The scton has become uc lagor than would have Hed to see, To include al ie pineal dovlopmant in thoory, stato thom with exampics snd exereses and yot ot allow the size to Inorease, proved too large a tack fr vs, Ony Ba amongst vs could hve atmpted Kft gully Wat Ihave not boon able 1 play ‘mere ative par icetly Rope that new odo wil conte fo moet the moods of scans Ind Ad saad. PL, Sucre CONTENTS. LL INTRODUCTION AND BASIC CONCEPTS 1.1. Sampiing Concopt 12 Why Sarele? 413 Samo Us0s of Sampling 44 Planning and Organization of a Suey 1.5. Sampling Desk 116. Probably Sampling Procedure 18 Properioe of Estimators 1.9 Samping Srateay 140. Cost Considerations in te Choice of a Sampling Statogy Frotrences ‘Append 1Tablo of Random Nombors 1. SMPLE RANDOM SAMPLING WrHOUT REPLACEMENT 2.4 Iosucion 2.2 Some Properties of Simple Random Saining Withoet Roplanerent 29 Nolaton 2.4 Unbisacnase of the Sample Moan and the Sample Moan Square 25 Variance of the Sample Mean 7 2.6. Estimation of the Standard Error of 7 2.7 Goaiidence Lents fr ¥ 2.8 Some Optimal Proparios of the Sample Mean 29 Estimation of the Sample Size 210. Estmaton of Proportions 2.41 Confoence Limits and Sample Size fr Estimating Proportions 2.12. Estimation of Propartons In Domains a 2 gags542 Mutaraty Rato-Type Estimators 5.19. Estimation in Stated Sampling Fotrences ‘Append i! Expected Vatues of Conain ‘ener Ordor Proguct Moments REGRESSION METHODS OF ESTIMATION 4 Inodeton 62. Tho Regression Estimator 63. Bias and Mean Square Regression Estimate 4 Elfeloney of me Rogression Estimator 8.5 Optimatly of the Rogeesson Estimator undar Super Population Probabty Model {86 Double Samping + X not known 187 Sanping on Two or More Occasions 88 Extnaion in Skatiod Sampling 8.9 Muthacato Regression-Type Estimators CHOICE OF SxMPLING UNIT ror ofthe 7A Chuter Samping 72 Estaton ol te Population Mean 7. Gtiiency of Clustr Sampling 74 Estimation ftom tha Sample of the Etiiony fof Custer Sampling 75 Resatonsip between the Varionoe of the mean of & Siglo Cluster ands Size 78 Optimal Unt of Sampling and Multipurpose Sunes 12 Use of Supplementary Information Ia Imgroung th, Eetency of Custer Sampsing 78 Unequal Chater + Estimation of the Population Mean 7.9 Unequal Custos : Comparson of Sampling Strategies Under @ SuperPopuktion Moos 225 en 238 2a 248 240 267 278 309 Everises Rtoroncos SUBSAMPLING 8.1 Invodcion 82 Two-Stage Samping, Equal Fest Stage Unis = Estmaton ol the Population ean 89 Alocaton ot Sample 10 the Two Slages Equal Fest Staga Unis 184 Conparzon of Two Stiga wth Ono Stage Saming. 185 Etec! of Chango in See of Frs-Staga Unie fn be Variance 85 Tivee-Siage Sanping, Equal Fret Stage an Second-Stage Unis 87 Alocation of to Sample tothe Three Stages 88 Two-Stage Samsing, Unogus! Fist Stage Unite 89. TwoStage Samping, Unoquat Fist Stage Unis locaton of Sample 8.10 Troe Stage Sampling, Unequal Fst and Second Sage Unis 8.15 Sratted Sub-Sampiog 8:12 Optimum Alocaton in Stated Suo-Sampting 1813. Elion of Skatfcaton in Sub-Samphing Fsoroncos 'SUB-SAMPLING (Contrued) 2.4 tnvoaueton 2 timation of tho Population Mean 23 Allocaon of Sample 9.4 Detomination of Optimum Probabies 85 Relative Eficency of ho Two Sib-Samping Designs 8. Sub-Samping without Replcomont 87 Statitation ord Gain due to 298 Sat-vebgtting Designs 310 at 312 a6 ata 389 02 268 90 370 370 3702.9. SubSamping wth Varying Probabiltos of Selection at Each Stago Sampling wih Replacement 8.10. Sampling Witiout Repacoment at Each stage Foteroncos SYSTEMATIC SAMPLING = Part A—Ereors in Surveys Houten 112 Types of Eos Part B-Observatonal Errors 11.3 Mathomatoa! Made! for sho Measurement of Obeowatons Enos 114 Tho Sample Mean and ts Variance ent Compenents of 145 Estimation of te 0 Vasance 420 295 440 a8 456 1115 Svatited Samping in which Enumecsors ave Assigned be Unis In Wel Respective Sala Unstatfed Sanping in wien Enumerators are Assigned Neighbouring Unis, Deteminaton ofthe Optimum Number of Some Studs on Enumerator Vaiabity and Respondket 68 tations of the Mahod of Repeated Sanpies in Surveys Part Incomplete Samples Hansen and Hurts Teetrigue Doming's Modt Foltz and Sinrons Techiaue Randomizod Response Techniques Part D-—Some Experiences In Sample Censuses of Agriculture Practical Difctis of the “Open Segment Concent Response Enors Folatve Magnitude of Sampling and Brera Fowroneus Ina 485 406 51 su 525(CHAPTER - INTRODUCTION AND BASIC CONCEPTS 11 Sawuns Concer Samgling, which is the selecton of pat of an aggregate to represent the woes Requsrty usod in everyday Mle al kinds ‘hiiwestgatone. Amos nstncvly, bere decd to buy a bt Wwe examine a Tow aries preferaby Rom siferent pats ote ot ‘Ante example ls prowdod by a hanctl of gran taken or Sacks dotomine the quay o! he gain Those ae examples are inference is drawn on the basis ofthe reas obtabed rom a samp, ‘Sampling is used most kequent in surveys, The purpose of 8 sample survey is 19 obtain inomation aboot popu. By opulabor’ we undorstand a group of unite delned according the aims and objects of tha survey. Ths te population may const of a tho folds unre speciod cop ab Ina and yd survey, fhe agioatual holdegs lamer than a spectieg sea, agseutural suveys, or athe houssholds having tur or more ‘hide asin soce-sconomle sways Of eouree, te popu ay tbo refer to human beings af 9 whole pepuistion of Urry Fa paricular sector af to county. Tho hlermaton that we aed bout te population fs usualy ‘he taal numberof unts such 36 the numberof fame in a Stato roming corn, aggregate values Of the varus Charactenstes suon a the toll ateage tinder corn, ‘voragos ofthe varous characteristics per unt such a8 the average ‘28 ofa household ae proporians of wits possasing speciiog atrbutes such a the proprio of houshelds having Income above fr cetin love Its clar that it harly mars how wa crew a sample, provided te population is nomogeneaus with respect tothe neces woe sud. boweve the unis consttng te populason ary onsieraiy in respect ol ta charactors undo ay, the method by whic we daw @ samp pays a crea aa end fe macnesary{0 pay due ation to the sslecion of te sample and dawing ‘niorence tater, 12 wy Saeue 2 A sampling method isa scone and abjecbve procedure of selecing nis from 2 population and proves a sample tat. expected to be reposeniatve of tho populaton as @ whol. asa roves procodures for tho ecbmaton of recs tht woul bo ‘obtained I's comparable suvey was taken on al the Unt nthe opulabon No mer how good samping mahod is used o Wa 5 sample, ts clear that a Sampo can aver reproduce onacty te ‘various characoiscs ofthe population unless the popula Zsa '5 taken as the sample and a census (a complste srumeraton ot {he population) is cartod out The resulting dsrepancies between the sample osimatos. end the population value that Mode Be obtained by enumerating al the ure inthe population tho samme manner in which the samo i enumerated, aa tamed samping nore. Their average magitide wi netialy ‘epond an the opuiton under study, the 3260! ta sample Ne manna a ich the sample is raw nd tha mebiod of atintcn, However, the fact hat eamping errs carmot be avoided sould nat bo dhrbn, 85 long a5 these erors are suticiety smal so that ‘he valty ot ‘he resuts obtained and the conchisone drawn ae not aectod Because of ited resources by way of tina, rained personne fand money, generally nt fase to allt dla tom more ‘han a fraction of @ popuaton. In @ sample survey, the data can be collected and analyzed wih much greator sped se the routs Made avaiahie at reatwaly sho notice wtich would not be possble 2 the case of consis. In certan Investigations, may be essential ‘0 use specialeed equpmen or Nighy rained ft stat for ata caleeton, mating It mpossble te eay ut such fvestiatons xcept 09 a samping basis. For the tame reason, Wis posable 1b emp effin supervisory tat and exercise greater siaviion over tha field wrk w ens conpstanses of rots end telaly othe data clecte. exors eter hen samping errs such those lisa though nonsesponee, Incamplelnoss and Inaccuacy of retuns are termod. nonsampling emors and are fly to be more widespread and inpotare in a concue than in a sample survey. Moreover, ter is no way fo assess nega era nice samole sunys where tis posable 1 aesess the tverage magnitude of the sampling ror. Samping methods ago provi tn ean ting in sdvance tho detals of the survey sign, such ey Me ‘procedure for slang te sare and tor choosing the sample za, f sul a manner tat wit a preasigned probably the average ‘magnitude of te sampling errs doesnot excood th spaced tink. in othe words, samping methods ena us to contol he procion| of samplo estimates witin ms Yaad lo advance, 13. Sowe Uses oF Sinn Sampling canbe used w avait of way. Howover, tis marty used inal ins of suroys all var the weil Depending upon ths ‘bjctves ofthe survey and te purpose for which tha data may be’ used, samp surveys. can be broadly classiod ovo tree categories: Geseigtive,anayocal or bh doscpve sn analytic In cescrpive surveys, the object usualy to obtan some descrpive measuies wit rospoct fo ha characters of ho ens Dopulaton under stay. Such suvoye ara very common and fe required for national planing ar seco-ooonomie development. fo colect data on agrcstal production and wlizton ef land ond natr resources, rust peodscton, unemployment 320 of labor fres, wholesale ar atl pices, como and expense per household, umber of Iteato parsons and schoe going chien ae Sso.n. On the ator tand he sbject in analy survey so blah descriptive infomation toe ieent sia groups ofthe population der fo test hypoheses concerning possi rexatonshipe between te subgroups, For exampl, in bor foo curvye one woud be interested not only knowing te avecago numberof hours wood er day end the wages pat but also whether men werk longer hous {tan women and whether fay resava Nghor wages than women fort eame type of wor, Sampling metods aro aso used ln population caneu. In fc except for coran basis tlormaton rogared in respect of evry indiveval, data on various toms such As occupation, parentage, mariage ety, come, migraton, hausing, fe colacted on a samping basis” 'Samplng moods are used t0 provide counter checks and sped up tabulation and pubiaton of rout, Sampling methods ave used extensively in tusiness and industry to ineoase aperatonal efisoncy. Thay play tt inportant rol problems encountorod in market research auch a2 ‘esimating tho Size of readerp of news-magazines ax newspapers or hcing the reactons of consumers to now produc rary intoducod Ine market. Tey ae also used 6 ascertain the epinions or atesrn ‘of he public to cerain issues in which they are interested. Surveys ‘arlod outer such purposes are fen termed pion Pol surveys Samping 19 also used wesly in puro experimental ‘nvastigatone a inthe determination ofthe Quay of rior the rexponee offerte to dilerent cops or Be cnemlcal compostion ‘of ais. 14 Pusnans sno ORGANZATIN OF A SURVEY Planing and organization ofa survey f not es imple a0 it looks and several prblame. area. Nataly, tose preter vary ‘teat depending upan fo nature ofthe lovesigabon, te typ ot {ata be eallected andthe condons prevalent inte region where the survey 16 to be cated out Some ofthe questons that need {0 be cally conedored ara braly dscwseed bol Objectives: Wie mest important 10 spocly as prectoly a5 possible the main objecives tthe survey and the purposes for ‘wach the tess of survey aro aly to bo sod. Unloss tis done to the eeeacton of the user ofthe Gata and ho saison Ft ‘urge oF the survey, Hu fy of the survey resuts may not be soto ‘ata tobe Gollctad Whe ti ipartant 1 cll al the' data tht ar rlvare to and consietert wh sho objectives oto suey itis naceeary fo ensure tht nly the Gata which ato Healy to Bo lnzed in subsequent analysis aro eoacted. Colecion of data about ome not oy to bo lized Subsequent, resuts not only Uimooesery waste of resoutcss but may aso Inarosty act the tualy of deta'on oar seca Heme ve Yo lack of adoquate oowiaton Under Investigation: Once the objectives of the survey andthe deus of daa tobe colectnd ae spc, te S008 ft the sampling vestigation 's more ° fis. dened. However acta! corseatons and convenence may suagest mading be Scope ofthe suveye. In eter word, the sampiod populabon may fot necessanly eines with tho poptaton about whi inormaton FE acoted. Tus, in sanping a popu ot hotangs for to purpose land uizaien suvoys, practical conscoraons may sugges he ‘xcusion of sal Hokngs rom the sample. Cleary, tho sampled Popuaton consists ony of to larger holdings. n such a shut, 111s necessary to deline a nokkng in an unambiquoss manner so that Me enumerator can immed deco whothor of not a parila hoting belenge to to population, inlay, in essmating the average expenature on food per househol, wl be necessary to date prectsly @ houseneld So tat the enumerator can ded fn the spot whathor or nota partcular hovsshold belongs to the Pepataten Clos Ia al these cases the concusions erawn tom {he sample wil apply 10 te sampled population. Additonal data would bo required fo decde whother these conciusons. ae also lpplcable othe population about which infomation is dented ‘Frame: For the purpose of solocting a sample, tis necessary that fe population #0 be sampled can be sued no afta umber of thet and Wertiiabla nts cad sampling te llevan forte purpose of sampling wat the samping uns are ‘Thay may be natural ns suchas inchs in a human popstaon| cr elds ina eop save, natal aggregates of such uni fomaiee oF vilogen, of aftia unis auch aa. ow of plans, of 4 plot of spaced sie in sampling a fed What fs essatal for the purpose ot sampling fa Ist ot al samiog nts in tho population. Such & it ealed @ tame and proves tho basis or tho seocon and enieaon of te unts inte sample. Exampos ‘fa Famo aro a fet of farms, and a tof sutablo ara sogment, Ike vilages In tna or counes inte Une Stas. The Vilage for county form's the camping unt and provides the mean for urther Seocion of a sample of arms, tlds and pos, The aval of @ sulle frame is otton ane of the major problems. 1 avaible, the Fame may be compote oF ot of date fr eomah diplosis’Intas ie tho ease, evory attempt anoutd be made to rove th delcts before using tha Ware fr te purpose ot sobcang a sample Motiods of Coecting Data: Even i the metiogs of cotecting data dopond to largo extent on tha tevostgaton undor study and ‘he typo of data 10 be collectd, thore are otlen several ways in wich th data can bo colecled. For exam, dala in togard 10 land uttesion and tering pracices can be obtained trough a aio questionaire, a persona nendow oft tamer, or ough ‘bjeciva mossuremont ard sbicd intron. Obvust, cary ata Brough mated questonnare is least expensive, However, te espance may not always be salary depending upon’ te cooperation of te fanner, he pe of questonnae and te design ‘ofthe questionnaire, n developing counties wher a larg proportion ‘ofthe popuation i iterate, tho meted of maled questonnare may fot even bo teacble, The menod ol colony daa. Trough Interviewing depends hay on the avadabty of well tainedthe most rasa, may ofan prove expansive At thoes factor wil have to be carolity considered below slectng a method fr colocing deta Organization and Superson of Field Work: The success of «a suey depends 10 geal extort on how wel he fed wark Sranized and superieed. The ted satan the superiors shoud be tamiar wth fe objaciver of the survey, ho ipo of esta to bo calbctod and oor mato eatnactad win the organization of tho avy. Thay shoud be gen tensive rlnng the motods to be used for calocing data‘and also inthe metod to be used n caso of ron vesponse. Arangements shouldbe mage supense {he fold work, pretraby on a random base, a ode fo check tho ‘ualty of he data collected, Precision: I is necessary to specy the deyrco of precision ‘ested wit spect oto varus populaton vais tobe estimated No mater now wou a turvay fs organzed and superaod, tw nt yl dated ress nos propery cred Sty 15. Sawoun Desi of finite population U consists of N csinguishatie units label (1,2, M.A sample =” om Uf an ordered sequence of fatale fom and may be reptenertd ae hore i, denotes the label of the unt erawn at the eraw, and 1 oto tat Ecdhar-ednaer =o en used) « Sar -edh me) andre |_Mary 108 di 0) itm etnatr@ is ebsasud for 0,8 masnsque err: of 8 reduces to the variance of and MSE (3) = Var (@). ttle now clear that # we have two estimators 6, and &,o 8, would be prlerted to bits mean square enor fever exceeds that off, This leads ts aainten 18 fs areey on or above teh Dotgton 10: Lt, ad, be wo eras oa pat othe 8 sc 1 be eto ter any H a oy seth « MSEC OF, Yo) wis ney oe net Yo-Yo wea Gow a eis, Datin 17: exit of nt 0 Ds estate sa ass Col cement omy ner nan vey ‘ter estimator in ass, whle wl be sad fo bo admssbie {Fite class does not contin & beter one, 19 Supune Sreatesy So far, wo havo dscossod various consideration involved in the choice of an etinator& to estimate a paramatr 0 based on 2° probabaty “sample oblaned “by “useg GosKn OS, P) OF DAs", #). Since, nowover, a samping design can be oblaned! in varity of waje, ht toome more nalual 18 consider @ samy ttsign 01S, Pr together with an eximator Bas a eampling sre) 140.) = HIS, P.0) 1 estimate a parameter 0. Consieratons Similar to those used nth cote of en etmstor can now bo ead to choose a sampling sbatgy ‘A sampling statogy 40.) Issa to be an unbiased strategy for estimating # 4 an unbiased estinator of & ctownse tS sai io be a biased strategy. A samping statgy Hy, sald Yo De Uanormyy boter tan another camping stategy H, cost being ‘amore, iho estimate 8 crrepaneng to strategy Ke unitary bot than the estimator A eonesponaing to stategy H,. Consider now a cass Gof samplog sates for esbmatag a parameter 8. samping sratagy H(0,8) is said to be the Gest strategy in tho ols of C, of tation H tho eatgy Ht unomly beter than every ether satogy in class C, and Wt wil bo said 10 be adniceble the clase does nol contain @ bor member 10 Cosr CONSDERATONS NW THE CHOKE OF A SANPLNG in tho previous section, we have discussed some of the consieratons imoived nthe choice of & samping Svategy WomEEE EEE en 8 lass of svaleges avalablo to the statin regards ofthe oa invvad, to cost varies conser fom satay to slegy, 180 Considerations ae clos nt aqua Kor ho pe of eae fa practeingstatscan ana it wl be necaseay to pay ue atenton| {o to cost volved in the choice ofa eatogy. A cost exon which i sinpie and reasonabi for mostof the practoal shes, may be voprocertd 86 Ga arene) (20) comprising two components. The fest component isthe oveead ost othe survey, while the second eompanentpropertonal tote fective sample suis te cost of ollacing ta, with, Dog the cost per unt. Sinco the tal cost C, varies fom sample to Sample, we may take ts expected value stl cosider for a sate HOM expocted value Cie a avn oe visas) (20) the total cost of he survey cannot exceed the pre assigned value Gwe shoul! choose ‘the best svalgy among a be ‘seteges HHO, 4) for which CH= Gy Atemalvey, # we have a dass of sbategis whic ae equally ecient, we should choose that Sategy for which he expocted Coe is tho lost EXERCISES 14 Consider a sampling dosign BSP), where As) = 1/60 for ‘very Be 8, to craw by simple random sampling wiht ‘replacement « sample of see 3 kom 8 pop of size 5 (2 Gataulate, the probably of including specie’ unit U,in the sample for i= 1.2,....8 and very that Ease ) Calodate 5, the probabaty of incuing a pai of writs U, and U, the sample for | ans vey tat fr ovey ‘and ¥, 8 respctvely, Consier al posse saios of size 2 drawn trom this popuiaton employing a design 148, nore.) i specie a8 (1). AB ah estar ofthe populist tolat SY, consider the estimator bezt (ah Verity that 8 is an unbiased estimator of the popuation ‘ota and obtains variance. (8 Wit cots 2 dota to observe a unt i ¥, > 5 and 4 olar oherwso, what is the expected cost of ebeerving 1 sample of size two. 1.8 A population has 4 fams whose ss ave in the rato of 4121: 3:4. Conedorsolocing a campo of te foams uh prooey roportonal wo size and wih placement using he Sampling mechanism defined by (1). Very ha this exuts in he samping design ASP) wner $s gion by (7) and the probable pts) are A) = 01, 8) = 4, pg) = 08, = 10, HE) = DI6) = 02, pe) = pe = 09, Si) ~ Psy) = 04, 8h) = pea)» 00 BS) = BS) 06, si = i) = 12 1.4 m dopondet samples, each of ste crawn trom Population ef 20 N sing a design (XS, P) Based on the 1 sample i= 1.2,...,m let (9) donate an anbiared lextimator of. LetBAS i Stow tat 0 eae (© E1804 mia) is an vibiane estnar wb (i) For any 20, PYBe)~e|>€}) = VIB aN se? 1. Goomies, VP, (1985) “A uned thay of sanping tom tite 2 Hut, TN (1062) eorc fete in sampng, Sankty 9 Homi, OG, Towson Du 198° genataeton of sampling who! replacement hom aie wives Ae Sat 44 Ko0m, Jc. (96) On the anos of sample fomatin ana fe esi on he cosinor sang Mest 5. Rana Corpeaton (158) mon random dg, The Fre Pras LNG. (1927) Random sampling numer, Ties er Compute, 0, Cambie Unies Pres ablsccconsene® porooer sts Became eee oman aes giceesneecertteitonseene Eeeowense oo gareeeneresens (Hee nownoeneets3 alee ntateeatnenten atte eta aes lemedenonesanaanannaevenad aRReanseane seguasszeg CHAPTER =I SIMPLE RANDOM SAMPLING WITHOUT REPLACEMENT “The simplest method of probability samping & spe random sampiing witout placement Athough tare are omer canwonion. tmatiode of prooeoity samping, desrabio trom to point of vow ff precision, wo shall stat wh simple random samping without ftplacement, sine serves 89/8 good itoducton tothe tary Br cample surveys and i ako hall in the study of oer more compton mato ‘An oasy way of crauing & sample by simple random sampling witout rplacoment so caw wns one by one by aaagring cosa prababityof election to each ofthe avaiable ats a he st and fen eubsoquont draw. Thus, here are W stn and tenable tins Uj, = 1,2,.-+yN, Mo West unit is selected with probabity {IN ho so00nd unk Is selected fom th remaining (W ~ 1) units ‘it prebabiy (03) and 80 on. in gonaal the Un of teh ‘raw wit be solctad om the (Wr 1) une avaiable atthe ih ‘raw wth probaby AW" 1. Ths best scheved wih shal ofa taboo random numbers euch as thoes publeed by ppt toe2n The procedure of selection cf random sample us takes the {orm af a) aeniyng te Au nthe peputason we the unbers to Not, equivalent gropaing tof uns the population in any onder and soil numbing ter; (0) roadeg nutes trom the table of andom numbers by stating at ary arieary pace: and (@ takng foe the sample oa unts. whose serial numbers ‘omeepond to todo aun witoutrepeiton on Me table orancem humbors. The fallow evarpies nil sata the procedure Example 2.1 + Oraw by. simpld random samping without replacement 8 sample of 15 schoo kom at of 838 echoos,Using te tee-tgue numbers given in columns 1 16 9, 410 6, tof to table gv in tha Append of Chap | and ees ‘numbers grostrthan S88 fac ago tho number 000) end repens wo nave forthe sample tho schools with sea nunburs gen bel 125, 926, 12, 287, 85, 251, 165, 191, 198, 3, 161, 208, 51, 52, 399 The procedure rejocs lage number of random numbers. A eves commonly employed to avo the reecton of auth lage ‘umbers i to cvide a random number by 330 and to chooee the Sofa punber fem | tweugh 397 coresponcng tthe remainder hen iti 9626, ard the tral umber $98 when the emainder is zara. However, is nocossary to feet random numbers 7? 10 800 (besides 00) in along Ps prooodure as othorise schools ‘ih sail numbers to 525 vl gota larg ebance of secon, ‘eval fo 8999, while Bose wit seri numbers 324 to 298 wl got 4 chance equal to 2839. wo use tis pracetire and alea the sama tree-gure rancor number as given in columns to 3.4 {0 6, et, we obi te sample of schools wi Serial numbers gen bow 125, 206, 826, 183, 12, 287, 95, 251 325, 990, 114, 231, 78, 112, 126 Example 22 : Nine docks in a conan atinisratve zone comain 753, 170, 970, 087, 1721, 1608. B04" 360 ard 826 househoisrespectvly. Oraw'a sample of € housshots. using tho meted of sinple random sanping withod repacoment ‘The total numberof households in all the 9 blocks 16 7967 ‘he fist stop in the selocon of rendom sample of households 'e to assume tso sera numbered frm 1 T987, by taking Suozesaivo cumulative fa 709, 960, 1883, 500, 4311, 6914, 6779, 7161, 7067 the 799 housenots in block 1 boing assumed fo Nave the seri numbers trough 793, the 170 Roueshotse i block 2 boing ‘assumed to have the serial pumbors 74 vou 69, and so on A reference to the tourdigt random numbers ebtained by ‘reading together clurns 8 vwoxgh 2. te Appenchs 1 of Chapt 1 wal then lve the folowing sample of heusaholds wit al fumers 7388, 822, 4112, 3506, 833 ard 3000. The conesponding households wit bo No. 197 form black 8, No. 129 fom block 2 No. 1822 and No. 1008 trom block 8, No. 699 Kom Dock 1s and another No, 1400 fom block 5s sbviousy not necessary to have allthe 7887 househots in the 9 blacks numboored sarily. Ory the howsehous in bocks 1, 2,5 and 9 wheh happen to coat the random sample of 6 households have to be serallyrumbered in this exampie. However, sara numbering of the hovtehoise ehodd be cartuy dono co a9 noo be ifuenead by the knowdgo of ‘he parol random numbers selctd fer these locks Ie wil be noted tia hi seloctn amour to a selon of the samplo in two stages, selocing a block in tho Wet stage wh probably proporinal tothe numtor of households in the Dok ttt rplacomont ad in the second etago choosing household by simple ancom sanaeg waht replacement In eat selec block on tho basis othe random numbers already selected. In ths trample, the Boks sotected aro 9, 2, 5, 1 and & in tot or ‘Net tai because of relacament, ack 5 has boon selected tee tines! tis obsious thet i wo wore to solct o samp in to stages, Selecting @ numberof fetstage Unis in prebabity proportional to the numberof scond-stage units fin aach test tage ont) While Samping with replacement, ac thon selecing ono secon tage unt ‘fom cach of tho slociod stage uns eee hina sec, we could. make the solocton in one stop by faking auccessNo ‘uml cals ofthe second tage usar sloting the required ‘nunber of socond-stage unis by emple random earpling. The vo procedures are equal. but must be emphasized tat ths ‘valence between the one and two stage sampling ole good fon whan the nuner of second-stage uns tobe select fom each fsttage unt of sampling Isto! tothe number of nes {he ststage unt is stectod camping wih veptacomont als is selection more than ence) 2.2 Some PROPERTIES OF SPLE Ravoon SauPUNG WinkouT Rentacouent Using the procedure of simple random sampling witout ‘replacement as desorbed in Secton 2, clos thal he probably ‘tang an ocd sage 9"= (Uy, ) mh Or Soot wh U, am a he team Ua sr doe 21a 20 on on by Pat) = PUY) PULA Ly UyPUI Uy Upon U2 = Paving U atthe eth draw, ven tat Upson U,, Ravebsen drawn in tho tr 1) rows) werd a Ie folows at onco tat as) = gata « MG ® Maran m Since a sample can ba crawn lal ways andthe probity of drawing a sample ina spootie orca is (W~0)/M, flws that ie prebabsty of obtaing an unordered sami of unt, PU, U, U,) where te ode In whch he uns ar caw is inaloar is en by Nh , ro = H=t / ro) We may, theater, cafe single random samping_ without replacameént a foo: ‘atnton 2: Simp anon sanglng wiht epacamen mate of sarping such tat every ene of te (paste sampios of size 1 from N has the same probably, namely RS 1A) of being select. fe shal cow prove theorem conoering two important propa of single random sampling witout replacement. Theorem 21: n simple random samgling without roplacoment (at The prebabity of drawing a specie unt at tha et ae te equal tothe probaly of drawing Hat te st caw (The probatity of incuding 8 speciion unit in @ sample olson AN rool: To prove (a), we noe thatthe probabity of drawing 1 spected und U, at Be rin aw Is the product ofthe probity that U, isnot drawn nthe frst (r= 1) das andthe probabity that Lyjscronn tine rn kaw. Now ne pobay hat ot slat fie tt craw f (9 probity at ts nt sect St he goood ra hon ft exacted at 0 Wat ans Bethy and coon Moles fatto probebiy Wal U fos selected ite fe = daw NY No? N-8_ Noftt WNST Ne Norse ‘io, prcablty of eran U, atthe Fah draw whan has not teen eran tho fst (7-1) cas i AN re 1. Mute te tes" pbabion wo tad sate prabaty of drawing Y, a he fh eran IN which alo ho prcbabity of drawing U, at 2 st daw ‘To prove (0, we ate tha hace a spostied nit may be inclosed in ne sempleat any ofthe draws, tho roquited proba the umf he probabltes ofr muta excusive evens, namo, the titi ivasced in he sample athe est oka, sand drew. tdraw. Since each of hove events hae probabiy 1M the ecu ‘ollows GeD. 28. Woramiow Let Ye the characterise understudy and donate by ¥, the value of tho oharanortie —sscociatod wih un Dede deco Furr, tet Sj the mean per ut of th popustion he the mean square forthe population n, RM = Yo, thepopuation iat 1 yyw? - Net # AS 0-W Me population¥-E3.y, the samplemean Le the mean ofthe "valves ofhe nuns in he samo = lay 1- FF. tho samp moan equare whore the summations taken over the 2 nits in he sample s ¥ From the definition of coniteney ven in Chapt 1 ts eae ‘hat the sample mean and the sample moan square s* are consstotestmatos of ¥ and $ respecivty. To see whether # Se unoasad we need w Sn her expected vee whic done ov = ‘the cooicient of valaton of ¥ sadn hE where yy denotes the ¥ value ofthe unit kaw atthe Fh daw Using tho tact that te expected value ofthe gum = be sum of ‘he expected valuos, we bial FOE Eb) Since any one of ho WN unts could be ekawn at Ad draw wit Probably 47M, dows hat 33 y- Hence, we concide from Ev that ay-¥ @ showing thatthe somrle mean is an unbiased population mean ¥ simator ofthe ‘an atrnatve approach Is 10 write in to form y- 43% tore 1.4 us icide into sample = {olanenee 10 9 But E(a) Is the probabity of including U, in the sample which we hye seen fen foows trom (11) tat, once again sample mean im an unbiaged estimate ot. To prove unbiasednass of =, we have eo [S ay 0 I fotows by analogy wits me proot of (8), i 0) Fuhor £| aah nf = El aap sake = wT E, Wated 5)where yy and yy dato te ¥ values of tho nits dawn ath and eth dave respactuly. Since yj, can assume any one af he values wth probably 1 and yy ean assume any ona of the remaining (N-1) values except tho one assumed by yy wi probably IAN"), 8 fotos that EV Kl = aE, Y% (9) arcane ae, aH EY) NDE, oy -Mtemaiely. the rest (17) may als be proved inthe following svannor. We have MDE wT EM 1 whore «, ad 90 dein as (10) Flan dnl warn Ee 0 = (2) a a (On substuting from (20) n (19), ws obtain once again the rest an Using (17), wa have from (14) a= Hy Svan way (5a) en wien expressed in toe of 7 ane ®t ean now be soon tom (2) and'9) bat aay 2) howng thatthe sample mean square # Is an unbiased estimator ofthe ponulaton mean square 25 Vosuwce oF Te SauPLe Mes 7 We shat prove the fotowingresu ‘Theorem 2.2 In simple random sampling without repleoement the sample mean yf» an unbasod estiator of Y and is varance e9) where f= wie the sampling tacton. Futhermore, an wnbased do Mat fan € eo rot The test part fotowe kom (8. To prov tho Secund pat we nave bj dofiiton v= a)-¥ ‘Subatutng trom 21), wo aban th rut 25). The lst pat follows from the ft that the sample mean square = is an unbiased esinatar of 5 GED. Corl 2.1 : The standard enor ot is V2 Ss eo OV Te OF Corollary 2.2: FY) = NY is an unbiased estimator of the perl iin viden= nan £.80-n£ enCortary 23 In simple rardom sampling wah replacement the sampie mean 7is an unbiasod estimator of ¥ and He variance i w Fort an bleed ena cna earn to-£ The tactor “5? in 25) oF (26) is a corecton forthe tte W ‘ze ofthe popuiaton and is caled te fie population comscton| ‘actor, Wien is small ab compszed to N, te fe populabon| ‘arecon factor wil not ar much kom uney and the varance the sample mean wil approxnatoto tha fr Yu aol a sample ‘tawn rom an inte population. 2.0. Estmanon oF THE StatoHnO EaROH OF Sinoo isan wnbiasud estimator of $8 natura estate the standatd error of by exisan VG" fare en Wo shai now show that this I a Biased eebmator of SEU, the ‘bias being shy negatve. For Bis pups, is eneugh to considor 36 an estinato of &. We have e= sa vac eo) se E99 ee (0 1 auficionby lrg, Sem aiatte values ely to be tess than 1. However, ever ths assumption does not hold for al Possible values of, me may sil expand s in powers of Ss and ko expacaton tor by tm 10 am a use approximation 10 the expected value of s. Honco, expanding ¢ in powers of 8, oglecing powers of Se hgher man the second and taking ‘expectation, we obtain, is) wasp 2] ow ‘intresting justteaton of tis rut has boan given by Hansen, Huw and Madow (1983). Soo exorsae 28 To attan the varance of #, we have we) = E427) -s* = oh aly ar - wip AE WP ot 2) Espancing and thing expecta tm by em as in Seton 2.4 an expression for VG) can be dave. The dean, however, imvowes much Reader agera tian ihe case oH Ths Caleaton of he vane of igh oer mans even mars ‘erous, The drvaton grat incase ty the us of moat Symmovi tens (Skdma P1099 ore ie of marae Symmotie moans developed by Robten (1057). The Gziesion ot ‘ose methods shower bey he seape his ook We ney te tat the varance of # ean bo dvd by ung eo of tes ‘sod and hat of exresson fre varanca ibe lasing cs ‘ha tno ny & gen by = BAe gg we) = BS (159. 9) 9 Ba = nyt (4) dy « 28. os fs the factor by which the vaiance ot sf inflated due 10 ‘on-normalty. Expected vaties of higher ore samp moments and Thoin products eve been worked oi snd tallated for realy Fetornce by Sulhstna, PV. 1948, Using (9), we havo fom (31) 1-9) = Bo= 6th | 9hon ats anata though as, SS Saree sede oa vo - 80) lee? a2} en fer 8 oral opin y= a wet wo + ges 69 We see that both Me and Ve) are lalate duo to non normaly tp te tame extent, by tho intalon tector fa l icn doesnot depand on skowness. The vaiance remains fated ‘Menthe samples large tase to lntaon factor almost eepondont of the samp eto. Ths result fof considerable "gnteanco to samplers nthe context of éatemtion of te sample fize using reliable ectinate & of SH tho infin factor is not, faken it account, and the popuion ¥s ot norma, wo may De misiod to botove that is rable a9) 2.7 Connosnce Lars Foe ¥ Consider the state z-—&Y, wag jk (1960) folowing te work of Era and Rn (1950) and aco (1048) has gen Necessary and sulcen condtons unex ‘wich dw dsnbuton of the sample mean tds the normal Uist, I Yolowe that Ho dibuton of he characterise ¥ tthe populabon doesnot afer foo much Kom normaly, ven fo moderate size capes, he detbuon of Z ean be approximated ya noma) dsvtution wih mean Zero and varanoe unity. fang 1H vale ofthe slandaed nomal variate coresponding to the value 1 of to normal proba integral, we can tn oxpect the inequality FP heart F f to hold on an sverago wih probability (1~«), The two Stor side of Y ae ead cation ts and theirs betwenn ‘om the confdnee raat The probably wi whi tho requaiy Hols, iz, C1 & teed as the conhdence covert When ls nat known, wo use Is estimator & Io ths cae, tho datiouton of the we) canbe approximated by Student's lstbuon with (n~ 1) degrees Gi readom (a1), provided not Yoo smal and the cstbuton of fhe characte ¥ ofthe population des nek depart oo much om raat te, tho vauo o Students fth rt) 50 Bate ee on = hn) = At we 6a Opel vecaty WETS SYST 2) to hold on an average wih probably (3a) The two Hints on fir side of are tho goresponding contcence tints wih tonionce coaiont (to “The question oatraly aos that the dstbuon of the characterise Vf he population deviates Hom he nor, now lae Shall be tho sire of the sample 80 ‘hat the tts given by insqualiis 40) and (42) wat prove canidnco intavals for with fonkdonce cooiclentapproximato¥ equal to (1). is iat fo answer the question ence kt dopends ypon the rata wth which tho Gatrbon of the sttete Zor t corworgos 10 the nodal fistibuson of he Stidente t dstrbulon as te caBe may De. [Eranigle 2. = Its requted fo esbinate tho total numberof Coweng wets ccnp by oumere We ety of Ames, ova, spread ror 541 Blosse. A sample of 80 backs was selected by simple