dy (in: Use the posterior distin, g 5 He < bp int Use the posterior disbuon, foind in Exec 794 (p38, ly). found in Exercise 795 (9 225) Test sttisic 937 Theo-tiled statistical tet s Type Lemur 338 “Type I eror Uppersiled est 346 ‘deena None 348 ‘Normal population 350 ‘Normal popelaion 376Tm EES ZZ 390 Chapter 8 Tess of typotneses | ‘Summary of Hypothesis Tests: Two-Sample Case Parameter oy Hypothesis Test Statik (md (am) =) =H Independent dt we saat between and then Dy = 09, whore Tis bane on P= +m— 2aand (w= DAH Doe d= Bydin p= aD det sane where “ a then =e Fords = 0 Awe wae nee seu diterence etree py amt Psthen Dy = 0) - 4 Fortzet > ok Fe Focitg "ona Samiesio _Assumotons 23m 230 Nowe " Fitern, < 300¢ Bath pops 399 ‘mvs 30 orth normal vith eq which # 02.88 the moicatons sen the bon op. 36) Allng(ltmy > 30, Population of 364 ttenthestandid erences, myaninglage Independent tough so tbat samples mina 4 Ally adng Independent ocmalpoulsionsLANGUAGE LAB. % z aa A eta 4 theta mht a naught a Siem-squed naught Chapter Summary Notes + Blements of a test of hypothesis: m "ejection region, p-value, and conclusion * Twotypes of errors in a hypothesis test: ype I error (reject Hy when pis ru), Type H error (a fae, * Pohablies of enors: = P(Type Terror) = fake Supplementary Exercises 394 Descreton Nall types Probability of Type Tere Prob of Type I ene Hyped eof popsltion pm in Hy Hypxesiad ah Population man in Hypoxia vole of popsation diference in Hy Hypodiein vale of population tae in hypothesis, alternative hypothesis, test statistic, significance level (a, pt Hy when Hg Reject Hol Hy tue), B= P(Type M error) = P(Accept lM * The forms ofthe altesativ hypothesis: Howertiled test () upper test (>), two-tailed test () * Observed significance level (p-values he sles alc of tha can be wc cjet heal hypothe * Dechion rue for rejecting My 1)! tats rejection regen of 2) ovale < a * Power ofthe test = 1 — fi = P(Reject Hy |My fe) * Key words for denying asthe parameter of interest: mean, averaxe + Key re rmo means using independent samples * Key wordy ears using matched pairs * Key words for identifying p a the parameter a intrest * Key words/phrases for identifying py fompare ovo proportions using independent saps. 3s for identifying aya the parameter of interest mean or average of ¥ wordsphraes for identifying xy ~ ply the parameter of interest diference between means or averages. com sired differences, compare proportion, percemage, rate ‘Pa the parameter of interest difference berween proportions o percentages * ey words for identiving os the parameter of interest variance, spread, variation * Kes wordvphrases for identifying er/er os the parameter of interes. ‘populations using independent samples Swoplementary Exercises "4 Morgolon dese ams. The Journal of Biogeography {Bs 203 pnd wy of a Mags en a Asi). Botanists pce sce has ative sites in he Dry Steppe region and sx sites in the Gobi Desert and b- feral he uber ofan species atte 1 ec se rence between variances, compare variation These data are sted in the ext table Is thee evidence to cmchie tha diference exists Bawsen the average amber of ant species fou st ses 0 the two Fxions ‘of Mongols? Draw the appropite cochasion using ¥ i 2392 chopter Tests of Hypotheses D comnts Ste ‘Resin Nambr of Spades 1 Dry Seppe a 3 Dry Sterne 2 4 Dry Spe 7 5 DeySteppe * Gobi Des 8 7 Gti Dee 5 ° Goi Dee 4 10 Gobi Desert s u Gobi Desert 4 {8.85 morgotan desert ants conte), Refer to he Jounal of Biogeagraphy (Dee. 2008) sty ‘ol atin Monga (Cental Asi). Eerie $84, ete you compel the ‘mean numberof ana eo dese ites Sine th sample sires were small the vrtanss of the populate {wo tes amit he eau 6 od fer te inference be vali Setup fy and for deermling wheter the vriaces sre the same, Us th din the tn forthe test €. Give the scestion region forthe est if = 05 Fil he apron pale oft ten OBIANTS hl finde ests. © Make the appeopit conchae in he won of the 1 What coon are eure for te tes resus to be vali ‘886 suening majors in colege. Whos female undergratntes sich from cence, mathematics and eninerng (SME) "aor dicpins hat ae at ska, a te ‘eason ileea fom the oftheir male soups? This questo was investigated in Scene Edeanen ely 1995). sample of 335 junusenir undergrads 172 females and 163 inlet tw lane esearch iver ties were Menifed as “vwitches” that they eta Asean SME nj fora non-SME mje Each etdene ised one oe more its that combed the suicing 2 OF he 172 females inthe sample. 74 ised ak oan of ierest ia SME (e, “tumed off by sins) a a major factor, compare to 72 ofthe 168 miles Com dct a et (ata = 10) to determine whether he po Ppton of Female witches who ive “ack of cee ‘in SMI" as a major reson for switching differ fom the cotespning proportion of mules Thiny-tee of the 172 females inthe sample ins "at tey were ascouagd or lo confience Bec of ow ges in SME dung their carly yeas com pare 6 44 oF 16 makes. Cons 804 snide inal forthe diference between the propor o feral and mal sicher who fst onde low gas in SME, Inert the esl '257 comparing word processors Suppose you want odes mine whether usr of dita proessrs have pefomnet Detveen word process AaB if ases hoe no ce frie of be wo word pressor ie fie to ‘ysems ae ec then the probably phat 2 wet ‘ef sytem Ai p = 3. Let Fe the number of {sample of 10 who preter sytem Ay and suppose yo ant ots fy 9 = Sapo Hp 28. One pie tet procedure tore fit = Vor Y= 8 2. Finda forthe. 1 Find itp =. Whaisthe power fhe te? Ral itp ~ 8. What isthe power ofthe tent? 8.8 Mensng paper tress The quay con depat set fa puper company mess the gts ames te of reflectance) of fins paper oma pero tas hough the dy. Two intents hata anal ‘meat paper specimens ae mjc! oro, at 0 be asd so that te mean readings far cnt oper specimen are the same for both insrments, Sa [oy ar concerned al th precision eh tn Stumennares tht nse 2 is ess pecs a Inu To check thie five mesure sine paper sample are ae om ho ramet The la are shown in the bl the at roi uit {sides to ndcte ht insane presi a instrament 1? Test ning = 08, D swcurwess insur) rstamert2 » % 20 a0 8 2 0 28 889 Schodutns te dabiny tests. The texting deparme of 4 te ad aber company schedules ick ad pose rs or dab es Carey, ties ae bse doe weekly Mleilegroesors (machines thi Ct hance cite cor pasenger tie) wing the srs !mocesing time (SPT) approach Under SPT the ie i the stot proesing fief chal ft. Com Po ‘searcher bave developed « new scheduling rales! they believe will ede the average ow tine (se, serge compleion time af ts and led 104 ed inthe average urtines of sched tes. To comps ‘he wo scheduling res ties were andoly sce) __ es scheduled using SP, fie provided a theta sul size Oo softs = arenes ow wr 1s28 8SI280 S26 48209 Is torsuliciem cvdece at = 1S to conchae that the average fo ime sks unde he proposed sched ling ul than der the SPT epncach™ [there stint videnee at = 05 wo omclde at the proposed scheduling rae wil lea oa redo he wera 190 tos capacity of Foams. Refer to the einfred cone Team cracking experiment describe in Exerie 873 (p82) The expednoral rcs were Compare he Metical resals obsned wing he fale sare red of predicting ultimate lad capaci. The atl ‘1 cortical lat tron moment for sx Thsuns thse sab with te given inthe table, Conduct sto determine whether the experi ean wkimite ‘erin moment der rm he theoretical es wate 9 mone — ee a a se se BowinenalRewh 470 520 S40 $40 430 80 Mevetsal Result 463 465 S440 S40 S62 363 See ie ere ec 19% atone water problem that cut with cei pet mining sere y-pednd oe m ‘Tadioactive and these pr ioe metimes get into our fesh wate spy Te EPA his se eli or Cen ini ne sna chatty ps ting water Parca the ms vel oa way cca tation 3 ios per eo wae, Aandi of 24 war sce fom 4's {es py profuced he mpl sani j= 48 pe Seas per cs 87 peur ie 4D these data provide scent eidene tn indlcate thatthe moan evel of alton sf (blow the ma um eve ey tho EPAY? Test wing = 01 Why stould you want wo ne sal eof forthe testi prt a? Calculate the value of Yor te test if ay = 45 pi ‘ois pr her of wate. o£ {ate stp he pe fr te 12 Overeating ina spa. Saying 00 log in a $s pool can esl in overheating, which in the case ofa preset Supplementary Exercsos | 393 woman, may cause fetal malformation But how Jong is too long? Based on thie werk in ths area, severl searcher hype that 75% of women, mes in spa with water tompeatre of 40°C, wil become unc Feetabiy ot when heir ear cna (cor) emer reaces PC. Arrest aubjective dicot is suggested a 2 posible seguir a Apparently contacted by an Austulian sly of 24 (46%) were uncomfortably hot when their core tempera ture eae 40°C. Test the Bipot ht the te po enage of hesthy,neapregniat women who become uncmnfertably fot when their ore trperatre teaches a0°C iow than 75%, Var = 4 vor 898 Contamination of sh. Refer Wo the US. Amy Comps of Engineer: ay of corterid! fishin the Tenner 2 Use arc number table abe 1 generne a random sample of n= DDT concenttom in fh fo te BD 1 The Road ad Drog Administration (FDA) sets the Appendix Bye lit fr DDT coment im indivi ih at Spar ot len pm). Dos te sume of par a provi sa ‘et evidence o conc tht the sverige DD con Ten of individu! fish inhabiting the Tennessee River and is cock bates exces 5 pp Sigifane level of a ~ 01 Sipps he fst of hypoth, ptt mas bse on & ramlone sample of only n = 8 inh, Wha ae te dia anges ending this sls est 1. Reed per using only the stormation on the DDT seis of sample of fish andy selected from the 40 observations of tw, Compre he resis of 1298 Bal bearng specieauons I the mamuacta of machin eit esetal tole path conform pec tins nthe pst, haters ofthe hall easing pred bya certain mamaactrec ad varanse f 4136. cat esi the manufacturers lesseapeive poae thon methid. Te yatnce of the meters of 10 a domly simpled bearings produced bythe ne press as 2 Doe ape ic ne ne ‘fe more variable thus tone produced ty the old proces? Teste = 05 1195 Atmasphericln bance A high conceattin of positive jm in a oom can induce fig, sem and respiratory lems inthe roms osapants, However sec as ‘heroom's amosphere (ough a negative ion genera, tn combination wit constant vention restores the a tal balance of fons th is conducive to human beak (One experimen! war cdc as follows, One andre394 chapters Tests of hypotheses employees of «lage Factory were radon elected and divided imo two proupeo 3 each. Both paps were tld ‘hat they woul be working ian espe wth a on ‘ulancs contol thrgh negative on Eentar: How ver, anknown to th nly in th experinenal grays woe aes Aithe en ofthe das the numberof pox epg migraine, nausea fatigues, o some ed ys sal discomfort was corded foreach group. The Reis are summed inthe ube ontroiGreup fon gens Experimental Group on genes ot ‘Namibern Sample my = 50 i Patel Discomort 3 a 4. Perform atest oF hype wo detenine wheter he Proportion of empioyect in he expeimenal gry wvho experience some type of physical contort the en ofthe dey i szniicim es than the sone sponding proportion a te ent gop. Use 1. Compt the phe fo this test '376 mores of WaSDs. Refer 1 te Scene (No 1988) ‘uy of inbreeding in tpi swarm founding waa, Exercise 798 p.329) Asample of 197 waapr opera, frozen, and subjected series gente eyed Sadar deviation of «= 84, Recall hati the way bay to tendency to ined, the tre mean itveeing cect ‘Teste hypxhess ht the te mean indice deny ori specs of wap ences 0 Use = 08, Bvaron 2. Compare he infeens, pat 8, 10 the inne ob une in Esersse 798 sing confidence intra Db thelnfrences age’ Exp 897 Strnats of uty poses The meus snd sandal de fens shown in the table summarize Infrmaton on he sents modules frp at rund in pup ‘qa inc) far 80 types of wooden pcs wed ye lity indy Do the data provide suicem evden indie a diference in the variance of the sot ‘ween ols made fron coil Cols ian cet Poe! Test wing = 02, = = sage Sample sample Sandee Soeces Sze Mean Devatin ComlDougcre 118 ano ket Souter Pine a7 got 898 Coa tar apo pressure. The le below proves isn ‘he theoreti (cakulaa) an expesimentlyales of 8 ‘gor pressts for dbenthiphene hee "atc Compound sins to hoe found in cal a He tert made fo vapor pres sa good model of aly, the te mean deren between the expe sn called valtes of Yapor pres fr BER Et ee Wl ual. 4 Do the dst prove sutcent evidence wo into ‘ht the mean difleence dies frm 0? Test ie a= 0s 1, Caleta interpret the pale forthe est wore | Temperatue(°C) __exprtmentai—_Calcuated 10060 an 21% 10136 34 sr 1040 as 330 1044 08 90 10870 4 a 0, sa. ses nae ss ss ust 2 0 rn Dts N36, 1S, 21-128 Crp 5 Aso Sine: Eda DR an Pras 1M." psa fa pa anor Presse Temperature rc) Experimental calcul 669 69 ws 93 a3 ss io 350 2 2361 tor 101 12490 107 108 ian 126 s 13024 ta 8 Bias ss Ls Seated panda of Cael Cenc Soe Repent wit eran or iStatistics in aevion | 395 yt Bates pans icc i i aac wc or tae igh ofa Bae ee os pel ocims Tm ome sees tee erie ee menos enti noee Seeccnatmiecmnl mer eainace be ole To mci probing to ey es ‘evel Atma edo oad mers ce pend yt unalone: ed oes pagua woh, rv niet ones (ec = ie el estado chek to decent a Sas te be cram maa of tg goss (Rane epenee prety Le proche Le ok tng ed acer eee ess So sci egh te Sippel ane oa acy ttt thn te concpning sear et fit 100 iter d's = Otc Does Sein ‘plein nde ta tt mace pecs fase ba ih ama np ot eo nsf ie tacos Cone ess prance ono oneal Tes = ae nee t000 Hess nace, Rescue iat FAS Day : = Recah Unit and he Depart of Area ee Engen te Uninesyo Plane Seve nats ae3 desien ere forthe comsraction of shade structs Sear: Relea 0.1 “Boe aio ‘hat they believe will help alleviate eat ste for dairy foraging ind its impact on seed deities lrmeserof pregnancy were divided ino 80 groupe AS ep. R092 Copp the remaining 15 cows wee deed cored the 10-day ilk il (in pounds of de. Researchers 8.102 Sola traction stay Rese a i compan le eeu eee Ee eat erat lo lee to indica a itference between the mean milk Tho data, repraduced inthe table, were wed to Compare ells of cows sven acces jo shade and cows de the mean, day-long, leash solr rion levels (in sae? User = 10 (assume the andar deviations of Ucar = 10 : BTUng fst two midwest sites shade No state eee Simple sce 16 is 2 es ‘Sted chnsty strewn stay. A field expesimeo was ‘adit 1 ascertain the impact of dose runes eee ce - (sete) onthe dest amd dition of 80 in a cc the sol (oles, Des. 197), Sine ome desert mes ay 2 in6 'e mown ou sees in sre caches, the tly March? 33 ae spect designed to Jetennine wher thse op igs - ches eventually palace mote sei 1698 #0. tha an ajacet contol ares. Foy small aes cael by mens were laced and covered ith Dt capes to prevent mde rom esi he caches. STATISTICS IN ACTION Comparing Methods for Dissolving Drug Tablets—issolution Method Equivalence Testing Inthe pharmaceutical industry, quality engineers are responsibe for maintaining the quality of drug prod cts produced inthe manufacturing process. The key to qual isan assessment of product characteristics ‘trough repeated measurements oft varable of interest When the vriadle iste concentration ofa par ticular constituent in a miture, the process Is called an assay For this Statistics in Acton, we focus on a ‘hemical assay to determine how fast @ Sold-dosage pharmaceutical product (eg, an aspirin tablet or| 396 Chapter # Tess of ypotheses tient, quality inspectors requirea test that accurately measures visoluton, 'n "Dissolution Method Equivalence” (Chapter 4, Staistcal Case Studies: A Collaboration betwom ‘Academe and Industry, ASA-SIAM Seres on Statistics and Apoled Probabilty, 1998), statisticians Russet | Reeve and Francis Giesbrecht explored the dissolution characteristics of a new immediate release Ong | Product manufactured by a wel-know pharmaceutical company. An immediate-elease products dexgres \ \} to dissolve and enter the bloodstream as fast as possibie To test for dsolution ofthe sold-dosage ug the Company uses an apparatus with si vessels or tubes, each tube containing a dssoWing solution, Out {ables or capsules re cropped inthe tubes. Then, at predetermined times, a small amount ofthe solos 's witharavm fom each tube and analyzed using high-gerformance liquid chromatography (HPLC) Te HPLC device quantifies How much ofthe drug sin the solution; this value is expressed as percent of abel strength (5) i i] | Cepsule) ssohes. ce varaton nthe soliton ofthe dg can have harm ie eects one ‘duce results equivalent to those verfles that any two sites using the dissolution test produce equivalent assay resets ‘Dissolution test data for an analgesic in Tabet form conducted at two manufacturing sites (New lesey and Puerto Rico) are listed in Table SIAB.1. (These data are saved in te DISSOLVE fle) Note that MLS les were obtained at fur different pins in teeter 20 minutos, after 4D minute sites eo i ‘ter 120 minutesfor each of the Sx vessels Based on the sample data, do the to sites prod, | lent assay results? | An intially appealing approach to answer ng tis question sto conduct atest of hypothesis onthe ot ference between the mean SL measurements tthe two ses Let ay represent the ppuion ean AS tor tests conducted at the New Jersey site and let represent the population mean see for tects cot, | srerag atte Puerto ico ste. the testresuts ath tw sits ar equivalent then ag The nl alternative nypotheses can be stated: | He (1 ~ 42) = 0 (ie, dssolition equivalence) ) Hi (us ~ m2) #0 (ie, nonequivatence) ] ZoammpHly the analysis the statisticians suggested conducting ths test at each of the four time pets St Bac time point, withthe resits shown in Figure SUB. TrePalues forte two talled test gle onthe prntoun fo 20,4060, and Lateetrires cre {me are 1528, 0395, 3499, and 4956, respectialy. wa select type arteritis heen h ‘oralectha (p-vakve > 05) fr tree ofthe four tie points: only when css oe tok so roes & ossowve TABLE SIA8.1 Dissolution Test Data (Percent Label Strength) Ste i ee ee a ; ; ° || © Fe " e s - » a © 96 » a 95 9% ” ii) tan ean ce Pade | o 6 “ a re ° 9s *” 98 os ~~ 2 = Ng ole SSR REE hap ns ae AS es pe eae Gan a a iy. 198 (Chipeta iStatistics in Acton | 397 Bonple Statistics Group N Neon Std. Dev. Std. Error Neo Jersey ‘Ss777 Tagen Poort fice & 85 4.3000 tire Null bypothesie: — flean 1 = Mean 2 = 0 Aiternat ive Mean 1 > Mean 2 == 57 Ter a3 Octsar 802 Conf idence Interval for the Difference between Two Means Lover Linit Upper Limit Tike oor) N Neon Std. Dov. Std. Error Neverssy 6 vas a.atsa Taaze Puorte Rico 8 eaiener atte ivase Hypothesis Test Nutt hypothesis: — Mean 1 = Mean 2 = 0 Alternative! Hoan 1 = Han 2 >= 0 It Variances fret statiatic oF Prot Equal ot Equal Par 9.27 oloata 30% Contidence Interval for the Difference between Two Neans Lover Limit Unger Limit tas sa FIGURE SIA8.1a | | or New Jersey 6 96.8999 2.3381 0.9545 Porto Rico 8 Seleeser 5.9440 14088 Wypothesis Test MoI hypothesis: ean 1 = ° Aitarnat ve’ ean Semple Stat istics Group eon Std. Error Nev Jersey 6 aaveseay Porte Rico § 9g S558 Wypothesis Tost Mal hypothesis ttt bop fea 1 - Neen 2 = 0 Mean 1 > Rean 2 == 9 If Variances fre ereerg 4360 FIGURE SIAB.1 (Continued)Statistes n con | 399° TABLE SIAB.2 Determining Dissolution Equivalence ean 4S Ossouon Equivalence Oca mre =18% nd 15% is tere suficient evidence to conclude thatthe mean XLS values forthe two sites die. nother words, ‘one might ressonably conclude from the hypothesis tests thatthe two sites produce equivalent results at Aissoving thmes of 20,60, and 120 minutes, but do not produce equivalent results at a csso\ng time of 40 minutes Tete are several caveats to ths hypotiesis-testing approach, a the statisticians warned in thelr chapter. First the idea of equivalence in te testis established by “accepting Ho” Recal that & measure of Felabilty for the conclusion ‘accept Hs" is 8 = P(Type terror) ~ P( Accept Ha His false), For tis apo, Cation, Bis the probabity of saying wz wen, fact, the means oe Since the sampling dstrbuvon Ot 41 ~ is unknown when the ateative condition, + sz, Wue, the exact value of pf unknown Second, the notion of “practical significance is ignored in the hypothesis test Thats, although the popula tion means may be statistically diferent ato = 05, the true difference may be small and not considered 8 Imeaninglul difference in practice. Finally, the test may have the unforunate effect of penalizing a testing site wth small smaller than average) ELS variance. You can see this by examining te formula forthe test Statistic in Section 8.7. when the diference in sample meansis divided by a small standard error which wil ely occur fone site as a smal variance}, the resulting value wl De large (and likely to be signicant) TD overcome these problems, parmaceutical companies have developed alternative approaches to the equivalence problem, One method, suggested bythe statstcans in their chapter requires that you frst fn @ 90% confidence interval for 1 — uo Ifthe confidence interval forthe diflrence between mean SL ues les within equivalence limits established by the company, then accept the assays ofthe two sites as being equivalent. the compary in this application uses the equivalence limits in Table SIAB.2. Note that the "mits depend on the magnitude ofthe mean SLs. Using the equivalence limits of Table SIAB.2, we wll accept the assays ofthe two sites as being equiv lent ifthe 90% confidence interval for yy ~ jg: (a) les between ~15 and 75 when the mean ®LS Is less ‘han 9, o (0 ies between ~7 and 7 when the mean LS is greater than or equal ¥ 90, Note that this ep roach is equivalent to testing the following hypatheses or those assays with mean < 90%) He (uy ~ ua) < ~180F (us ~ ma) > 18. (Le. nonequivalence) Hai —15 < (uy ~ wa) < 15- (12, dissolution equivalence) For ths reason, this methodology is referred to asthe two one-sided t test (TOST). To apply TOST tothe data of Table SUAB.1, we find the 90% confidence intervals for yy ~ ya. These con- fidence intervals, as weil as the mean SLs values, re also Sneded inthe SAS printout Figure SIAB 1. The Confidence intervals for each of the four time points are all within thelr respective equivalence limits Ge. between — I and 15 for times 20 and 40 minutes, and between ~7 and 7 for tie points 60 and 120 min tes) consequently the data support dissolution assay equivalence between the two sites for all four dis. soliton times. TOST is now considered the standard method for bioequivalence testing of pharmaceutical products {and is becoming widely accepted in process engineering. chemistry, and environmental science. An excel lent tutorial on TOST is given in “Beyond the Test: Statistical Equivalence Testing.” Analytical Chemisty Lune 1, 2005). There, the authors provide insight into TOST sample size determination and on how to hocse the allimportant equivalence limits. a fern Se AM maa TeCHAPTER Categorical Data Analysis OBJECTIVE To show how to analyze count data obtained by the classification of experimental observations from | a multinomial experiment CONTENTS | 9.1. Categorical Data and Multinomial Probabilities 9.2 _ Estimating Category Probabilities in a One-Way Table 9.3 Testing Category Probabilities in a One-Way Table 9.4 inferences About Category Probabilities in a Two-Way (Contingency) Table | %5 Contingency Tabies with Fixed Marginal Totals 9.6 Exact Tests for Independence in a Contingency Table Analysis (Optional) STATISTICS IN ACTION ‘The Public's Perception of Engineers and Engineeringble Nay Ne 92 Estmatng Category Probabities in 8 OnesWay Table | 404 9.1 Categorical Data and Multinomial Probabilities In Chapters 7 and 8, we discus nale population. Recall tha how ro make inferences about a propetion from a Population propetion pi the probability of "ste «e35"in binomial experient—an experiment that resis n one of two possible ot comes on any one trial In this chapter we ae interested in making inferences about the unknown probabilities (or proportions) fom a multinomial experiment with k possible outcomes. Tha is, we want to make inferences about pyspa-- py, where p 1s the probability ofthe th outcome and py + ps + (See Section 47 for a detailed disession of multinomial experiments.) Toillsrate, conser motor fan blade company tha manufactures impellers on ‘one of ive praduction lines, A, B,C, D, or E. Assume that the lines proce nnpellons ue rate an volume. Ina sample of » = 103 impellers found to be detective manufactured online A, 27 on line B, 31 on line C.19.0n lite D, and U1 on line (see Table 9.1) For this mulinomial experimen, there ae five outcomes, or ach defective can be classified, one corresponding to cach of TABLE 9.1 Classification of = 103 Defective impellers According to Production Line reuse ractcal question to be answered inthe sti is whether the proportions of defective impellers der among the five production lines, Do the data provide ev dence 1 contradict the null hypothesis Hp = = ps, whore pis the pro portion of defectives manufactured on the ah production line? If the data in Table 91 ‘ontradict this hypothesis, the manufacturer would want to Know why the at of pro Auction of defectives is greater on some production lines than others and would ake ‘ounermeasune to redace the production of defectives, This chapter s concemed withthe analysis of eat that represent the counts for each category of multinomial experiment. In Sections 9.2 and 9.3, we will arm how to make inferences about the category probabilities For data classified according 10 a single qualitative (or categorical) variable. In Sections 8.4 and 9.5, we consider inferences about the category probabilities For data clawed according to two qualitative variables. The sltistiewsed for most ofthese inferences is one that posesss, approximately, the familiar eh-square distribution. although the proof of he sdaquacy ofthis approximation is beyond the scope ofthis tex some specs ofthe theory can be deduced from what we have learned in earlier chapters, orca data specifically, data °2 Estimating category Probabilities in a One-Way Table ‘Consider a multinomial experiment with outcomes that correspond 10 cateyories of a Single qualitative variable, The data (ie, category coun) for such an experiment ‘would appeae similar to that of Table 9.2 where nis... «mu represent the category TSLE9.2 One-way Table wns an m = my m + * + me. Such a abe is often called 2 ome-way table fategory Counts sine only one qalave variable swe form the categories, or outcomes eters Toestinte category robbs na one-way table, conse hat ¢ maim ——_ Seam experiment can always be reduced to a binomial experiment by isolating oe cat al thers. Since we know that in a bina ns epory St, category i and then combining experiment with number of successes, ¥.) = ¥/n isa good estimator ofthe binomial parameter pit follows that aa Sen gone ofthe probly soci wih carn ina mune exper iment. solv th wl poses the same roe as ean, ea whens ee, lb apex oma ctr (by he conta tit thee wh Fi)=(iy lee ne | aa | vo ‘Consequcrly, a lrge-sample confidence interval fr , may be constructed ss shova | in the box | A Large-Sample (1 ~ a) 100% Confidence Interval fr pin a One-Way Table Table $ of Appendix ae We edimate the difference between a pir of category probabilities, 7 = stories and ( # ), using (iy ~ f,). This linear funtion of and fj wl Be approximately normaly distributed with BA pie and PB) = Vi) + 1) ~ 2CovtG.5) Since the covariange of two category counts, say and my (i +), given by Counc) = ~ may (the proofs eta an exercise atthe end ofthis section, it flows thatthe covariance Derwoen the conesponding eximatrs, and ) (n,m; | Cov (i d)) = AUG, ~ ple - A { 1 1 2 Corte m) = (apy) ‘Therefore | Vai ~ B) = WA) + VG) ~ 200A PAL=p) Rm) any and lrg sample (1 ~ 100% confidence interval for () — the bax is as indicated A Large-Sample (1 — «)100% Confidence interval for (p, — p) ina One-Way | eases . [oi =F) * Bap + 2A w aa OE RT vi‘9.2 Estimating Category robabiities in @ One-Way Table 403 CSET wre nna 54 ts rel icon 9 cc nes incon be tra ocucton A Noe et Psat prenTton cimpelae pode roduc fea mat ae eleva, athe operon a dace Mle ey ue pretkcon ins Seluion | From Table 81,we haven, = 1S anda = 103, Thera, 295% confidence interval forp ia 5 -m 1s fs wher fy = = 1S as = a6 + 199, [CONES eS ey | 0°46 4 068. Tere. ournterval estimate for pi fom.078 to 214 ‘That i, we are 95% confide tht the true proportion ofall defective impellers that are produced on ine A falls between (78 and. 214, CET | Pio srr 91 ts 2 cdc inal ~,t oen bencen Proportions eet mses arouse to craducton ines ae esoecnry Soiton | From Tale 9.1, we have m= 27 an 7/103 = 262, Then a 95% confidence intra for (pp) _ +R A) i (i -b) dey oe (CraaKasay = (a = (146 209) + 196) “HODES * =.116 4 121 ‘Therefore, our interval estimate of (7 ~ pa). he difference inthe proportions of the defective impli atibutble to production lines Aan Bis ~ 237 OOF Sloe thisimterval icles 0, theres inset evidence at the 98 confidence vel is conclae that the to proportions df Applied Exercises 28 ew asincuon study ceo on dental pases with emooroandbulr (jam) jit dysficton (IMD) was ‘alia in Genera Denitry an, 200) Aram fame f 60 patens was selected for an experiment Letowit of TMD. Pir to warmest, the pte ied 8 survey on two nonfunctional jw fists (eth rinding) and ec clenching that have en ‘inked wo TMD. OF to 6 patents, 3 ated o brie, [Hadiined oth clenching, 3 adie o both abi, 2416 claimed they bad ether babit. * Desctibe the qualiative varahle of interes in the ‘tay. Give te level (categorie) snc wih the anal AG sri one-way tal for the sample data & Find and inert $3 confide aera forthe propeton of deta patents wo admit wo bth Find and interpret 95% confidence interval fr the Ailes between te tr proprio of dental ps ‘ents who admit to hot habits and the tra prope tion of deta pens who elim they have oer hab 92 Turns a ierseetors. A alc study found that of 972 automobiles entering aus ntseton ding he pete fom at 107 387 tre Te “21 ered gh nd 234 drove sight tough he intersection 3. Comsrct sene-vay tbl forte say Find» 955 confdence inva rth tse proportion ‘of suomotiles tha ve suai ough the merc ‘ion dig this peri © Find a 95% confidence interval fe the difference be- ‘ween the proportions of eons htt leftard turn righ especvely, ding his pes. Inept the interval.404 chapter 9 categorical bata Anais. 99. cADtectnolay Each month Mechanical Eusering wr ey is readers with an cline “Questo af the Month The uy 204 se reported on the resent the ques Yin, "Do you fel you know enough sb te ae on pteraded design (CAD) techacogies wo do Jour jon? ‘The ress 44% anwwered “yen” 1% ante "hs at mot wowed abot,” 38% anvwered "na cr eros me.” and 9% answered“ don eed to kaww CAD inmy job" Assume 1,00 reades respond to the lng se Find and ites 98% confidence interval for the Proportion of ears who fe ty know endogh abt EAD todo hiro 1 ind and interpt 95% confidence itr for he sieence beeen the proper of ads Mh a ho answered no, ad i concerns ne 94 Dial slr ae image process. Dig inal and age processing (DSIP hk wide varity of epic, inching entertainment (eco on demand lansing, sccuitysrveilance, military taret cot, wiles ‘ommuniatios, and inteligent transportation sym CConseucaty, dee iss rapidly prowing need ft eng ‘ers tied in DSI. The International Jeural of ee trical Engincerng Eduction (Ape 31) repre oa ‘vastion ofthe experimen DSIP ondopradate te QVM). sample of 80 student repunded to heat ment “I hebeve at his esac experience i vey vale able t my profess fare” The resis 47 set agreed 3 students were seul, ad 0 students sagen ‘ih he tere, 4. Estimate the popoton of WMU stants wh ace "hate DSIP research experience i alae thot Professional ftir. Use 99% confide ima 1b Entimate the difference betncen the propervons of ‘WMU studems who agree and who ae mal abut the stsemeat Use 3 9% confaene imerea 2 Ponce 9S Chumetersties of ice motwonds. Refer to the Nail Show and Ke Dut Comer (NSIDC) sllston of dats on S01 ee metpons io the Cioadan Arte, Example 21 p16) The tare sanedin he PONDICE file One var ae of ire 0 eovionil engiesrs tying the Imeionds isthe pe ofc observe fa ach po. Real that type ic hase Ft year ees mlgear eo ‘SAS Output for Exercise 9.5 lant ice The SAS summary ile othe ce ype el $ mekponds sori te ton th pope 12, Use a 90% eofidene itera to exinat the propa thon of mein inthe Casadlan Arctic ht ae Use 090% confidence iteral to estimate the die ene etwcen th roportion of rtd nthe Can in Arctic tat ave fst yea ee and the prop ha bave mates ice 94 Onontaton cus experiment. Refer tothe Han Fatt ‘se 198)sndy of eolortrphness os nbd sito: slow, Exiie 7.9 (p. 301). Ninety college dee lining cm the tacks inthe dr mere Genented whet Prsitoned ons ring platform unde lowly tating dk ta bloke thei fel of vision, The subfcs wee shed to sy “sop” when they fl as if they wer he side wo. The potion of the brighines pte the ds ‘in ean o each tue’ body elation was ed record Sujet lei oly thre sk ges pt tems as subjective yer clues: (1) tiger sie (2h darter side up and) trig and are ie ald miter sid ofthe subjecshead. Te fesncy Sms forthe experiment are given in the accompany ble ‘Consus 495 confidence tral forthe diferent ‘wee he proportion of sbjects who sect brs sie 'upand the poerion who select darker ide yp et ‘hes Inert the esa soovewe = isk orentaion or rr se Up_ Darker Ste Up rgnt and Darkside AS ss is 97 Gastoentsrts outros. vatetoene ponbcteil 3 ‘ocneitsoutueak ecu in Corde sa ro Tonestandng her deieoey and malfunction of * aps teatment plat, A sty wa conde © dete ‘ine whether the incense of gastointesia diet ing the epidemic was vested 0 water consuming (American Wier Works Jounal, tn, 196) A telepote sary of houscholdsyieked the accompanying iB ‘io om dy consumption of ounce pace of wate 8 stool of 40 resents who exhited gto ‘ympioms during the epidemic The FREQ Procedure leervPe Frequency Percent First-year Londrast. ye natttyear 2 ales93 Testing Category Probables na One-Way Table | 405 Socm ‘ental Plerids.Arandom sample of 10 agricul nh = ee mes cers was imerviewed and asked which of the strategies Daly Consummation of he or she believes would yield maximum productivity 8-ounce asses of Water A summary of hei reaposs is shown nh ble © 1-2 34 Sermow Tort oy mmoare sitsympuo| 6 Bp om 4 = Fe tee om 2 Frequency Ut so» saan Wer Ws arc NL TB Nn {Be Copan © 1986, American Wats Anociaon @ Finds aleogacers who rsermend sttety C a for the diternce = ween the tm proprions uf agricul engineers es who dink 1-2 glass of : who recommend sess End 8. 8. Use 199% confidence interval to estima the pect ae of gates ho ink 1-2 and glasses of water per dy. °% isting crepend. Because of cmsic mifll paterm Theoretical Exercise {2 low water olding capacities of soils in Fluids, 9.9. Fe the muinonial ebability distribution show tat, Spree imgation is required for pring most Stops. & research eam his develope five aleroatve °3 Testing Category Probabilities in a One-Way Table ‘Suppose we want to test hypothesis abou the category probabilities forthe defective impeller study using the data given in Table 9. Specifically, we might want to teat the (aul) hypothesis that the proportions of defectives atribuable to the five produc tion lines are equal, Lew Hospi = P2 =~" = ps = 2. against the altematve by pothesis that at east woof the probabilities ure unequal. Intuitively, we would shooxe a test statistic hased onthe deviations ofthe observed category counts. ‘ns rom thee expected values, or expected eategory counts E(n) = ni = (103)( Lange deviations between the observed and expected cutegory counts would provide evidence t indicate thatthe hypothesized category probabilities are incorect The statistic used to test hypotheses about the eatery probabilities of a category mokinomial experiment, one hased on the weighted sum of squared deviations be tween observed and expected cell counts, is Substituting apy for Ein) and expanding the numerator, it can be shown (proot ‘omitted that: | | acre 8 \ [oegeuanghae the statistic x will possess (prof omited) approximately a chi-square’ sampling distbution.® The val of 42 wi ib larger thn expected he deviations [n; ~ E(,)) ‘arg lage. Therefore the ejeton region forthe testis a* > 2, where isthe val 2° tha loses anaes ain the uper tai of te chi-sjare distribution (ose Figur 9.0, The numberof degrees of freedom forthe approximating chi-square dstbution always equal less {degre of freedom forever linearly independent rests placed onthe category counts. For example, we always have atleast one Tne tition on the category counts because their um mst egal the sample nee A Test of a Hypothesis About multinomial Probab Hi ies: One Way Tale ates = P20 k= Pa Where Py Pay. Pag Tpmesel HE !hypshesied ales ofthe multinomial robbie Me: Atleast on ofthe mukiomial probable doesnot guts ypbind value where £(m) = pos the expe ed numberof outcomes of type 4 assuming H i+ true, The total sample size isn Rejection resion: x? > 2, where x2 has (k ~ 1) at Assumption: Fo the chisquate approximation oe vali E(,) | Leon: Sire cisare approximation to be S forall very Probables assume specified Vales el in only single near resticton om the clegry county mac mr ao fe 1 = 1. No category probabilities need to be estimated besause this value are opt | Sed in Ho, The test procedure is described inthe preceding box, We will vst is simple application ofthe chi-square test in Exae 94 ‘ease splot expected coun cnbe eth Mo ts jected ‘per Cth 08 iene owner orcasCE | ea: Solon {9.3 Testing Category Probabilities in & One-Way 7 407 he cats poten Table 91. Teste hypothe ta the proprio fl deface im We wantiotest Ho: py = pa = ++ = ps = 2 agunst the altermative hypothesis Atleast to ofthe cotepory probabilities are unequal. We ave uieady cucusted rt 8 (= 1.2...,5) TABLE 9.3 Observed and Expected Category Counts for the Data of Table 9.1 Ome 8 ‘The observed andthe expected category count (in pare Substitting the observed ad expected val for x we oan eves) are shown in Table 9.3. sof the category counts into the formula 206)? 206 = hum (1 = 206)? 206 336 The rejection region for the west is 42> ys. where xs ie hased on k — 1 = 14 degrees of freedom, This value, found in Table § of Appendix B, i ats = 948773, Since the observed valve of 2° excoeds this value. ther is suRerent {evidence (at a= 05) to reject Hy, It appears that atleast one production line we spoasibe fora higher proportion of defective impellers than the ther lines, Wore The test fora one-way table can be conducted with slatstical software, The SPSS | primout ofthe anaysis is shown in Figure 9.2 Sine the p-value, 010 (highlighted) Isless than @ = 0S, we eject Hy) une Dasenean | Bpeceon [Rena x me] 2s] 86 8 Z| (eaten vce 6a fe a1] 208| 104 D a] ms) 46 : u| as] “98 Teal 103 | | Tost statistics c TRE THOTT] Ts a 4 ee 3 0 ces 8) have expected trequencies less than 5, The minimum expected cel frequency is 20.8 FIGURE 9.2 SPSS analy of drain Tale 9.3 ie eoT airr 408 chapter9 categerical Data Analysis Applied Exercises 910 Scaming mtemet messages: Ine: Tchalogy (M18, 1997) reported thers of on Equitas Consumer Privacy Survey in which 328 meme srs nated et level of spreement withthe following stem "The government eed 1 be able to scan Inert mxagee and vser communications to proven fad and ate rime.” The rumber f wet it ach sponse cap summarized se to dteine If the opinion of Cos the et bess yo woul arc ney ae egos pata using U5. © In the content of th robe. wha i Type Lenoe? A Type emer sumptions ust old in ert eur the ality ofthe est prt? B scan “agree ‘agree Oisaaree Okage” Storey somewhat somewhat Soa en) 2 » Ponce P51 Characters of joe moezend. Retr the sy of ke Imeyo inthe Canadian Art, Enecne 9p.) The SAS summary il forthe ce Ope the Sn ‘wpe atte bow ofthe age Spent engines hypethei tit 19% of Caan Acie net pm have freer ie, 405 hae lanl and 986 ave miyearice. Tenth engine thaey ing er 942 Menagement stm atures. Refer to he P Bsr. anagenen syaem Cause cxtewory and US. Chemical Safty ond umber of insets Engineering & Design a Process Pactces 2 “Management & Orersighe 2 Training &€ Communication 0 Tott s ‘SAS Output for Exercise 9.11 Cadtest” 136 nie 20 Hara nvestigsion Boa sady caused by manngcrocat stem flues, Exere 24 (0.1 The accompanying ale gives breakdown et soot causes of sample of 49 incidents. A thee ant diereces inthe pecenige case ctgorien? Tung 0S, ‘Spaces hotspots. Refer tthe Nae (Sep. 1993) ny animal and plant species “hotguts”in ret Ba Everise 275 (pI) A hase if defied as kn ae areata ispecies hay popu by the species of mere Simi olspt i He ‘pare ara that specs por. The fllwog tae ge the numb of butety heaps and numberof te olds ina ample 0288 IO-km squire ess Inthe. oy $% ofthe eas shuld be utery hoapts, 32 sould be baer cosets, withthe remaining (29%) new Tx he hry wing = 1 J worseors Bawey op ona Bary Colypas 7 eat reas a3 TOTAL 88 914 Ts at iersctons Refer. the wfc stay, Exece 92 ay, 4. Dothe data disagree withthe hy phos tht the afc 5 sully divided among the tee iets? Tt ‘nga = 08 4 Do the dat provide sutcent evidence to indica at moe than enti fal atoms elerng Yeection tun ef? Tet ng a= 05 ostoentents cutreak. Refer wo the American Wie orks Joural Uae 1986) sy of a gsroees Esetie 9.7 (p40) Condit esto Stine whe the incidence of gastrin sexe during th CF ‘demic seated to water cnsumpsion, Use = 96 Onertaion cue experiment Rete the Human Fst Pee 198) uy of ristao ls. Exrche 68.10)on Iv ac square sampling Bmuss Conic et compare the proper Whether th dee proportion if « op lana. Refer othe survey of asculra en 94 Inferences About Category Probables ina Two-Way (Contingency) Table 409 1m Optional Exercise 6.118 aed that 2 s+ anda ona ands n. Deavie the two cleo for bien 9 ~Y) Ther fore fires, Exercise 94 pi) te date se sf esperimemasm) = Yan Theoretical Exercise manda ty a 938 A gna pet of the at tat 7” press apron wtih | degre of som, Show cea hat 94 Inferences About Category Probabilities in a Two-Way (Contingency) Table ‘ne-vay) classification ofthe data Fo example, the categories fr the defective im. peller data of Example 9.3 correspond the “Values” assumed bythe qualitative vt. able, prduction line. Often, we may want to lasiy data aoonding to two disctions Te methods presented in Seotion 9.2 and 9.3 are app of elassifcation—tha is secon 10 0 guaitative va a classification wsualy ito determine whether thet ies. The abjestiveof sch ddenendent ‘To llurate, consider questionnaire that was mulled to 4 sample of 180 house. holds within 2 weeks ater a nuclear mishap oecired in 1979 on Thuee Mile Island near Harrisburg, Peonsslvania, One question cone full evacuation: “Should there have been full evacuation ofthe immediate area?” Residents were classitid according to the distance (in miles) ofthe camtmunty in ned residents atitades toward a reside ftom Three Mile Island ann their opinion on a fall evacuation, which the A summary of the responses forthe 150 households randomly selected is show in the two-way table shown in Table 94. This table i called a contingeney table it presenis multinomial count data clasiied on two scales, or dimensions. of cansifce Mile Island and responses to te fll evscuation tion, namely, distance fom Thee question Fach coll of Table 94, located in a specific row and column, represents one of the £ = (2)(3) = 6 categories of a twordircetonal classification of the m ~ 150 observations. The symbols representing the cell coats fo the experiment in Table 9 are shown in Table 95; the comesponding cell, row, and column probabilities ate TABLE 9.4 Contingency Table for Three Mile island Survey ve 72 z Tomas fall i is 5 3 acuation No 20 » 4s x ‘TOTALS 8410 chapter 9 categercal Data Anas. TABLE 9.50 Observed Counts for Contingency Table ‘asanco anne eine es 6 748 2. Toms rit Ye a ms ms me action No Ss m ry ns TABLE 9 5b Probabilities for contingency Table Distance ftom Three le tslna, res eae Tos alt Yo Pu Pa Pas m Pvacuation No Pa ba P TOTALS me Shown in Table 9.5. Thus reeset the numberof residents wh lve wii 6 ails ofthe acident and suport ful evcuation, and py represents Un corey ing cll probably. The ow tetas (designated atm, and ng) cel colomn coos er lenated at, a, and na) ate shown in Table Sa, The corresponding row an column probability tals re shown in Table 9b. The probably ls Rete somt | sn columns ae called marginal probailies, For example the sarin po is the pecabilty tht a esident favored ful evacuation, ate ve ea oral “ste probably that a espndent lives 1-6 miles rom The he feed, Tes Pie = Pu + Pa + ps and P(fovor fll evacuation) P= Pui + Px = Plive L-6miles fom Three Mie Island) Yon cons thatthe expernem we have desrited is x mutinomil expres sia otal of 150 nls and (2)(3) ~ Seateporice See te Ink eordaes cee ‘kam chose, be tls ae considered independent end he potataions we veel a remaining constant from wal al Tey Stes fhe study iso determine whether the two classifetions Sziistam Tee Mile stand an opinion on ll evaciaton se tapeedens rat te knew the dstace fs Tice Mile nd dics dat areca rove oct about the resden’s opinion vaca ation? Th a probabilistic cence, we KoO¥ Sham 2) tha independence of events A and tnplice Pla C18) = Pea) PC) Snr, ate condageney tbl aly if the to cain atons oo issco 2 erebity that an tem is clasified in any panicle ell eb he po co he comesponding marina probes Th, unr ins nen of ose Pendence, in Table 9.55, we mat he P= Pe 1 PAT PERE Bae pips tim fonh Therefor, he oul hypothesis thatthe detions of clsifcton Independent is quan othe hypehsisht every ell pobase eal B! wile{24 Inforences About Category Probabilities ina Two-Way (Contingency Table 414 Dreduct of is respective tow and column marginal probabilities. If the data disagree With the expected cell counts computed fom these probuiliis, there i evidence wo indicat hat the two directions of classification are dependent. i we were wo calculate the expected cel counts for our example, you would im. mediately perecivea dificult. The marginal probabilities are unknown and must be extmated, The best estimate of te ith 0 minal probability call pis Row itt timate of Ue th marginal coluran probability is Colum total Therefore, tbe estimatéd expected cll count fr the cll inthe ih ow and th column ‘ofthe contingency table s (5)(2)- ag ‘The general form of an Xe contingency tale (one contain columns) is shown in Table 96, When nis rows and ¢ “555 will possess approximately a chi-square distribution, The rejection regio for he test willbe 2 > 472 (See Figure 9. ‘Todetermine the number of degrees of freedom forthe approximating chi-square Aisribution, note that = re, From this we must subtaet degree of freedom be cause the sum of al cell counts must equal n. We also subtract (r ~ 1) ease we Inust estimate the (r— 1) row marginal probabilities, (The last row probability will then be detemined because the sum ofthe row probabilities must equal.) Similarly i TABLE 9.6 General x c Contingency Table 2 6 Rw Tals tom meme ow: Column Totals ” nen412 chapter9 categorical Date Anais. FiGuRE 9.3, Rejeeon region for the hi-squte te for dependence Ae eile ‘we must subtract (€ ~ 1) because we most estimate (c ~ 1) column marginal po abilities. Therefore he degres of freedom fr chi-square wil be a (Scinet Tinany independent e-W-(-)-(-1) wrr-e 4 (= e~ 0) ‘The chi-square tess summarized inthe box; its ws sisted in Example 94 General Form of a Contingency Table Analysis: A Test for Independence ‘The two classifications are independcat 1g. The two classification are dependent es lng — Atay? AR Bm Test statisti where Bing) Me = total for tow ¢ ij = al fr column Rejection region: yx? > x2, where x2 has (r — 1)(e — 1) at, Assumptions: 1. The n observed counts ae a random sample from the population | of interest. We may then consider this wo be a multinomial expet | ‘memt with r X possible outcomes, 2. Forthe 7 approxi expects ET | 0: ctr ade 9 oon eter rit resents orion on a evcunton of ‘le anc oecencsonrow er es te eet Ne fom Goer an Soluton | The first stop in the analysis of « contingency table is to calafate the estimated expected cell counts. For example, tion oe valid, we require that the estima’ be greater than or equal to $n all ells94 inferences About Category Provabites ina Two-Way (Contingency) Table | 413 ln (84)(78) 150 4368 The cel counts (op number in el) andthe corresponding estimated expected values (ettom number in cell are shown in the SAS printout ofthe contingency table analy sis Figure 94 . oe this study, the 2? wes. taste i computed a follows; 2 lu = BoP tm ~ Ena)? = Hoy e- + be ST | Elms) Evins) 3, ong Aa Em) Table of EUAC by DISTANCE evnc —ISTAMCE, Froquency| Expected’ [1-6 [7-12 [Ovent2_| Total cy 20 13 5 oy 21.2 | 19.08 | 43.66 res 18 1s 33 66 were | 14.36 | 94.32 Total 30 ™ an) Statistics for Table of EUAC by DISTANCE Seatiatic oF Velue Prob Ch) Square 2 8.2558 Oars Etter iheod at io Chi-quare 3 «Oakes Gere Fontelagenexe! Chi-Squore | 0.2687 0.8110 Pri toeftieient otoa2i Contingency Coefficient coat Cranes oroaat FIGURE 9.4 ‘SAS contingency ale ali for Sonple Size = 150 Example 94414 | chapter 9 categorical Date Anaysis Substinsing the dats of Figure 94 into this expression, we obtain (181679 672 Note tht this value, x = 2688, i shaded atthe bottom of the SAS prix, Figure 9.4. The rejection region for the test is x2 > yg = 599147, where xp 6 based on (¢ ~ 1)(e ~ 1) = (1)(2) = 2 depres of freedom. Since the compl ‘value of 1°26, falls helow this ctl valve We fil to reject Hy here sins nt evidence to conclude tha the two diestions of data clasificatton are depend 1 appears tat opinion on fll evacuation is independent of eistanve from Thrce Mik Islan, We cam ave shaded in Figur athe same conclusion by observing thatthe p-vav for the est 4, exceeds = 5 Suppose we conclude thatthe two directions of elssifcation in a comtingensy table are dependent Practically esntages of observations ‘depends om the level ofthe other sean that the proportion three dstaee groups, Tod confidence interval for the diferences, (p, peaking, this implies thatthe distribution of he por falling in the categories foe one ofthe qualitative variables arabe. Inthe 2 3 table of Example 9, ths of residents tht favored fll evacuitiondifered forthe ccrmins the magnitude ofthe differences, we coud ons = Pah (Pa = Pes) al (pa ~ Bs) sing the method of Section 710, Applied Exercises 9419 Suuty of ereine development In Teco (x. 2008, _2solgiss published thi ecrch on he formato a et ‘lines (ured moun ls) nthe eal Appalachian ‘mountains. comparison was mide oo nppes che, frock tht hive moved oer 4 ge ert dstane) ‘nen Petsyvaia and he tern Maran Rock sam les atthe mourn rn of bo aos wee called an the foto intersection ates (FIA) preserved ihn lage mineral rains was messin for each Te socom, yng able shows the namber of rock saplec ine or ferent FIA measurement capris atte two leans ‘The seoloxns teil whether the nibaion of FA trends wee the same forthe Panmayhania Naga Maryland ape sing a chinquae et of ndepemoce, LY onocune Reverie wand Topo nanpe one 20 6 ra we v 0 150.179 0 " ‘es i. eal Apaches, N23, {Give the nal and aerate hypotess rth test 1 The searchers opr thts ems? 157, Do you are’ Find te ection eg forthe sing a = 8. 4 Make the epropate concen in tne mes 0 86 rote 9:2 Hh Sect sty Acct tse eprint ural of he Neral Cancer Inia (Ape 1916 int fs igh in ber may Help ete sist sue Te ce oy die 30a fs fa gps Meas Al as mere et 4g tat cr tent ue er ook tae fd sit ffs tiber or 5 wos However he veh {atm Sher vate om group go td 8 feline ete saner ort wha ts deemine for exh goa. The ua a sana ‘he accompunyingconingeay be 5 race age | > ee eros Wah Fay ‘High Fat’ Low Fat) Low Fot! NOFiber Fiber NoFiber Fiber TOTALS Cancer Yes 27 ~» 9 4 a Tumors No 3 10 n 16 = roms Sieber & Docs te sampling speat io satis the ssumptons for s multinomial experimen? Explain, 1 Cate the expe cel counts forthe comings tableAi 9.4 inferences About Category Probebilis in aTWoay (Contingency) Table 448 Cleat the 2? ati 4. Ts hee even to inde that dit and presence! abuence f cancer a independent Test using = 8 Compa the percentage of rts 00 high fa her faufiber dict with canoer using a 95% confidence ine. al inerpet the esl, 921 nagoing gran motns tn on experiment desea inthe ural of Agricaliral, Biological and Errored Sis (Dee: 200, bins cor wer socked with var fous parsits Ce. grain et) in le wine ta eaty summer (Jan), hee bowl shaped taps were pede ‘8 pan face im oder to cape the ots Al ice taps vere hited wih ase pherorone ge Bower ap wed an unmarked sky aubesve, ome as mired witha Murescent ed powcr amd one was maed witha Suorescet ue powder The aps wete set om a Wednes hy andthe catch callie he flowing Thurdsy and Fay. The uble shows the muber o mots cared exh ap om cach day. Cos es (at a ~ 10) 4 de mine if the percentages of maths cought by the tee "ap depen on yo the week, Gworstear ‘Achesie—Wo Mark ReciMark Blue Mark 136 a 0 Sore Wo EP, wal SAakig vapor welch ke 22 Bat soutaon risk order eval ti station arom phe raps att in atl lows. At a eandom pst in te tithe itor 5 fromen nd data on situation goes are immediatly colecs The sination ster contin ntl mot, Pesmance (e.g umber of kill) s mead. A sty Speedin Haman Factors (Met. 1998) investigated ‘mete: emperaiy sopping the smaion resus in any ‘ngs in plo pertonance, Ti ete designed 0 tht ‘ome simulations wee stopped to oles stain are es data wile ethers were not sapped Each tal ws then clasifie according tothe mute of Kills made by te plo The dat foe TA rss ae summiod in the Sontngeny table below Conde! a coniageny table ims and fly interpret the es. Gomeus ‘Number OKs a oe Pei SS NoSwps 24 3688 A a 923 Grounawatercontainston i ws. Refer 19 the En ‘nmerial Since & Technology Jan 38) stay meh ere NRE) contrition in publi rivet Nev Hampsie wells, Exes 210 19), Recall tha date on wel cas pu rp egies ether or usconsaida), nd dese level of MIB low limit or dee wore collected for «sane of 233 wel ‘These aa are sed in the NTE il (Da othe Fi 10 seicted wel a shows inthe acompanyng able) mre WellClass _aauler Detect MTBE Status Private Badrock Below Lie Prive Badnck Below Limit Public ——Uneumsaliited Detect. Pub Unconoidated Below Lim Publis Uncemoliduted Below Lint Public Uncemoliuted Below Limit Public ——Uncemolited Dees Public ——Uncomsolidted Below Limit Public Unconoidated Below Limit Public Bedrock Deiet Public Back Detect ‘erst eet oceutense ad elie acces puis an thet New Hamp 4, Use the dts inthe MBE feo rate seningeney befor wel clan ad detectable MTB stats b, Conducta test astern if dette MITE stats eponds on wel class. Tet wing = 08. © Use th dat ithe MBE file creates cotingency able fr ager an detectable ME sc 6. Conduct test determine if detectable MTBE sus pens on auier Test using a= 0S, 92 Fight response of geese. Oflhire vil dling estan Alaskan estuary bas let ines it tien luge heticopters—i th ares. The U.S Fi and Wile Service commisioned 8 stuly 1 invetgate the impact thes hebcopters have onthe lacks of Pace brat geese that inhabit estar nal ene migrating (Sal Case Stier: A Collaboration between Academe and Instr, 1996) Two lage blicopters wee flown pet lly over the estuary ot ferent ies an aera! tances trom the flock. The Might espe ofthe geese {eorded ion” oe “igh ade neo mes), and later! cstance (handed of mics) for each of 464 Ielicoperovelshs were revere and ae saved in the PACGEESE fie. The dts forthe fist 10 orig ate shown inthe nex table)416 chapter 9 categorical Data Analysis LD pacctese (rt 1Ooberations shown) Overnert —Atude ota a i os 499 a 2 st ica 3 ost 338 ce ‘ oie 2k Low s re ‘ a1 ah ? 305 als 8 610 338 OH 9 a 10 iy 6 oat 8. The reseurtes categorie ale as fas: ks thon 300 ters. 300-500 mses ad 0 oe me seers, Summarize the dat inthe PACES ey ‘reaing a contingency table orate category and 3s 0 determine if ight response of the re depentsonaltade othe heer Tew eit The escarchas categorie lal stance slows tes thin 1.000 meter, 100-2040 matey, 2000 31000 meters, and 3.00 oF move meen. Suminatze "he data inthe PACGEESE fl by cca ome BEOCY ube for lateral distance catepry and Mie Conduct ates © determine if fight respunse ofthe ese depend on later distance of kecope ne Mock Testusing a= 01 The curat Federal Aviation Autocity (FAA) min ‘mum aie stndard fr ying over te etn 21000 fet (appronimately 610 mes. Based wy the ‘ous, pats a, hat changes tothe FAA ropa 2 you recommend in onder mii he Pace ban geese? fst to { seeouns 925 Subarcue pant study. The ts of ced beating pans Inlgenous 10 sobantc Finland! way suid in Arey Amnrete. and Alpine Reeurch (My 20M). Pots wom ‘ategvizedaccrking 1 peat ab, ho eh bundance of seedings (no seclings, tate sedge, og ‘shunt seeding) generate owen vegas rodction,vegeutiverepnaton poe, veptaie te Produc infec or vegtative reproduction eft, sec wg eas (0 S.5.10,1.0-80, and 280 gras, and aspre morpho (i ec ap is, wngs, fleshy frat, or smnuhoks. The da ra sample of 7 plants are sane nthe SEEDLING fe 8A comingacy table fr pln type aa ceing sae snes, produced ty MINITAB, follows (Noe NS = to scalings. SA = scilngssbundane, and SR ‘tedings we) Suppne you want to perform ci uate tet of independence to deine. neh seeing abundance deperih on plat type, Fd he expect cell coum forthe comtingeny tale. Ae he ‘seuss aired othe es sats? Reform the conigeny table ty combining th NS and SR cages ein sburdance. Find he expected cell ours Yor ths new contingency Ale the aston egied forthe et sai © Reform he comingency ae ofp bby cumbi ine the dart si and rans caters plat Find the eset cel ous forth contingent Are the assumptions rue forte tt sified? a Carry out te chi-square tes fr ndepemicnce oh otingeney tale, pate. sing «=. Wha onl IMINITAB Output for Exercise 9.25 Tabulated statistics: Abundance, Type Columns: Type us a 1 as 3A 5 “ug & se 5 230 17 an 3 yo 7 uate cust umes. Fugive dust plans seer ‘ac eqipmeat can hazardous oman eal 1a dourel of Agricul, Biological, and Eminnrsl Setences (Mat 200), environmen gies developed eel for dust puicle concentrations i pues Pe ced by a ractr operating in 8 whet Fel. The Hae ‘vee song it pall eitensth paths in he fl Aen sensing israment witha laser fea, pas the eds ofthe el, measured the porate mater he thst every 5 seco. Unformaeya ew of he TE ‘ces Were cerned. hgh than he signal ee ofthe nse. This say cece when the ‘ea sortase rm the isomer ner bea, ThE ext table stows the mumber of ceasred eum foreach ofthese atone ‘+ Calelate and compre the sple proportion of c= sored nesurements fir the i ct Hes. Do he data provide sficient cvidence onda the proportion of cesond measurements fle: ft ‘he sx waco lines? Tes sing a = 95ngency Tales with Fixed Marginal Totals 417 et swocrecrs te 927 Satvnre defects. The PROMISE Software Enginsing Bocce Reni. sed ye Sh fin Tec a suche with ds st fr baking pede sofa ir ae Stoel (See Sts in Aron, Chapter 3) Das oh ! : spas isen ved be SWDEFECTS Se 2 1456 Recall hat each module was analyzed for detects snd a elasied as “oe” if contained fective cee and B i ake if One alg fo edicting wc ot 8 mie a detects ope” (ented 5 aa ENG), whore 4 molule with atlas IS blow gro aan te mn ‘han te mtd rete sdk the ped ENG Tors san r sue yer odie, on Acomapoey ole Speier ey forthe two vale acta defective sts ad peice apenas VG is shown nthe acompring SPSS prima Inter: Mac 201 or dail provided by saa Bt Helen ‘modules? Explain . “ ‘SPSS Output for Exercise 9.27 DEFECT” PRED_EVG Crosstabulation coum [eee as a 3] a tue a 2] a Total aa] | ae chi-square Tests T Age So] Badse [baat vave | «| “Goin _| Graces) | (es Perm ON SRE | Te 7 2 Contnuy coreenore| 306 1 00 Leineod rato 28 1 a0 Fane's Seat ret a7 | a57 NotVaie Cases «| ‘Computed cniyfor a D2 table ©. | cells (26.0%) have expected count less than 5, The minimum ewpectea court is 98 +8 Contingency Tables with Fixed Marginal Totals Inthe analysis of contingency table data, one or more ofthe categories may contain tn insufficient number of observations. To illustrate, we will consider the study (deserted in Setion 94) ofthe relationship between a resident's opinion of fll evac sation of the area surounding a nuclear accident sn the distance the resient ives from ‘Thre Mile Islnd. Ifthe random sample contains only small number of residents at ny ty cannasin fueiaa iBT NINA418 chapter9 catogorical ose anaiysis TABLE 9.7 Distance-Evacuation Contingency Table with Column ‘otal Fixed aa Distarce Fan The Wie Wend ries 1% 7 oe Tos Fett te Ee eae 6 Fracuaton No 385 208 TOTALS. to 10 00 live a certain distance awe; this may cause the expected ell counts for tha istanse to be small—perhaps less than the equired 5, To guard agaist this possibility, exper {imenters often fix citer the row or columa totals. For our example, we would fit te column totals by randomly and independently sampling a fixed numberof residents in each distance group. Ths would increase the likeli cell counts would be of adequate size, For example, suppose we obtain the evacuation opinion of random sample of (0 residents in each dance prow. The rest ils might appeae as shown in Table 07. Noe the diferece between this sampling procedure and the one describe in Sesto Sc Where we assumed tht a singe rundom sample of = 150 residents wae sceced fom ‘among th population ofall people residing near Thtce Mile Ink In his section, we ave randomly and independemly selected te samples, 100 residents Irom each ds tance. Therefore, the data of Table 9.7 result from thge With k = 2 ells (apport ord nx suppor fll evacosy Aistances, 1+ ils, 7-12 mils, A chi-square nod! that the estimated expected uitinomial experiment exh ion) corresponding othe hee and 13 or more mils trom Three Mile Isl est fo detet dependence between row and column classification, ‘when citer the column othe 1 tots are fied. is conducted in exactly te sate way asthe test of Sction 94 tan be shown prof omited tha! the sais will Fosse a Sampling distribution that is approximately chi-square diebuton with (r © (6 ~ 1) degrees of freedom. The test procedure is summarized in the box. AX ‘pplication ofthe txt the comparison of two or mur Binomial proportion fl trated in Example 9, A Test for Independence with Row* Totals Fi | General Form of contingency Table Analysis: I row tous ae fixed ‘Hg: The row proportions in each e ll donot depend on the row; dat ste ist bw ons of observations in the column categories are the sme for cach ‘1: The row proportions in some (oF all) ofthe ells depend on the rows that |e sey ~ Bog Para Test statistics where ug) = Rejection region: 32> x2, where xis (e~ 1)(e — 1) at =e ta btn ie pete fe contig i i er 4.x amy dol, i nny oo dp95 Contingency Tables with Fixed Marginal Totals 449 Assumptions: 4. random sample is selected from each population for which the row totals are fined. 2, The samples are independently selected EEE | 1 conpre nepreparins of deface Inporsprodued by tee prodcton Ines, a aut eon ra engneerranaom samples roars om each ne The rumours eis te tace H marer.en | ines were tun woe 2,17, a7, respect, bo te da prove asf endencee oles dezences inte proprtons cf decor pts proceed y te he pret Ines? oer Solution | The da were entered as a contingency able in MINITAB, withthe resulting pintout shown in Figue 9.5. The objective ofthis experiment is to compare tree bona proportions of defectives. pp, and ps, based en thre independent binnmialexper- ments, each ontaning 50 observations The nll hypothess i that the proportions of defectives for she three production lines ae identical My m= m=m ost the alternative hypothesis 1g Atleast wo ofthe proportions, pp and cfr [Note tha the null hypothesis we have specified implies thatthe numbers of detec tives and nodefetives ae independent of te production line. There, we et Ho: py Pa Ps using the chi-square est fora contingency table analysis ‘Tabulated statistics: STATUS, LINE Rover STATUS Columnes LINE prrect 1217 736 mmm i fe 398 408 $05 1363 aun 500 500 500 1500 ‘$00 S00 500 1800 cell concent Pearson'Chivsquare ="4.269) DF = 2, Fovaiue = 0.116 Likelihood Ratio Chi-square = 4.999, DF = 2, P-Value = 0.111, FIGURE 9.5 [MINTTAB coatingeny thle analy for Example 95A et ed ie a 420 cChapter9 categorical Data Analysis The estimated expected cell counts are computed usin the formula Therefore im) = | ana Etna) (36)(500) (26)(s00) 7300 1500 These long with he remaining estimated expected cll counts ae show (ighlihed) ‘on the MINITAB printout, Figure 9.5 The computed value of x (also shown on the printout) is ere PRS Bap 02 2F 7-19 (asa — aa0y? a @ = = ass The lesion son ote? > yh whee yy 58017 is ted v0 ( Nfe= = (NQ) oes not exceed 2s (a 2 degrees of freedom. Since the computed valve of nd, since the p-value shown on the primo, 118, exceeds «© 05) there is insufficient evidence to indicate differen in the proportions of efestive impellers produced by th three production lines Nove that we donot aePt Heth is, we donot concioe that py bout the posi Applied Exercises 928 Performance of soter joke inspectors. Westinghowse Eletic Company has experimented with ferent nes of evaluating he porormance of slder join nacre ‘One aoroach involves comparing an individ! ape ‘ors lsifistons wit thse of the group of expe he comprise Wesinghoe's Work Stns Commitee In ‘ne experiment 183 solder snoestions were eal ty ‘he commitee and 111 were clasifed as aeepale: As inspector evaluated the Same 1S connections and hn fed 124 as accepuble OF the tes rejected bye me specter, he commie aged wth 19, 8 Comirct contingency table that summzes the ‘lasifiations of the commits el he inspector Based on visual examination of the table you con struc part dees it appa hat hee lao Ship betwecn the inspects clsiteations and the that if iferences exis, the {oom each prodvction line >—because we would be concerned ‘of making Type I eri, fling to detet differences inthe r= Dertions of defectives ifn fit, differences exist, The text concise simply Were (00 small to detect using samples of 500 impels sommitce's? Explain. (A bar graph ofthe pero ‘sje hy commites an inspector il a 901 © Conduct chisqute test of independence for Bes? ita. Use a= 05. Ctl interpre he eas of Ur test inthe context ofthe proble 929 Anew cena bonding agent. When boing ith ot lente ms macs a dy fed. Anew boding 3B sive alld Smartbond) has heen developed to cise the acest ofa dy Hel. However there is coe Ut the new bonding abesve may not sch othe wo as wll the curent standard, composite ase {Trends n loraterials& Arid Organs, fn. 200.) ‘Tests were conc 2 sample of 10 exacted ttt ‘ded withthe new adhesive anda sample of 10 ex ‘cath bonded withthe compute adhete. The Aes aeRenan dex (ARN, which messes the es be sive of bonded oth on a scle of 1 5, was deter mined for cach of the 20 honed Leth ater 1 hour of shying. (New: An ARE score of | implies al adhesive rami on the teh, an acne ofS means none ofthe adesive resins en the oth bk ofthe mn terof bonded tein inthe fe ARI categories ash fo thee Browns ~ AaesveRemnantndex Score iL 2 9 s « 2) ero je Bea 4 Expain why urging, 1 Conuct an analysis to dete ifthe dntibaton of ART scores ler forthe two pe of boing ae © Are the astumptions ofthe tt sated? Ino, how es his impact the aly of he inferene dvired Fre the est? comigncy able is one with fe {0 Manganese i tne arts crust, Te ssurce and esetal etl, manganese, has been fd aba inno es on the deep sella To vestige the elticanip esweon the magnetic age of Er’ crust onthe cca Noo andthe pana of nding manganese nls in tht lcution. cust specimens were sled fiom seven Iagnstic ape cans and the peentge of species emiing manganese cles was send foreach, The state shoe in the aocrpsaying able Is here sl sit cridence wo nate tt te probity of ding Manganese nodules in the deep-sea Fan's crs is de Pendent om the magnetic age of the erst? Tet sing SG mancanese = Es ve Specmons. Manganese Nodes Miocene resent 389) 38 lass M0 v9 Bie 214 6a — 4 24 eeCitaceom aa aut Evo Midi Crewseous 1.120 a2 # no Ieee H.W. “Tin, hn, anne in fess "AL" Ancrcn Stet Sep-Ost 197 95 Contingency Tables with Fixed Marghnal Totals | 424 9.31 Doubled crest Seldane-D,pocod by Marion Merrell Dow Ine san oven coumer dug designed to relive steering aa congestion ad other smpoM allege sini. General avers sects of Seldane-D ‘vee investigated ins double-blind, somtolled sty of ‘ver 0 piers sateing om llc hs sample of 374 patiems were given Sele D, whereas second Sample of 195 potets were given placebo tno di) ‘The numberof potentsrpeting ima in each of the ‘v0 groups are gives in dh table. Tet to determine whee the proportion of painstaking Sekane-D sho expense isomers Ina the comesponing pre bon or patients recsving the placeho a= 10. 2 sewoaneo care Tasca ” Nobeemia 7 1st Tons 5 193 9.32 Bvauanng ngn.evel programs: High-level programing languages (eg, C"7, Fate Algo and Pascal ae ee ical 0 1 computes’ opcating sytem In one sty a searcher developed smeasarement ester for eating ‘so bih-evel programming languages, Algol and Pascal ‘The eprid rel include a isbn of he relative frequency af cceurene ofthe flerene pes of sae mens wed in typical Algol and Pasa programs of ap oxime the sme size. The reported percentages ete sed to tabulate the nation gen it he able womever = al Pascal Tpeef oF 2s 204s Stoners FOR 96380 0 Bs ae Asigoment —$92)——4768, Other ror es rons toi. 9470 “Swe: Aap tom De Prskor "Oaths dcop f= Assuming fied magia forthe to programming Tmguages conduct atest 10 detemine whether the pe etiages of the citeren ype of programming ste mens ier fer the to languages. Tes sing a ~ 05, be. Coosuust a 98% confsence irl forthe iflerence Into percentages of aeigment sates sedi the to langues, ¥ f SSRI WOT422 chapter9 Categorical Dat ta Analysis 9.6 Exact Tests for Independence in a Contingency Table Analysis (Optional) ‘The procedure fr testing independence ina contingency table in Sections 9.4 and 95 is an “approximate” est due to the fact thatthe y° tet statistic has an approwmae rnive fxs tat spots of independence; theres sufficient vide? ineffective in treating the MN strain of HIV SL pve = 04076 ~ 050, can be more ue Iki shown (highlighted) ander the “Exot OF the SPSS primo in Figure 9.7. (ove: TH ‘exact p-value fora wold testo Fisher's exact p.vaue for thi cesly obtained using statistical soft Sig (L-sided column a he boone independ sao shown on the SPSS pi SHE IS Yale, 077 i cbuined by aking the hyperpooncane probaly, ‘forth coningsoey table, one with 17 unvaccinated pees tity postive #27.6 Exact Tests for independence in a Contingency Table Analysis (options!) 425 GROUP" VINSTRAIN Crosstabutation OOP RE Sane 7 Expected Court 310 __ Sigected Count 5 Tatar = Expecten Count 26.0 Chi-square Tests [Ame Se] Baise [Sansa (sea) | Gisiaed) | (tse Peatson Ch SGUTe Ose Contnuty Correctorr 036 f ikelnoad Rat 1 036 Fisher's Exact Tes 77 80 Notvaia c x 2 po tae ». 2 ceils (50.0%) nave expected countless man 5. The minimum expected count is 258 FIGURE 9.7, SPSS contingency tbl ana or Table 98 7 vaccinated patents testing postive, to ho one ti lve This table wae not considered inthe solution tothe problem sine it results in sample proportions that contact the altemative hypothesis that the proportion of postive HIV picts inthe vaccinated group is less than the proportion in the unvaccinated group.) Fisher’ exact tet fora 2 x 2 contingency table is summarized in the box. For Aol the methodod hese relecences fr this chap. Fisher's Exact Test for Independence in a 2 x 2 Contingency Table for s more general 2 > © continge ‘Suppose you observe contingency table of the form = ‘ume? Couran? Row Ta / Row Pn ™ 1 Rew? Pa " Hil Column Tt ns " f ‘Step 1 Us the formula forthe hypergeometsc distibution to find the probability of the observed contingency table (eee EVs (lee) |426 chapter 9 categorical Data Analysis | sto 2 comet sl 2 2 cogent have hes mar tos as thc ble Step 3 Use the hypergeometric formula to ind the probability of each cntingemy 2. The contingency tables with probabilities less han reel the probity ofthe observed table are atlas as comico t thea | __hypotiesis of independence asthe ebserved table, table instep ‘Sten 4 Sum te probebilties of al contingency tables tha are at east as conti ull hypothesis of independence as the observed tale: (Nat: Include the probability ofthe observed tbl inthe sum.) This sum pests Fisher's exact p-value for vented test tory to the Applied Exercises 959 Drie wer quay study. Refer othe Disasters. 2%, 200) say ofthe effects of tropical eyo on he ‘quality of drinking water ona remote Psifc sland se. ‘hae 9p, 6.-One pat ofthe study evant eel nes fa imple paper stip, Rogen slp (8) tt it for water quality in determining the prsenes fecal aces, (Notes The HS tet paper i dese em Mack when fea bacteria i presen in the water) Exch ie 2 sample of 17 water pore ze 500 maliiery oft ‘ain 3 das after Cyclone An hi the nd ws ted for fel bacteria Bah the conventional fea cali let andthe simple HS tes were apt tosucth ater spesimen. The test es are summa ine the, oy wastesr Bactoa Deieced Conventional et Blackened aa HS Tee Rest Not Blackened ieee 4. Elen why Fisher's exact ts sto be wa de termine wheter he H,S tout ult depos om wheter ‘mt bctera present in the water pecs Consrac al powsible contingency tables wih the same imaginal eal she observed tale Use the hypergeometric formula oi the probability ‘each of the abs, path cxeuing Mey ete ‘hes tha have probable less than or eal tas probably of te observed be (These ante ble that provide more coavncing evidence jt the mal hypoth of indepeene than the observed able) Sum the hyperpeomenie probably of the tables deified in part e-This sn represents he pve of Fster'sexact © The researchers concute that “the H:S teat showed ood agrecment with the convertion fecal elim ext” Do yom agree? Test using = 10, Y swoersers 13 Stare cet Rte xy on dicing des Insofar cle writen in lage era NASA crt intent, Eee 3.27 (p41), The SPS con fem he forte two categorical vara at dbfecive as and ote deve aussi EVO Ista atthe op ofp. 27 8: Show at tee ae 1 pb coming we cating cine unc) who se tial asthe cared bi Use te iypegeomerie mata ond the pobtiiy fess a ab inp «sete pies, prt ton te pa Fe sx for inependens Very your as by hcg the pal so SPSS pnt Sime the supe sie get p-value er te apts est ld be appro ed iofister seat teip sabe fete Ae dtl brn age. oth Prd Ba ‘eis § Ariel Organs an. 2003 sao a8 ing ese fret, Exeice 929 fy 20), Rel Ut het adei called Shon scone Be Sanlardconpoe chen, Te hive Remote (AR scorer oes ond wth he cw aes 10a bd wth he corp adhesive nee Me seed The coming le orth data erode & son0me 2 938 “ihe Renate Ninn 3 ae ear come fmpoie 1s 318 composing set” Ten Doerr i Ory W183 fn. 2 ae) Exon why Fiber’ exac test fer indepeseace 8 (40 stout) be ppd wo this contingency table 4 A SAS prinout ofthe contingency table ana sown tthe ttn of p27 Use the information ot tbe pinto condict Fisher's enact tes a = 5‘96 Exact Tests for independence ina Contingency Table Analysis (Optional) 427 SPSS Output for Exercise 9.34 DEFECT PRED_ Eva c t mE aa ian sol as i= ee = +m = & SAS output for Exercise 9.35 2 The FREQ Procedure iB e Table of #OHESIVE by amt a fovesive on { Eroaueney ae i ee eee ee cwrosre| a] 28] 3) 2] re ee aware] 2] .¢} .e} | rel ek S| oe Torar agi ial raps 2 fatiecint ier tao hime iy ga eee Del Wales rece Chi-square [tkel hood fot io Chi antel-Haenczel ch =Square Phi Cooffietent Convingeney Coot ictent Croner’s v MaMING: 75% of the colle have expected cots less than’. ChiSSquare ney not be a valid test Fishor"s Exact Test Tobie Probability (P) 9.0209 Pree 02616 Soaple Size428 chapter 9 categorical Data anais's | Quick Review . Key Terms Note: tems marked with an asterisk (¥) are from th. optional ection inthis chapter) Chissjuare distribution 401 ‘Fisher's exact test 425. Marginal probabilities 410 One-way table 401 Contingency table $09 Fined marginals 418 Maltnomial experiment *Probability of the Expected category count Dependence of two 401 contingency table 422 “405 lassifcaions 409 Observed cll count 408 Two-way ta key Formulas ‘news te a Confidence terval for Tos Stat Fin) = m0 ~ hypothesize value fn Hy 412,418, | LANGUAGE LAB ao Valeo main proba ype in x Chiaguae Tex static wed in analy of conn ane Namber of cred cucomes incl foe way taeai. Supplementary Applied Exercises | 429 Ba) ei specie umber foes in elt oneal when ys tne BEES roti an oxen tm nd clo wy cemtingecy table ae Number of observed ovzomesin ow ad clin) of trv comlageney ab Hi) Eamon) Estima ened amr fone iro and column two-way eomingny ube i ees Tot uber of eukomesinson of acon bl ip neers Tol mnie feo clon jf contingency ae Chapter Summary Notes * Maltin + Popes ofs multinomial exper lata ae qualitative da that al into more than two categories, classes, or cls nent (1) ideal ia, outcomes remain the same from tilt il (rls are independ, 2) k possible cateomes to each tial, (3) probebilties of the the variables of interest re te ell counts * Aone-way tables summary table for a single qualitative variable * Atwooray table, or contingency table, is suri table for tivo qualitative variables * The chi-square (x2) statistic is usd o test probabilities associated with one-w and two-way tables * Conditions required for a valid: (1) multinomial experiment, (2) sample size ni lage—saisfed when expected ants ue al greater than or exalt 5 * significant x2 cs or arvo-way able implies thatthe two qualitative variables are depend * Giesanre tests for independence cannot be ase infer that «causal relationship exists between the two qualitative ‘arabes, * Fisher's exact test can be applied 2X2 or more gener 2c eontngeny ables, Supplementary Applied Exercises Hite Exercises mad wit an seri) rau methods fo the opinal ton ia hs caper) 7% aporowriate hosp acssons. New Eland Jour tal of Meicie stay (Nox 1, 1986) fund hata si Smal poron of acute hospital cae is reper fo be Ioccesry. The pias Who condoted the ye ‘ewod the medical ens of 1132 points hospitalized 458% different eatons actos the oun. Ovecl, 0 ‘the admissions inthe sample were julged to fe apo rie and 23% were demo inappropriate, when the "emitng 17% could have ben aided by these of Petry sngry Let py, pap represen the te Per Satges of hospital aunssins inthe tre caters ‘errs, inappropriate, and svonsble by abatery ‘rae, respectively, 4 Eximate py, 25 and py wih 955% conten intervals DB Testhy: = py = slaps A ing a = 1 52 comeessed work weeks. Compressed work wees ace eds “akeratve work schedule in which wae tae between the numberof hours worked per dy. and ‘be mamber of ays worked pe week in oder w wok the ‘ard ouaer of woehy hous in lee than 5 dy Aield say was condueted alae, midwestern, con- tiouspocessing (7 daa’ hous) chem pa that Sd caperimentd with fur diferent work schedules 190 ‘Stich were compressed Two i2hous ed ike (ay evening mangro Three Show oting shits Two Si hundred sve "rank the for work schdles inode sccompunying ible gies the mimber of fst place rank ings for each schedule. there siicent evidence ind ‘ae tha the Bort employees have «preference for one ofthe Werk schedules? Testing = 01 oe hourly employees we & wornscreo Srourined Ghourotatng 2 hour feed hour rolatng 0 20 938 Computer abuse. A.tyear uly of compe sete was recetly completed by reserchors. The tember of out Iypes of bus epee and verified by year are sown in the neat tale. Do de data provide sufficient evidence to indicate atthe proportions of citfeent pen of sbune ane changing over ins? Tet ing a ~ 08430 chapter 9 Categorical Data Analysis, emi $f) ae © aause Tee Ae The of Unautored Finareis! formation Usp of Froud or reperty ——normstion vandalism) Tora ua 5 & 2» Year 3 Is aT ‘ 1s 9 61 ori a 3 3s wr 999 Atomic weapons exposure. Researcher atthe Ok Ridge Tennesse) Natowl Laboratory hve developed a ale ihm 40 extimat the marters of espoced ad ences ‘es of thyroid cancer eccring in the Htime ofthe ‘exposed atomic weapons est the Neva Ten Sen the 19505 Heath Pris, lan. 1986) Ofte app mac 2100 pram oe wes a ‘out, 58 were expected eo evel hyo cance the remaising etme. According to the slg, the $8 ‘uss canbe catsperized hy 4% an eel of ralsion Sappose thatthe data feet random sample ot 38 thyroid cancer pains selected fom the caget popu tion. Condit ts determine whather the tae de tons of elasiicaon, sen and dose at tine of expoune ue independent. Ue a = 01 Barone Genser Mele Femoe = TOTALS lesthan 63 Dose, rad 1-10 » % Hormone 30 b See Zag EA, ad Morn MD Till Smee he 940 esti use in orchards. Four reties we in dermant Caltoria onchands ae chlopyior, dizinon, met ‘ahion and parathion Eninnimenul See & Teche (Get 1993 reported the umber of applications of tee spray chemical for Beary t ne 1980 in alferne The dita for cach of tice type of ait or mat har se shown nthe accompanying able (Pethion ha ics bs tamed for we om didn fat and ot woes) Coed atest to determine (at = 01) wheter pes ‘ie wsed depends on ype of erchan 1, Becawe ofthe lige mmber of pesticide applications reported, the ttl sample size fr the test pt 8 cxremely lage (n= 417.697). Consequealy a ‘watt sgnican” result may not be "pactealy signiican.”Psform an analysis to show he mat tude of ifrences in the rates of metho pi liom forthe ve rca pes. Pesmicbe Choma ‘Aimones Peaches Necinbas Chores 519979 usa Discinm ———s0n93$ ast som 210 5198 190 13606453388 aa ‘rpimpoepatemeicides ted on aman echurds nde Son ‘eau Vly. Caifna” Eminent Scone & eld iL 2T. Ne 10,06 199 236 Tae 9.41 CASE tecnaogy. Refer wo Execs 63 (, 248) a he ersten dy of computed ctw ee ing (CASE) tectology Jewral of Stems Manageme Sa 19891. vey ashe ech na sample of CASE Se a often hey me da ow diagram a eo ad Sti they are wth he devin hating ech ‘we wing. Ths peerage sponding each of he el the comingency ble ae given here Analy the SE) yc SSN LOD CASE wer pute inch wa. Bease BS ae Win asia 4 8) pedal ‘ser Ststecion of esan ‘Carag Techies ee Daie —Sattit unsatisfied TOTALS Fow Dagan Aeon 3 oS Most ofthe time 313 2 us Orcasionsty 343 abi 380 Never as 2 u roms a2 75 1000 Sew: Rei, ald Main, M "Sc aan Wei he” Jama Se Manageme pamsice,8 Use a ciaquae tet dtemine wher tere is Aiference between the peptone of males and fe ules with carts lathe snot Use = 01, timate the dierence hetmcen the proportions of ‘mules and females with car tate in the et. Use 3 7% confidence inert theres © Apply Fisher’ exact test the da Compr he de H wecowe - ~__Wpeot commer @nSeconds) a Secon) Recs of Yn Is 2 roms 7 4 922 Selenander sno wounds. Dear enenyrecognion DER) 948 isto em ost by natura cokes forthe aes sive behavior of bids, mamma, and ans when te ‘wal bounties ae viola by one cf thou pee. DDER is fen flowed by elated tacks en he nang sina. recent sy explored the possibility tha dee Sash slimanderemplins DER hy asng chemical si "also dings fala from unfaiiae salmmsIn ‘scaled ees slander ill temp an op Pooen'ssnut—an ijry tht eval ec eames abit ease prey. mates and erent, One rot be sty focused en compari the rpartions ¢ ales ond teas eating wounds iste smut One {or led, an npc for car sein the sot. The el are showa in the tbl elo Gon = = ise Fone TOIAIS Sar ison sn aay oscar tise in SII roms ena , Supplementary applied Exercises 431 20Secords) Tors o 2x0 eo tie compression. Vio engineer xe tne com ression to shoren the tne reuired for broaanng 9 ‘clevnion commercial. Butcan shore commecihBee fecive? To answer this ston, 200 college sedeats were randomly divided in tse ponps, The ie ry 57 sudems) as shown a visctape ofa television pe am that included 4 30.second commercial the second, withthe 24-second dine compressed version ofthe com rea an the thin rnp (9 sale) wes shows 0. second tine compressed version ofthe comet Two ays lier ving the tape the te eroupn of students were asked fname the rind tht Was averted, The ‘umber of stakes ealing th brant name for each the thee poupsate given inthe tbe athe wp ofthe page, 4 Do te dat provide suficin evidenee (at a = 5) that the two dictions of clanfcation type of com ‘mesial and recall and ame, re depen? It 1 Const 955; cenfdence inter orth difeenc be maa 21sec incomes commana, Coal vs. nuclear onary An elec Imus choos tween fro techology ope 1s customer he fits: conl er lear energy To ase the atts of kcal lize the Power eompany confucted s public opinion pall. Four sectors were considered: news medi, cau! mine ution avroamettlss abd conservation, and lca groupe Fifty pesos were randori sample from each sector / and asked to give their opinions. The vests of he sty ane show inthe ble we the hte ofthe page 4 Dows public opinion cepaing the choice of fie technology opbons for enertng eeciy ier fats tothe tet, pa among the our groupe? Tes uxng a = 1, S commue 2 ‘News Coalviners Enivonmientalst and Loca Meca “Uno” comeratonists —Gioups—TOTAS Spee Coat Option 24 2 u pe » Suppor Nuclear Option 18 2 16 3 ” Nerat 6 a 2 2 Ii} Tomes so 0 so 50 200 |432 chapter 9 categorical Data Anabsis (055) isa compueriza! notem dengncde ea eae use” bags ofthe candy (eching Stas, echt forconpre)ndmetnctinge ke Cohutta oan ee aaa cen the Ur and he stem A sate ighues sees mam ‘evel instal a DSS 0 help meme isco pad ne —————— corey mien nt Ota cme Gr Bony Yaa Med’ Compe Geis TO wermy selected 151 of the mow wsely eoemeeeg Problems tat couidbe traced deci wie DSS ances Neco it ke conclu fom ihe da inthe able ate pan ies pope or aa DSS components? Te Ms. 9-48 Accident rate of workers. Does the upensity for woret 8S iniarics depend on dhe length of time that a worker is om ‘manufacturer gave the results shown in the tble fof ‘duced to compare the fidelity and seectviy of radio me 29 ACCIDENTS ~ cceivers. Thirty receivers were tested snd classified ow oe Hau of SAP L oy 6 s Letra Megius a Ae ee "sityand kv? Use ihe sents 8 Doth tinny athe proba of woe SB r3010" lea dots cipher in some time periods tant Tex wing = 10, aa 1 Date is provide slicient evidence oni the probably of sn aciem during he st 4 Fidetty La 8 sit sara han ding the ft 4 hows! T ma sing = 10. Gene Tes Py: py = 5 were ‘rohit of an accent dig te at 4D. STATISTICS IN ACTION The Public’s Perception of Engineers and Engineering 1.203, the Amoran Asocodon ol Engeetng soca [reteset need ae art eters a irstn ergeeing he pany joc a {AAES), with a grant from the United Enno rors een pis rows 8 -ompare the 2003 su sults to those obtané oa noes Aas slow et eae aoeStatistics in Action 433 TABLE SiA9.1 Some Questions asked in the 2003 American Perspectives on Engineering Survey (Qk: Pease tllme how auch you aie with the following Nata show cana 2 Engncers are largely responsible for cur high standard of in 3, Engines ae response fr creating ings tha ee barf soley. lanorer: Ager sional, gree samehay, disagree smewhat age strongly] so fave had) om or dar what wold cause yo oe pleased if our won cer chse engowings a potesion” (Q8: Pls meow insted you hak emgnces are nthe creation fhe following: 2 Highways bids, an nels Spacer © Computes 4 Inte, © Necletechologies (Answer: Very mold, eet iol nt ery mold, mt al noted) (4 How infred you shou enpnceting and engineers (QS: How key ae you o read aout the devloeent of yb uot ta se fst ‘lle intend of pole” Unsver Absouey coma very les somewhat ies nt a al Wels) (What media rcs do you use 1 follow ses mover Cable TV ne, lca TV news, lee For this Statistics in Action, we'l focus onthe responses to questions Q1 and Q&. A summary of the Tesponses is shown in Tables SIA®.2 and SIAP3. Cleat the responses are categorical in nature, Conse. ‘uenty, they shoula be analyzes using the methodology of tis chapter. Fist consider te responses to the statement “Engices are responsible for creating tings that ae harm fulto society (Qi. An overall summary othe responses, shown in Table SIA 2a, are inte form of @ one: may \eble- We can use these result o determine the percentages in te four response categories aaree strong, ‘agree somewhat disagree somewhat, disagree svorey are diferent. An SPSS one-way chsquare analysis of the deta is shown in Figure SiA9.1a. The test statistic (x2 = 150.8) and pwalue of te test (p-value = 000) arebot highlighted on te SPSS printout Thus, theres Suficientevdonce lata ~ 01) toindate thatthe true Dercentages i the four response categores are ciferent. The sample percentage shown at te top of Figure ‘589.12 roveal hat the majority o the public over 67) lsagee wih the statement. To determine whether the level of aereerenticisagrooment with the Statement in question Qib depends nother factors, such as gender, education, and famiienty with engineers, we use chi square tests for nde Bencence inthe two-nay contingency tables show In Tabs SIA. 20 SIA 2c. and SIA 24 The results are ds Blayed inthe SPSS printout Figures SIA9.20, SAO 2c, and SIA9.20, The pales ofthe tests Cughlghted) are 23,350, and 073 respectively we conduct each ios at « ~ 01), we fal to reject ther hypotesis fn ependence in each case. Consequently theres inuficient evidence to indicate that the percentages ofr Sponses inthe four tb categories depend on gender, education and failiarty with engineers. Anayses such 2 tsied the AES to concude intel fia eporton the survey hat-attues toward engineers are main ose and do not vary much when looking at familiarity with engineers, gonder and education” Now, consider the responses tothe question “Wat media sources do you use to follow stores about eng Dering and engineers? (G6), Does the percentage of adults that flow engineering stories onthe Internet de Pend on famianty with ences, sence, education or age? To answer this queston, we conduct cy square {23s of independence onthe summary dala in the two-way contingency tables spayed in Table SIA93. SPSS Betouts forthe eontngency table analyses are shown in Figures SIA@ 2a, IAG. 2b, SAD 2c, and SIAD.20 = ‘pectively Note tht al four tests have palue ~ 000. Thus the proportion of Armen adults wo use the In ‘ere o follow engincering stories early depends on famianty with engineers, gender, education, and age.434° chapter9 Categorical Data Analysis 1) nammrucars GY ranmevare 29 wanmeo1o 2 wmrermergea wremner ass wremerase D mrerner gen TABLE Sia9.2 Summary of Responses to Statement “Engineers are ‘responsible for creating things that are harmful o society’ (Number responding in each category) ngrseSwongy Agee Somentat Disagro Somewhat Okage Svongy » 22 an as Category Fila wid Engineers (n = 98 nesses) row 0 eapnas now a eas TER ‘Agree Some Somewhat n au Dona gre 0 $9 ©. Cae er (e = 000 responses) ak Fema ‘Aare Siete 3 168 Domo agre 25 or 4 Cnegori by Edecaton (x = 97 tego) Nig School cess Some Colege Colege Grad Post adie Agree Stongh/Somevbat 100 100. ° ° 10 1a iat honey a ep Pa rhs as TABLE SiA9.3 Summary of Responses to “What media sources do you tse to follow stories about engineering and 7 number respondingin each category) ne one eneneers? Nu 1 Caegoried by Fania with Engineers (n = 71 eegonses) row Orne row at ast Vere Flow on emer 5 1 Do nt follow on tient 153 90 ‘. Caegovied by Gender (n = 878 expoases) Wal rena Follow om internet = 4 om fliowonItemet 7 “s Categorie by Eduction (n= 850 responses) ess than Colege aa ‘ohage Ga = Were Foto on tert x a ona flow on tenet 43 26 Categorie by Age (x = $89 responses) 18-2 yeusDo-tayeas ae ahyeas Gyro Oe alent a s 2B a Do notfollow on theres 121 190 2 195, SATE Sees amteene em 13.200 StandOver (n = 965 responses) i Response T Canaae Fi percent ea TTA SY a3 Tor see Somewhat a | 220 rages Sonentat | ar | ga es Disagee Strongly oa | 955 0 Tota ses |_ 1000 Response Guenein [Spain eS ane Aare Somentt 22| a3 Osage Soneunat a Deapee Svongly aa} 2413 ta 238 Test Sone moe em @ 3 Asying. Sig. 000 2 Des (08) have expected requencis es nan 5: "neinmum expected cel hequency 82413. OtpResponse *knowEng Crosstabultion TET or vnowo | Teast | tora TERESA aT Donat ree se 63s Tot na] 73 | _s57 chisauare Tost [Asm So | batS9 | eactSq | Fone vane _| ot | “Yhiteeay | Grsneay | Crauea!_| erabaptey PaaS | ST ro Continuity Corrector® 4.826 | 1 026 Uietnou Rate sar? | oe) ozo | ors Fisher's Exact Test | 028 ons Asecizon™® see 4 = NotVata Cases ost 2 comp » Ocets (.0%) nave expected count less than 9. The minimum expected count is 64.47 The tondarzed states 2 290 FIGURE stag 1 sls reas in Table SIAS.2— "Engineers ar responsible for creating things tht re harmful society 435 SHEENA HISAR |«: Categorize hy Gender (n — 1.00 spose O1bResponse * Gender crosstabulation Pees \ T ‘Asymp Sig | Exact Sig, | Exact Sig Point | Seer near are 1 351 art 193 | ) have expected countless than 5. The minimum ex I eg counts 196 22 iit © The standardized stabstie i 993 4. Caeporised by Eduction (n = $97 responses) | eroResponte-Eéveaon crotstabumon ee i Ean | HS [ae cte comer [reaaa | ro i TRE Aa TEP ae peas Peas toa iii | vonmcaee | ws! ol] as) a] Hii re avs iM | aa ok | | H | ‘Chi-square Tests “ome S0 Betsy [Sana] Po | vue | | "iteay | red [SESE | Oy | Peas OEE | — eon] — se | ws Hi Uieircosraea | rio] 3] am] 8 i ae ara 1 052 055 028 004 if Not vata Ca: 997 | l s tan 5. The minimum expected counts 61.79, ©. The standarsizes statistic 1s 1.943 FIGURE SIA9.1 (continued) >. ‘atleast one « Wy trom Figu 89.20) 23.2 (rom Figurefe i on Statistics in Action 437 (Categorized by Funai with Engineers (n = 871 sponses) ‘bResponse ' KnowEng Crosstabulston Tore wo | ett | raat wainmvcoveng | 190.0% ‘chiesquare Tests pitas any Bea BOE Categorized by Gender (@~ 975 responses) O6Response Genser Crosstabulsnion [re au Chusquate Tests ee Norval Cases 1 Ocels CD) nave BoETte countess han. The manu exectes court FIGURE SIA9.2 [SPSS ali feats in Table SIA?3—"What mda sores shout engineering at eninge” At east one engineer use the Internat, compared to 32% of those who co not know any engineers. similar ly Grom Figure siA9 2b), 19.9% of males use the internet compared to 9.1% of the femles; (rom Figure SIA9.2c) 23.2% of college grads use the Internet compared to 10 5% of those without a college degree, anc {trom Figure sia9 20) 21-22% of younger adults use the internet compared to 9% of older adults illl i 438 chapter categorical bata analysis Categorized by Edveston (n ~ 870 responies (OsResponse ' Edueation Crosstabulation ee eee Categerizad by Age (0 ~ 84 responses) OsResponse Age Crosstapulation seitian | Colegearsa cotepearaa_| “orngner | Total DaREspoTsS Use HEME CONE = oo 135 within Eduction 105% gam | is6% Donotuse Count 482 265 TIT % witin Eoucation 695% ream |e Tota Count 505 BE 50 *% within Eaueaton 100.0% roo0% | 100.0% ©. ets (0%) nave epectes counties than §The minum expected count ot Pearson Cr SqUae Luketnaod Ratio Lngar-y-Ur Not Vaid Cases 849 2.0 cels (0%) have expected countless Man S-The minimum expected count fs 22 13 FIGURE S1A9.2 (continued) Seaa_| 45:55 [ear wer ‘GaResparse Se REMEt COUT 7 2 TT within Age ax | soa | osax ut 180] 221 195) r _% thin Age {776% | sos%| 94.7% Tokar Count 245 2a 206 tin Age s000% | 100% | 100.0% Chi-square TestsCHAPTER \ Simple Linear Regression OBJECTIVE To present the basic concepts of regression analysis based on a simple linear relation between a response y and a single predictor variable x CONTENTS: 10.1 Regression Models 10.2 Model Assumptions 10.3 _ Estimating and ,: The Method of Least Squares 10.4 Properties of the Least-Squares Estimators 10.5 An Estimator of o? 10.6 Assessing the Utility of the Model: Making inferences About the Slope 6 10.7 The Coefficient of Correlation 10.8 The Coefficient of Determination 10.9 Using the Model for Estimation and Prediction 10.10 AComplete Example 10.11 A Summary of the Steps to Follow in Simple Linear Regression STATISTICS IN ACTION can Dowsers Really Detect Water?440 Chapter 19. Simple Linear Regression One of the most important applications of tatstes involves estimating the man Yale ofa response variable or predicting some future value of ‘lg oF set of related independent® variables 444. y For example the manager of « wachowse tha uy automated vehisks nigh | ‘want to relate the congestion time (the dependent variable) 0 such variables | the numberof vehicles in operation and the size of the lead being moved (te Independent variables). The objective would be to develop prediction equatin (ce model) that expresses asa function ofthe independent variables. The weal on j able the manage to predic for specifi values of the independent variables ana. ‘imately, to use knowlege derived from a study ofthe prediction equation t nt Policies to minimize the congestion Gime. based on knot | 10.1 Regression Models | } As anoter example an engineer might want to relat the rate of malfunction yal 8 mechanical ascembler to such variables as its speed of operation andthe meenar operator. The objective would be to develop a prediction equation relating te de Penden variable tothe independent variables and to use the prediction ation Predict the value ofthe ate of malfancion 9 for various combinations of speed of op ‘ation and operator, ‘The models used to relate 8 dependent variable y to the independent vartabes "ate calle regression models or linear statistical models becuse th) expres the mean valu of fr given values of re stof unknown parameters, These para proceso be explained in Section 10.3, The concepts ofa repression analysis ae intealuced in this chapter using 2 Vx) simple regression mode eth flies 1a singe independent vale We : sitar bow of hs model ase da wing tele ke ae wil examin in el he fect pe of infretes tek ean nen be at fais. In Chapiers 1-12, we wil apy this Knowledge to elp us undersea ve teoretical and practical implications of a multiple regresslc, analyse problem of ain 0 moo mae nde } | betnon 10:1 i Th vrata be pred or moldy cate te Gepeent Tinea foetion of sets are estinated eo sample data usitg reaponse senate Devas 00/8 ed ndependon variables a ee cece ‘The preceding chapters provide | Although the ‘and, In rany instances, wil ease you to find he raeans an vie cor of eae am pling distributions, ‘Your theoretical knowledge will be useful for “another reason, The theory of st i hysis engineering, economics, et.) to the soon of ee maid obleam i a ihe he wor adie her tan pte ee, rile oh rae. an nde 2 cE ike10.2 Model assumptions 441 from assumptions will ect the resulting statistical inferences, snd you must be able to adapt the model and methodology ta the conditions ofa practical problem. A basic ‘understanding of the theory underlying the methodology will help auto do this 10.2 Model Assumptions TABLE 10.1. Compression Versus Pressure for an Insulation Material Ymsunsrion soosnen y FicURE 10.1 tenn simplify our discussion, we will postulate the following fictitious situation and a set. Suppose thatthe developer of anew insulation material wants to determine the amount of compressa that would be produced on 2inch-thick specimen af the material when subjected o different amounts of pressure. Five experimental pieces ofthe material were tested under dillrent presses. The values of + in units of 10 pounds per square inch) andthe resulting amounts of compression y (in unis of 1 inch) ate given in Table 10.1 Apt ofthe data, called a seattergram, is shown in are 1h F Suppose we believe that the value of ytd to ncrease ina linear manner a. ae ‘reases. Then we could select a mee! relating 10 x by drain fine through the point in Figure 10.1. Such a deterministle model—one tht dows not allo. fre of predition—might be adequate if all of the points of Figue 1.1 fell on Tine. However, you ean see that this idealistic situation will not Table 10.1, No marer how you dea a ine through the points of i some of the poins will deviate substantially frm the Fite Hine The solution the preceding problem io construct a probabilistic model rl ing» tox—one that ackosledes the random variation ofthe data poinisabout Tine (One iype of probabilistic model, simple Hear regression model. makes the as sumption that the mean yalue of fra given Value of ¥ graphs ax» aight line and that points deviate about this Hine of means by (positive orn qual to y= Boron + ised I Toe the data of 0.1, at least where yan are nknown parameters ofthe deterministic (nonrandom) potion « the model. IF we assume that the points deviate above and below the line Of with some devistions positive, some negative, and with Ee) ~ 0. then the mean value ofy 9) = E(B + Bix + #) = Bo + Bix + Ble) = By + Bux = NSE i if‘graph of he dats pins of Tole 101 andthe ype line ‘442 Chapter 10. simple Linear Regression A= levee Therefore, the mean valve of y fora given value of, represented by the symtl #9)* graph 38 traght ine with yinerept equal to land slope eal to By A mph of th hyphal ine of means, E(y) = By + fy, ik shown in Figure 102 ‘A Simple Linear Regression (Probabilistic) Model Y= Bot Birte where ~ Dependent variable = Independent vaiabe E(s) = fy + Byxis the detemninis Ps tic component (the equation of 4 straight line) (epsilon) = Random error component Br (be H-0) = Malar fe ie, point ih he ine ns cu toga To Bi (beta-one) = Slope of the line, ic, amount of increase (or decrease) in the dermis componcty rset Tae he et we) eae Pipe car aason ml ast da we mein {or the unknown parameters. fy and &,, of the line of mean, Bly) = By + Br aint ot yf ld on te ee Ia which num pend on he pay See ee Somer fn mate ce macapee a summed ae bc weve aac a | Assumption 1 The mean of the probabily distibution of i 0, Tha 5 te ors over an infinitely long series of experiments i O for Sh Sting ofthe independent variables. This assurmeionigdce tha the nea vit of, BU), fora given vaheof is Ely) = By t pre Assumption 2 The vatane ofthe probability disttion of ei constant forall seuings ofthe independent arabe x. For our sraight line mode tht soup tmcans tht te vaiance of eis equal toa constant, ay, of al alae f= “Themes alo oy hen veo she ese ton come cnt hen he mode onan ‘laevis the noon and wprccat ya paar , RGURE 10.3 The prota dir 103 EstimatiFIGURE 10.3, The probability distnbutin of that the elatveFequency distibutions ofthe errs are normal wih "mean a «(all the distributions shown have the sme amount of spread or variability) The straight line shown nFigne 10.3 s tbe mean vale of for en value of x. (3) = Bo + Bux Various techniques exist for checking the validity ofthese assumptions, and there aro remedies to be applied when they appear tobe invalid. We discuss these tech higus in detain Chapter 1. In actual pease, the assumptions need mot hold exacly for least-squares estimators and tes sttsties (to be described subsequently) sess the measires of reliability that we would expect from a regression analysis. The assumption wil he satisfied quately for many practical applications 103 Estimating Ao and f;: The Method of Least Squares ‘Tochoue the “bes-fting” line fora set of data, we must estimate the unknown pa rameters, Bp and B;, of he simple lincar regression model. These estimators could be found using the method of miximum likelihood (Section 7.3) but the cise rmethod—and one tat is intuitively appealing isthe method of least squares. When the assumptions of Section 10.2 are satisfied, then the maximum likcibood and the Jeast-squares estimators off and are ential * The reasoning behind the method of lesst squares can be seen by examining Figure 10.4, which shows a lie drawn on the scatergram ofthe data pints of Table 10.1. The vertical line segments represent deviations ofthe poins from the line. You an se by shifting a ruler around the graph that ts possible to find many lines for hich the sum of deviations (or errors) sequal 00, butt canbe shown that there is ‘one and ony one line for which dh sum of squares ofthe deviation is minimum. “The sum of squares of the deviations is called the sum of squares for error and i denoted by the symbol SSE. The lin scale the least-squares lin, the regress Fine, the least-squares prediction equation Sern 102 a vile Ie eorptens ena ts,444 chapter 10. simpie Unear Regression FIGURE 10.4 rap showing th deviton ofthe Points shout line To find the least-squares line for a set of dat, assume that we have a sample ‘°F m datapoints which can be identified by comesponding valves of + and yay (i 91), Gas99)--.. Gi) For example, the m= § data points shown in TABS 101 are (1), (2, 063.2) (4,23, and ($4) The saight-ne made forthe repame yin terms of xis Y= Bot Bese The Hine OF means i E(y) = Bly + x and the fitted fine, which we hope to fins represented 8 3 ~ Bip + Bt. Ts, 5 isan estimate of the mean valve of» Et) ‘and predictor of sme fate valve oy: and Fora given data point, sy, he pit, 9) Dredicted value of y would he obtained by sak 5 Bo + Bx And the oad 8 are estimates of and 2) the observed value of is y and toe tung into the prediction equator vation ofthe th value of fom it Or 5) = by ~ (Bo + Bix) Taal i of Sates ofthe deviations of ey values shou thie prec vols for all ofthe data pins is Prodictd value is SSE = ¥ 11 ~ (Bo + Bray? vein es Bo 10d By that make the SSE a minimum ae ealled the Heeger Srdmatet ofthe population parameters Aly and Bi, andthe prediction opine Y= Bo + Avr iscalled the leastaquares ie Definition 102 ‘he teemequaet We one rt het sr SS ten yor saat ede ‘The values of Band iy th ince 858 ~ S,~ Gio + aah are obtained by seting the two {0 Dand solving the resulting Toillustat, we frst comp artal derivatives, ASSE/apy and aSSE/ap, etl Simultaneous linear system of least-squares equation ute the partial derivativesSolution ase. A 7 Bo Budi) a SES ats ~ (Bo + Brand Siete CRs Sexing thes pl dvatves il 0 Oand stoping, we obtain ie et gues gutter ABs= So -nb- Ay Sui Ada ade- fa Solving this pur of simultaneous lincar equations for By and fi, we obtain (proof ‘omitted he formas show in he box. Formulas forthe Least-Squares Estimates ve Stope: Bi = 5 pimereep: By = 3 ~ Bik ss. Siu Sample size the las squres estimates of fy ad A for he dts of Table 10.1. Thea so pu Ss Gives practical interpretation ofthe rests 4 Preliminary computations fr finding the least-squares line for te insultion come pression data ae contained in Table 102. We can now calculate ‘ine mans wile etn fen is pola. i othe hai co Swen he us nator nln le meses he sae, when hey sin we will rte som eannes in duevaaitin | : : :a || , 446 chapter 10. Simple Linear Regression TABLE 10.2 Preliminary Computations for the insulation Compression Example’ | ae 1 S85 10 = | 7 and he y-imercep is | Pek ti | Frat Bem —1 The graph ofthis lines shown in Figure 10.5 Th observed ad predicted values of y,the deviations ofthe y valves about he Frotcted valocs mod te suns ofthc deviation fx th dss og ae 101 8 seenane, abl 103. Note tthe sum of square fhe dviton, SSE ie 10 scp iat ele of SSE the woud be baie by fine ony citer posse Straight ine 10 the dats, “lations required to bi jhtfocwand, | cular, the proces is laborious a | sizes lage. Most engineers and | linear retession. The SAS, Ml sin Bo, By. and SSE in simple linear re "become rater tedious. Even with the use of i! ssceptible to exe, especialy when the sample Scientists rely on statisti software tor the sit INITAB, SPSS, and EXCEL outputs forte analysis FcuRE 10.5 edoFIGURE 10.5 odes 103 Estimating and The Methoct of Least Squares | 447 TABLE 10.3 Comparing Observed and Predicted Values for the Least-squares Mode! 1 1 6 ‘ 16 3 2 ° 0 5 4 aa ays 6 6 Sumofenas = 0 SSE = 140 the data in Table 10.1 are show in Figure 10.63-d. The val highlighted cn the printouts. (Not that these values agro ex culated values) by. Our interpretation ofthe least-squares Slope, 2; = 7, is that compression y will increase 7 uit for every I-unit increase in pressure x. Since y Is measured units of 1 inch and in units of 10 pounds per square inch, our imterpreation is that Compression increases .07 inch for eve increase in pressure. We will auach & measure of | Section 10.5, F The least-squares intercep, y = ~1, is ourestinate of compression y when pressure is set at ( pounds per squre inch, Since level of compression ean verte negative, why does such a nonsensical result cecur? The reason i that we fae attempting to use the leataquares model to predict y fora value of (x = 0 that is ouside the range ofthe sample data snd impractical, (We have mae to say bout predicting outside the range of the sample dat—called extrapolation —in Seetion 10.) Consequently, Bp will not alivays have & practical interpretation, Only when x = 0 is within the range ofthe x valves in the sample and is a prat cal vac will Jy have a meaningfal interpretation, of By ys and SSE are ly with our handel 1-pound-per-square-inet init to thi inference in ‘To summarize, we have defined the best-fiting straight line to be the one that satisfies the least-squares criterion; that is, the stm of the squared ersore will be Smaller than for any other straight-line model. This line is called the least-squares Tine, and is equation i called the east-squares prediction equation, anas on dye att sess iean || 448 chapter 10. simple Lnear Resression ] a The EG Procedure Frodet noe Dependent ar ia Number of Observet ions Read 5 Ninber Sf Observations Used 5 fralyais of Var tance gm of ean omroe or smorse saiore Value Pro F Mode 1 4-80000 19.88.0354 Error a ocateer Corrected Tota! : Boot nae 4.50859 o.ais7 Dependent Moon 20008 orest ~ Peet Bar 30.2680 Porenoter Estimates Parmeter standard Veriable or "8oRTagte Error Vole Pr > Het Intercept 1 -2. 10000 863503 0.16 ang Preset i O:70008 oss | Se Sa? 8. MINTTAB out he eaceasion equation Genstont =9:1000 0.6351 -0.16 v.00 Press x POSTON olisis sige Oceee | $9) 0.50380. Same ag Oo pa tealy ie 6h source > Mreil, ; Bageession 44.9000 4.9009 Residual Reese 3 41003 osec0 FIGURE 10.61033 Estimating fy and py: The Method of Least Squares 449 SP88 caput Mode! summary aa Sa Ear oF Model |e __| esqusre | Sausre 2 7 Se] a7 756, 336 3, Preditors: (Constant), PRESS_X EXCEL op FIGURE 10.6 (Continueg) Applied Exercises 1 saving for ayand ‘The ution fo sight ie (de »= Bat Bur 4 ie fine passes through the point (0,1) then «= Y= musty the uation Tha = B+ BO) Sina, ibe tne passes tug be point (2,3 then 1 2.y = 3 must sais the oquatin: = + 8,2 Use these two equations to solve fr fy am Bd ind the gain of he ine that passes hough ihe po (0 1) sd23). 0 2 Fine foe eauators In eich case find the equation ase through the points. Graph each lceratyngyrercet and sope. Pot he folowing lines450 chapter 10 simple Linaar Regression | 104 OUAMS OF Quasi sttar Objects Astmocen call « | tetween aehift level and apparent magne Ge, | Urighines on logartune sel) of e Quast Scr Ob Jeet QSO) was discovered ant repred in the Jamel of Asimplpsies & Asronooy (Mae Bun 2003). Paget. Basu (Carson Univer, Oxiva) ape simple cat regression to daa colle fors sample of oes 64a) | ‘QSOs wih continnat reds. The analyse ye the flowing tests fora pete magni cigs: = 1813 +6215, where y magniide and reds | level Graph the least-squares ine. ste lope of the tine | Inept the estimate ofthe icrep in he wont the problem, © lnerpre the estimate of the slope in the wands of be = pmblem 105 Arsenic in so. In Denver, Coloao,envroeniins have dscvered a link btwoem high amen evels ina) j snd raga Kile ised in he 1950 and 198 Er i ‘nmenal Science & Technology. See. 1.2000) Th te se diconery was bused, in par 0 the seuss shown below The graphs plot the lve of the mes Scattergrams for ‘ Exercise 10.5, 106 Distance south of Ge Pann) ‘nium and uneni, respectively, agai the Stn froma forme eel pat e a Denver essential proper 4, Nomally th metal vet in si decreases tae From the source ez scr plan) incre: Pe os asnehtlne model ein tl evel wd tens fom te plan Bas onthe they, wld ‘expt pe ofthe lint be postive rng Examine the scatepot for eam, Dos the supper the hay. pa?” Evamine the sates for aseni, Does the plat at he hry, para (Vote: Ti ining ed ime Eso dicorer he ink been high sen ech ‘nd the we of te crabpaskilor) Vehicle congstion study: Mader warchooses empl omputres nd auomuzed pide vices fer nace handling. Consequently te physical out of the mae home must be cally designed 40 prevent vhise congestion and opsiize response time, Optinal design an automsed varsse win studied in The Jounal gineering for fdasy (Avg. 193). ‘The nyt loved aswmes that vehicle do ot Bock each er ‘en they travel within the warehouse tte a Ta 0 1008 ate anc south of Gabe Pat10.3 Estimating gard py: The Methed of Least Squares 454 checked by simulaing (on compute) watchouse ope ‘tions, In cach simulation, the number of vehicles was var Piel ~ iat and te congestion tine (ot Gm owe neice Taparent Mean ore baer Sled ahr wm reonl, he date sown be sce prety meron sccparyng ble Of net sheer hs i ike a vehicles $ * bd . c 163 23 wanenouse = > 6 33 Tube ol veides Cresson ne mies z mm 3 a = 4 0 : a i ‘ 108 Srsdng to of std st. A coma engine a 7 * DuPoot Corp. studied the rate at which a spilled votutile cH ‘ » tg wl ead cre x tan (Chel Elec 5 1 a fm evel stat ctinrs, Th engi wedded RE ‘empirical formulas (assuming 3 state of vurbulent-free eon- bi ° ‘etn Ww ae th mas pune he spl tera ¢ © ot Dated fie agi or 0 mi Te se # Caine vnean pon de seepeyig bc n os y a Y woyo0smu 13 a3 Time Mass Time mass a 1a on (rinites) pounds) | Grimes) pounds) iS 1s as 0 68s 2 16 ‘omer or optimal warchous out sin > 608 6 37 & Consinict a scattergram for the data. & SBhste frie, nee 8 Find ie etre ning ner fis Ee) 3s ow Go cingeion ine). wan | ow ox Putters ton empha - . a 4 Ince he values of Band Aaa [doo 8 tend te oan miu sete pot. A ies een a — | Sean of th propre of bk noe latina a ier pon wasps The Ameren Commi ple |e oo | Senet Bein fo. 2008), Sale commer a bricks wee ead Te ie gh ser pot > —— Penson proof tek Hin he es oon Sa een a eran | 1 ge hi) soe ew ms xen i eth appre poosty of exch rick speemen 5 Wl - ree ies Step dam seu Ok Te ae CONAN oe a hy = Seine i 0. Fin eg ie ling at) te he instepiatns Ties ele poanit {ni te ee ¢. Plot the least-squares Tine on the grap, part amr Inc vars Bad 2 pte ee eerie le Pe gs © laure he toe of ie 109. New metad oF esemetng rani. Accurate mess 4 Prati the apraent prouty percentage fr brick ‘ih mean poe ameter of 10 mice mens of rainfall ue eval for many. hySeloical st meterlepical projets Two Mandar mets of4 montring minal we ran gauges and weather ra, Both however. can he comminated by hate sn co ‘omental interference. Inthe Journal of Data Scone Apt. 2004. researchers employed afi neve orks (ie, computer based matheniical mal) ex at rainfall at & meteorological sen tn Mowe, Rainfall esimstes were mage every § mimics ver ‘TO-minae perc by each ofthe thse methods. The Sats Gnmilinetes) ae se inte be 8 Propose a simighine model relting i guage 1 Fithe mel w de dt the RAINFALL fe sing the mba of eas equres (Grape lesa’ ie nate ed, [there vial evidence of relaontp between te ‘80 vaibles Is theolwionshp pov x meine? 4 ltl he esis th nape pin the atl neural netwerk rain estimate (3. Rept ts fortis el PAINALL “Tine Rader Ran Gauge Nawal Neto Boas 36 o 1s ao 20 12 Ls so 12 “4 sists B 19 30 1a la iw 830 a2 20 21 so as a5 26 B09 a9 40 40 9mm 65: 62 «2 10 ba 78 12 sweetness of rage ce. The ua of he range jie Prodiced by 2 manufacturer (eg Minit: Mai Toph ‘ia icon moet Ther ar waren seep snd chemical component tht combine to make the hose. ‘anting orange jue. Foe example, one manufacturer has developed a quate inden ofthe “wets of oe ange ie (be higher the inde, th sete the jie) Is there 4 reatonshp between the sweetness ine an ¢ y ton ‘chemical measure such the amount of waters ec (prs per miionn the rang jue” Duta clad 8 these two vars for 24 proton runs a3 le ‘menuscuring pla ate shown the mex abe Sapee 3 romuicterer want o oe simple linet regres rit the sree) fom the amon of psi 2 Pn the astute for the data. nee al nthe words ofthe probln, Pri the sweetness iodo ifthe unt of prt the ane jie 300 pp ovuce fun Sweeiess der Pec pow 1 32 =m 3 60 2 5 ss 2 * sa 21 ° 56 0 0 Th 4 su io 8 38 06 4 55 = s 33 a 0 87 at 8 55 264 » 57 Eo x 3 26 a so m 2 ss 26 2 so 241 Nov The iin te abe ar whe, Focoaenliy Reketite exerinene Two proceses for byt ailing tock dyeing awe iing. 13.27 hole, somes sis rsd dawn he del es 088 ‘ects nd dive te hammering we hoe fel dows An experinent was conducted dae ster te ie takes wy sl waste of St inock incre wih dep (The Ameren Suc Feb. 199) The mao fr oe pron ote expo archon ia the nent a’GY omuncce “Deper ae wen bing Te wrt ‘ag fot, Footy, mites o 490 2 744 0 ou 10 si | 200 os 250 44 300 a9 us 836 380 1179) 285 02 2 Construct a scatergram for the dt 2. Find he least sqtes preston equ. Grp the essa ine onthe stream Inert he values of 8 nd 8 1012 Therma perkarmance of copper tubes. A sly was on ted wo mode he thermal performance of inegral-in ube sed ip he refigeraon am poess ind \Uoural of Hear Transfer. Ag. 190). Twenty ou Silly manuf nepal fn tbes Raving restangla hapa Fins made of copper wece used inthe expesiment Viper war reese downvan into each tbe and the "porside het taser cece tse onthe Ouse sutice ares ofthe tube) was measured, The dependent ‘able forthe sty i the eat unser enkansement ‘tio dein ether the vapor sie colicin of fe fin ibe wo the vapors ceticet of sma abe ‘ald atthe ane emperors. Meee et fe wll be reed to the aes af te fp of he ube hats tafoeded” by condesation of he vapor The dat in he Uhl are the uno re ai (ade sans en aacemert ) vals esd for be 24 ata ia es. Pind the Irs ques lise celating beat transfer en ance yo ufo are ae 5, Plotted pont and graph the least-squares ine 8 eck on your calculations Inept he values ffl and ~~ 103 Estimating p and py: The Method of Least Squares | 453 & oerupes ‘eaded ‘Area Ratio x 195 138 4s ist a 22 6 Lm 49 ss ‘1 aa 70 237 “7 200 sa 12 42 a7 38 is 29 33 51 Sironth of masony irs. Civil engines fen se the suaphline equation E(y) = By + By o mode te re [uohip betwee he icon sea seg Bt) of masonry joins and presmpreson srs 4. To te this theory. A series of sree tess wns performed on sold beh ferangsd in tiple sed jined wi oa Proceedings of the Isinae of Cl Enpncers, Mat 1980). The reson frettion rest was varied foreach pet wd the nat Shear load just before failure (clled the shar street) was resordd. The sess res or rps measure in imme) ar show inthe ext ble 1 Plt the seven datapoints in a xatergrm. Does the relaomiptevech shear stensth and precomrss lon ses ppt be linea Use the method of eas uses o timate the para seer of the linear mi 6 tery the values of andi, evnartin, eeriemeters Hoop pine: j = Ce ae mere esimate lope ofeach in ieee a . i . Inept the esimeted -nesep of ech ie Sew Srengih y 1.00 218 224 241/259) 290 305 soon tee resus. which ype of ibe Me Precoprecion prea to be stomget and cre fatigue ian? | Sine 60 120 138 143 as us Exphin \ Theoretical Exercises 10.18 Wind tree bade sires. Mechanic engine othe 105 The maximum ikeiood estimator ofthe mean yo Univesiy of Newest (Awaba) investigated cot tormal diibtion i he sample mean. Comer inte in higheticeny small wind bine Bak (Wiad model ty) = Show tha the an sures sina | Exginering, Jan 203). The sgt of vo YPC of ivalo 7. ter Mini SSE Sin reame limbet—radat pine and hoop Pim wer compared “wea spinens (alld “won each tinker ak ae fig toed by messing the tes in MPa ne 10% Comite air of siutancous near equine ste various mers of ae ces Asim le y+ BZu= Dy als eh tsa ice fre ing hs ow) = ne BX De = Deo, timo mame yen Dative the formulas forthe leastsuaes estima. Radiata pine: = 97.37 ~ 250 ond, 10.4 Properties of the Least-Squares Estimators An examination of he fummlas linea functions ofthe nae ton 102) thatthe random enue asta aah A i dependen. somal sett random vaables wih anda vas Ht} follows that they valves willbe normally disuibuted with mean E(s) © Ba + Bo i and variance od tat ad wil pours as esa | ‘ally astibted Chee 610}, forthe leost-quares estimators reveals that thy 8 Yalu. 032. je Since we have assured (SH 5 ¥ Vale, 5, -- A Hema andthe ain of the sampling dtibuon of are gve in Sen 106 We ‘re doquited in Example 102 example 10.2 Fi he mea ond tare he sang stone i solution Hh | i { The quantity $8 that appears in the are assumed 1 be knowe sant when we Find the expe ‘dom variables, 3,9, fea ces aly the val, ..norsom, Therein 85, abe eed 262 ‘l vlte of 81. conta, SS, i a funetion of thea Ye Ts, $8 = 5 ~NO- 7) = Diy - 3Liy-H Ab) ~ (& — Dy) Bes-N= Beate Sn y,104 Properties ofthe Least Squares Estimators 45S Therefore, 8, Substring this quay int the Formula for, w S8u This shows tha is. nee Function of the normally dstibuted random variable, ds ofthe random variables in the linear es ST te ae The final sep in finding the mean Ei) and the variance M(,) of the sampling distribution off it apply Theorem 6.8 which gies the rue for finding the mean and the variance of neu function of random variables, Ths, vay = (Dy , (ed (9 MED A ase OS gey Seo whore yy. te obsined by substtating the appropriate Values of + ito the formula forthe linear model, ie SESE 1 eleanor Qs) = Bo + Ba | = Bot Bite tee ant £0) = Bo + Bite Tere, jy = St Pagyy + ODp , =a, Ree sae Sg = Ges 2-3) $e (Bo + Bin) + 5B + Bim) &-9) (eo + Bi Bo 5 8 B3a-9+ 2 50—m bt Seq = Dl B= Tlie - 10-9) 23G— 5-3 De ow) Since we hve already shown tha Xx) ~ 3) = O, weave SSq,= Sx) ~ ¥)xjand | neotor, Be B BiB) =0 + gets This shows that i is an unbiased estimator of By.456 chapter 10 simple Liner Regression Applying the formula given in Theorem 68 for finding the variance of a lise favetion of random variables, and remembering tha the covariance between ay pot fy values will equal 0 because all pairs of y values are assumed tobe indepen, wwehave - an ~ ¥ bs fi) = 2B yy + ' | (SS) (SS,, Accoring Jo the msumptions mide in Seton 102 V(a3) = Vs VO) = 0% Therefore Re will use the results of Example 10.2 in Section 10.6 to test hypotheses about and {e-consruct a confidence interval for he slope ofa eression le, Te poste implication ofthese inferences wil also be xp Theoretical Exercises (win: 8 A= be 5s, 1018 We shoved in Example 102 that i, the leasesqeet SP ie. £(B;) = ). Use the result from Exercise 1017 1 stove =p ae 101 ta nei 107 yw at ye we a sims ann i i be SS, Mi) - © (2 10.5 An Estimator of o? In most practical and must be estimated from thes sets atc in Ely) = Ba + ft deers mullnely asec ocateate by dividing SSE by an appropriate ruler. Theorem 10.1, an extension of Theor 6.1, willbe useful in obtaining an unbiased ene,Lexampie 10.3 | Solution 105 Ankstimator of 2 487 [THEOREM 10.1 | Lets = S8E/(n ~ 2).tThen, when the asumption of Sction 10.2 ae sats he _ SSE possesses. chi-square dtibation with» = (n From Theor 10.1, follows that and we conchide that isan unbiased estimator of ‘The procedure use in Table 10.3 to calculate SSE can lead vo lage rounding rors The formula for? and an apropeate method fer calculating SSE are shown inthe box teow, We wil ilstrat the calculation of ¢ with Example 10.3 Estimation of «2 Ss SSE © = Degrees of feetom forearm ~ 2 S86 = Slo,~ fl? = 88 — Ais, (Zui 88, = Doi-5)'= Boi - Wining: When performing these calculations, you may be tempted to round the calelated values of SS,,., and S8,,. Be cersin to cary at last si significant Figures foe each of these quinties tO avoid substantial erors in the calculation of SSE, 1 Inthe insulation compression example, we previously calculated SSE = 1.10 forthe | teastsquares fine § =~. + 7x. Recalling that there were n ~ 5 data points, we haven — 2 = $~2 = 3d forestimating ¢? Thus, SsE__ 110 = Hee 67 n-2 3 ¢ Fi BNI KASSfs the estimated vaiane, and | 5 = V6 = 606 estimated standard deviation of e, Bath these alues are high MINITAB prio, Figure 10 ed on te FIGURE 10.7 spe inarersson of dts in Table 10 “ou may be able to obiain an inttve feetng Fors by recalling the inert cas att deviation in Chapter 2 and remembering that the ase ayers le sautmag te mean valveofy forgiven vale a Since stmcasuncs Ge spt of he leaseguares ine, we should not be surprised 0 id Te Pee uat th observations i within 2s. ot 2(608) = 121, of the ease aqumes line Fortis simple example ony Five tan sein eins al wins SE estas line. a Secon 10, ve wil ws gm avaknaste creo Aistibutionofy values about the Interpretation of s, the Estimated Standard Deviation of « ¥ values t ie within 2s of their respective It Applied Exercises 1020 mepring te stad dovon Cake SSE. 2, 1021 New fone pce An tno ev ie an forthe estas ine taking etl (led Tk) panes gh 2 Eee 106 Exerc 107 im misses det frre re an sel Mi iI © Bene (08d Ekewne 109 Ensincering (Ot 2008 pulse he es of pt Exercise 10.10 Exe 10.11 undue onthe new process: Force pas of the a the carbon omen ot plant test. a8 serch 1012, Exerc 113 scan see pred pi pat et Eee ‘mpaodto tha fom laberatay fumace tents. Te Ierpre the vale of seach ing oc 25 pilot ex ar Iie the ie _,b. Fit a simple linear model rclting carbon conten in a Piet test. to he caro coleman ab fern Tnerpesie estimates ofthe model parameters, . Compute SSE and 1. Compa a inert its vale caroon FiatPant—LabFurnace | Plotrane Labrumace ls aa a 3 a 3 M4 M 30 30 32 Fa 32 26 34 1022 brug controleelase rat stu, Researchers at Dow ‘Chemical Co. investigated the effet of et surface aes and volume onthe rate at whch ads isles in a tlledeewe donage, (Drag Developed Indra & voworue nig Release Rate “Surface area to volume o 150) 8 Los ® 0 » @ 105 Ankstimator of? 459 Pharmacy, Vol 28, 2002) Sic simily shoped abies ere pepe wth iferem weighs a thickness and he rio of srface ara to velume was measured fo each Using» dislton apart, each abet wes ploced in 900 mls of deme wiser andthe itasoal J release to (percentage of drag flesed divide by he Square rot of time) determined. The experiment d we Ise inthe secompanying tbe Fiabe simple inci mod, E(») = y+, where ~ drug release ra an x= srfice-areatovalame Compe SSE, and Interpret he value 10.23 Moreuy posanhg hh kes I esense tox ea aisory regaring ney ptening in Maine lakes the Envir gi mental Preston Agen sind eds of 12 ey 1 Aco nd dary, Ata Sale Aa ¥. thm, 1998) I ao to the meter level (ars per ae tn ofeach ake, the EPA ese the ake kevton pal (fe Te dts are sad inthe MAINELAKE le (Dats ‘a forte int Oakes ze shawn inthe ascompanying ble) i & mamerane ‘Allen Po 1089 405 Alligator Pod 002s 49 Arasgumicok Lake 0570, Bastahepan Lake 0790 Beaver Poa i Belden Por 330 ‘Be Annis Pond 0180 m2. 2 Fite simple linear model, Ey) ~ fp + Be, where meres lve and = elevation €. emretthe value os. Theoretical Exercises 1028 Show that Ms?) = 204/(n ~ 2). [Mme The seat fo lows rom Thcoer 0.1 andthe Fat that (x2) = 2 1025 Verify that SSE = By, ~ §)* = 8,