STA 2312 t REGRESSION MODELLING t Curse objectives: ot “fh the end optbe couse helete due beetle 4s tnvetigets hear ve tons ps ween wvadelles using covrelation anelysic and ste gression acalgts: Course Cutline - Bate concept of regreccian > stomple linear vegrecsion irodels, maltiple \rnear reqrescion models , vafious designs for reqreiston, | err: accumplione it eo ae € and and| ysis hehermacedsanie ty, inperence about sed eliparametees and predictions hagontstiec af rewedtal mee'suree apd a ell independent varctle selector, aothteellertarly en ult cdldneor ity preroge’ site Lysine ads os ; Sth RROL - pobalilily and stokisMiec 1H ‘gtx 2300 = theory of ectimation Rererences- 4 ; TN’ R -Draper , H> Smith bpp regression avalyciey Neayork- wiley 1998 2. ¢- charkerjoe , 8: Paice : Regression Relic, Negek ae 2 DA. Freedindlt titted models: Frees vr ott > (Theory and practice : Cambeidee Uriverey Press 200sthes aI in any chavacleidie or e-guantily, or a number = quankicetive «Searle Noinal: theb-gou can measew or count ~varialle candor Introduction 1+ Represion ands te Aatitica rede the ltcas te tlio p felweon dice er more quantitative variables: «The method ogy willy eal | in bicinece, social and behovioural ceience, im ta tens ond meng eller dt ephne: pai, Analy fe aed or expleining of modelling the rolalionchip ee ee ee he, Pepe etl teal Wwarlable- and’ one or more pecdores hel tepaele or Dinte nf variables! Kr,Xo,-e+y Xp * 7 «When pet, His walled simple ‘near wegen» bet then Prt, it i ha oul 2. linear mnoes een 4 single Purpact oe, Bagrindon Peal sis - " , Regression frdysic tes cover! Objectives tacleding: ela cn 9 tre cleo gq» Ascecsment oy eff eck: of , or ehbonchip Pesce vil variables and he responce ails. 3 siGeonl Aan op deb. es oy variates and the Hee ep egyeccion ned besa ve aktenion , fc ene response variable only + iy dada waiiable CRecp orse) iideleaa yard any vale aillan a given ‘wteral - 095 if ~e0Ky Loo * normal lineay regreccon if OA Ykt ts enigonm Bele vogrescion Gio Discrete vattable (Response) = taker en cidinch luce nly. - e953 ig counky ye Otj2y-6 % frisson ditg = = poisson region. icin binomial dab logs rerescton ig oun; yr0,1,2,° ip ents 201 Foal ded Leite yen ae =General Linear medel- enon term Fe ascumeite be omecced gly vores) =o" Oprealtced Wier medel- rer artecn ors oO assumed + be heteroscedaticity satel ese, pdesondlant Un tate givet section of this couse, ae uill concentrate on. continuous dats which ic nocmally dicbibuted Caencral Linear Model). The cthiv lye of sible ail be Eckl ender Gueand Liner Models Simple Lineay Regression : + The is a becie egiéerion model aith one prctclor| explanetery variable aid the regrscion punction ‘es linear in the parameters. ‘ The modél is stated ac : : - Yer Pot pues te where yi ie the valee of the vesponse vorisble at the 7% Bal: “The gis ave independent + eqh wether fe a lethe intercept 5 Fegreedion patametor- ubich te the vealye oy y when Xo 1 zisthe dope + reqrecsion cé-epzecient regression pavam eter Pe fen ed dng ky cage yf The nature of change all be Weestned | e ‘ign of Br 24 2 ig q krown constant’, valee of the precitor vavabl ov explanatory variable inthe ih tat: ’ e; 2 ithe enor term which ic aesumed 4s bé normaly dishibuter NC0,0°), this aso assemedte be encore lated co¥ C8} ,¢3) =o 10) chore the ard j are not equal | oth cs - Chand 23 ae independent 9 Pte ne Hemoscedacticity = enone have the same variance jg; 3 ~N(0,0°) « Hetero sedactieity — each emyterm lave chgerent variance 125 Ew NCo,ot) te 1 5R,--7yn Var (@1) = 02 . Var Cea) =Propertis of the enerterm vlbte sormally ebb In. other werde, it has ¢ meqn ep 0 and vadance OF cod eumean 20 3+ Variance = 0% wp Ureonelated NCO) =O Lay 3 ef and XP ae wld The wll 4 Inetoie Ua péronotes bene 90 pusmeter-appeoie 4s an exponent: er velled [del by ancther parameter Important seakwes of the medel- Yi > Part fate fe: Qi ~NCoD*) var (to nant )= 0 ort 7 varty)=0 EY) = EC Bot Bi xi +02) ECconstant) = conctant > ECBe)t ECB, %i) + E (e:) Elei)e0 = bot Pr xt Var(Y¥;) = Var CBe + Pr xy ey) = Var Cfo) + Var Cpr) + Var(et) so po Meg YE~N Chet xi, wt) +Extimation of parameteis inthe model “There are 20 methods ates pormete ia Hite medel: 1. Ordinag Leact Squares Methed Cous> a. Maximam Like bee Methed ¢mue> + Ordinary Least Squares Methed Cots) «Fer the obsertions. CKE YE) or cach case, the methed of (eadl-spares censidesthe detaalinn of Yi fom its expected value t-e, Ct ENG ~ Bo ~ByAd Yi - Por Bre) + Un OLS, we minimize the sun of oquaved envrs Ni spt prites CNG - Bo- PiXi ey = Vi-p-prxz) * " Zee 22 Ni fo- bai)? iss Fe let @ > 3 er? an = CN} -Pe- BX)? ae atdipperenticrting by chain nde wrt Bo | BO og ZOYi-Be- Ba) En) a I atts tia, equate the dizrerential to 0 a +22 0 fe pide =0 a dd both cides by ~2 xt? ape Prt d > o ae =2 ; . 2 (Ya -fe peti de uy n 1 4 ~Z fe ~pr Zt zo ier a ay ZY: -4fo “fr Batre ba | 5 / Mette tee SNe 22D | fe is © tedigeealinggurt fr Bo. 2 2 Mi -fe-pr)'CxE) oR is $2625 Sal (Yi “Be Bi)! CH) se 2B, at . = =pn = 22 CHEN; - Boi -pyXi*) * 0 ae = 2 CGNs Port Fis )= 0 ie 5 % ayi)- be 3 xi ~ fz Xi 20 ai ot pe 2 xiep 2m Ea 1 @ ier 4 edving eqn one © simultaneously’ A vixt pr? 1 Ba 2 ZANE xn 1 ‘ obo +B2 4 2 22 Demi -_ izt act i nfo 2 z etep Eat 29d aM it npe 2 4 + Pr dai 3 xi n fe a 4 npr xi p(Z ma)? + ae ag By os et RiLad xt- ve «4 -2"i 2 Ps atSExy = ROH); 4) = EX — EXD /y Sm = & (Cee Bae = xn » Frem eqn © n . nBe +Py 2x52 eye A ain ape = 2 YG -B, a te 3 > mh ptt Assignment: Given Mi=Bo +Prxt + es - show that fe z0a- 2-9) ba Poe 2M -f. 2x a | Example " Xand Y values are @ scores on a pte preliming qiral tect of dudente in ddatabee dee pies X | 1¢ 12 t se 54 42 6% 80 GI 4 [714+ 33 20 70 16 33 40 Yo GF Use the above date te draw ¢ setter plot 22% Gin Te gi a rrescion equation ch) To prt” Y then = 05Goo we wot x J = so woe De 30— i) 4 : : fet n 2ave - exit yp K nN 2x - (xj)? F T == RY: Ike 70 at 5390" \aqoo Tt at. ‘ease | sfyp, mw 33 336 5) 3H cs 30 Btto | 4eay a) u ea BB oy ii 16 tro” | eae # 94 Tae e1Ry oe vo Bi20 40496” ae Jo |1R0e | coo ct eq | e20g | ara 2xi=693| Zhe M3] Doyj=scpsy 2%3*24TH05 |Pr = Wo) se, 087) -(¥3) (2/3) (io) (#7 #08) -(e93)* SGORqo - 5555279 = Sen = OTH RY] AT HOS ~ HCEHDF 156] 413- (0: 1481) (643) 212-506-856 to to » Yu Boe + o-7HONx » (i) Y= 30-81 + 0-742 Ces) * Tate oe nme Given Yee poePe wtpeey «Show Led? = 306+ Pe BGR ACEI) fee 2b 2 0 | n a" Kw min 2 Cer)? « min 2 CY Y; JA "I Ys “fe fx lia 2 Cie p “Be x3)? te 1D = 23 (Ys -Bo -& x3 MC) = ap a3 Oi EP prt IO1)=0 v egn® =QD sad (Yi -fe -Pt )C-Ai) =o 3p az ( f pat )O%i) 5 eaten a RE) oa oa ‘opiag o From equation © n = te 2p “ aYi-2 Bs BV opex oe ancs) aAgiitng eqn 2 in 9 we have; 2G-CY- pl proses is 3 iat - 04 “pik 1a ~ puis =o i x F Wo Ba 2948 43 pozastd pat Blea vt el wt an 2=1 6 ZO) A - 2B ERD ogee mr20% Yat ~ pd (xt 8) 26 oy Pr NM by YGd =) = 2 Yexe- iyfe “Bir he = Demi -2 x i a 3 iF) ik) = Bi 24K) ORE 2) CY TQ> res Pte i + For he estimation of 0°, recall that prtsample Junge, -eoJq ist = 2 Cyi-§)* n-! ~ We are dividing by n-t because we have be Plog “tre in lasing G te estimate fl. “fn @ similar manner, o* = 2 Yi - é tf ag))8 n-z -We a1 acing n-8 because a eqteed oe are lectin edimating fe ard pole Mes foeBex: heme S.-2 9-9 = paula ion oF 7-2 z ef = mean stirred error CMSE) Se = PeddualSEB NS ERR Exercise Uding the previews exaniqle, ext mate af te Cede) cE KE fe Der OTT HRI ao (i - 3) u 10 §255] ¥ B56 a5: 3: |. 20°) tH ve 4G SREF T4553 aT weit ao BR aH oF1g TTHOF! 3G tees * o srotaa = BLT | 1-27 nee u 8H 73-514 & 1365 | -R-GE wa te BE yoo 3pc toes | 538 pao lez a qyteaae age | 7347 ay ae oF Talo TES TH a ges fe a0 84-913 24042 | 0-02 o-0c0% at” er qe-318t = 15-89 | - 6-39 467-33 Viesia | 325 693] Rg G ores] 3 (ge-G)"s ary ce 5 ce neto Ga) -begawse Hora Residuale oe The 2 raiduat daisted @ Vigeeeike vated gi andthe covresyn QT QT RIS belueed an Srsi-bein anc sqresicval «7 piramelers, Board Cir isthe diggerence Letivein, de domed n ing fbbed value “gen: by age 4 Gi =gi- ge walt vPropertiec op the fitted ine ( Z2e@ico i “since the ketal devietion of the predicted vdues qrom the adual values muse the came above ane below the linesthen-the sum af the redidvals uill aluays be o- 2 Deviation othe Lead: Squares Ecti mabore gearantecs ester the sum of the cquared reciduale te Tris tc the minimum valee sprang auch model: 2 ei* By athe sum of the preticted values ‘ic the same acthe sam ot the ebce red values. Ron oan 245224 u ta * The reqresion Une possesthrnugr the pont iY) 13 esx 5 lZeitizo fay Zearizo er 2 Maximum Listed Esk mation ~ adh . + When-the -runctional orm ofthe probability distr ution cy the erretterm is g veiedhe adn dorleg pe mt can, be' obtained by the method of maximum Velhoe: «The tikethncod guncione of m observations Jrsye,r--ayq ie the rduct of indtvideal dencies Beall TE) ; An this case, sine the ermre are normally dictrbeted (65 ei ~NCO,07) | hence each ys te rormally chbibebed Ce, Ji ~ CBot Pr%i , 0%) es!tye “ Cyi-m)? t= Ley fan wy ax ao BA yi WC Bet prs ,o? eget -L CS = Be- pias )® a via “Hf aot Bat f 4 LC’, Pe poe i fog) Teeter i 30° i (oa0')® exp {de 2 ipo“ prxa)? G spt! | ji + 7 = The walter of BoB, and 0% tnt maximize the fkethood function oy" are the maximum Bhelihbed estimator andl grodencted ly Poh andl B° + 8 = We chose fo eerimize the Log by tikeibsed suaction arc 1 -Mhlnan-"alno® whos Dtyi pop f | : = fat 2 Ad 2 (Ji porpeiy® zo jas" RQo% gat BtBee aot BOF pepe per a ete, te 3 cys - Po prety Ne iy we = 43 (Yi e- pra® et . -2 2 Wefan ch 2 fo 208 al oe st 2 ts oP xi) = 429 pe-pzi) 0 eG fat 278 z (yi-pa ~ psi) =i) apr ee SN eg 3 ice Pepi xidePo i ® oe ay 4+ sdving on © aid © Sirvalteneorsly ee have: Pel = BZ xe a3 pees ot oF a z a Bags ~ beZa “BEM | }s cin a a ——Jes 2$ -npe - pr =a faba - Buaigt “BEM AZ, SNe Yen é al paniseds tebe be he pa , 7 A | peer rpr xia: & Z miss baa q 1 ofA 2a 7 243 zai eae “ngs be oh Kyte = ab ep Me abe FE ni 23 oaPngi By be ie h Lad nsey RxD 0 2 xiys és th e mH Beee Assignment: © Uce the method of moments 4 ostimete Be and Br © PDipperenticte bebueon: do Maximam Uiteliheod: meted and Ordinary Least Square Method, ain Resideal and error v @ Slate the propertiee ascumpions othe ene tem. © Sele T assumptions Regression line. ¥ ©)Ginen the date 10 1 12 G8 53 SHIR CH 40 Y | ua 43 40 1 1 33 G0 qo 49 ci) Witte @ code to read inthe deta Gir Write the Reed fo zit the regression line. Yee Ba tp id te aii) summary) then diccass the output ve Diggerericke betwee heroscedasicity and, heterosced abi city» 1 Ghee the causes of hetero seedacticthy.and discass * a ted jor ‘dentyng heterescedasicity. a Exercise: 1 When acked te late the ample Kirear regres sion med el , a chudent wrote jas qellous: ' he e ELT Bt Bete? — De you agree with thie 1 No, et it ie cuppeced lobe Nis Por Bites 2} Consider the normal ene ciple Linear omedel, seppece ectiaee! values of 280, PS ote iy Exam minh | Poort By inthe are Wn Find the expecte valve ef N uhen 2X2 10 See ftp ae Gis R00 ts ry © Be jc the intercep oy and ‘ic the value 4 Y when x0 > Elseo0 SS | =p. = 52 -value oY vhen X inereasec byt unit: ~ it ic the ope of y V2 oot eee xi or gradient of the Vine iy ECD = fa Be ECYi) = @ 00+ s2ax X10 ECYi)= 2004 5-2Ct0) + 200 +58 = 252 (ngerence in Regression aHere, we make inserence concerns the regression parameters feaed 1, considering interval ecimattion oy thece perametere ane] hupsthesic tect concerrfing them . ‘J : a cecton we assume “he nevinal error ciple linear rogrestion reel (Mis potpai tes, ia NCo,0°) ieliminary revvlks: r/Br icalinear combination eg Yi 0 p Pe ez¢ ay) 7 2 Cai-%* { j & / s - 2 2 (m-Dyi - 2 Ci -50g 2 aD mH 5 an)= § (nk 02) +0 td So ee ‘ 7 Ze byi | Paces & aS 5 Z (xe rit t... 2 (2-5)* | le wee yg i Boat. I = fe Zkigi hence By tc aie cero ai shad if en Ny : | a4 este tha he AERA ly ropertioc of Ki r r 2 Hl t Zk =o ea te 7 To i 3 (6-2? [ — is 2 (xi BD? -— 5 2 aD 2 pf-at 2 a : 1 20 = hence reluas 0 4‘lea ft 5D La L TL HD oct fa pa (ai 5® a | b 2 (17 )* oe [Denominator we tne B® 2 CM -O)CAGR) = (ai - CDK ol Fee wi redues b 2D | 2 Gee ie 1 4 (ZT G@ea)K=0 . js » ‘ ' * 1 | | 1 F | - 5. @-H)? Baal :Q: Show that B, ts a anplased esliamler + Fe ( Boe: ' =! ' [? ey |% F038) Sampling Dishabaen of, E> be [Fey il Eq > e[2«si] te, = 4 Ki ECyi) ie 2 3 K Cpotpixi) oa ‘ 2k il s fo Zki + pe 2 AL >okings| f ot i { 2 peed + pico ll mh hence 8 ‘is anbiaseed edimator of h \) Vor Cf.) = var( 3 ky) Me gan Kl vareq) > @® var cx) fi = 2 varcyiy ; !Tex? a Since oF ts nek known, icon be esbimabed by mean quand ener (ose) the regore var (B)= MSE 2oui-i)* a ie x 4 Stimses get = ZCu-Y)* n-2 ak as Sampling dehekin 9 9. poste Ef): ): ELG -p p=] LE Gor ia Bh: | PEE eRe op nego - ea) at + atte = a 1%) BLE +f day-ega ; . } au " ALM pe pga - px ; fe Pre - pe lp + fe is af aliieel Main ikVar Cfo) = vor J ~RH = varcg) ¢ X? var cB) 3 i Ke | nee Z(u-D) & From the date, hee can be estimated 4 Nise mst [1 4 2% awe " 36)-»* ie Assignment Find the var of fo end Bc using the previous exam ple lot. ogerence eoncemins f . + One form of by pthesic tecting ic wally of interest to mary researchers Yegarding Br is given by He Biz ty Bete ; ; The reason te interest tn tecking whether er nek pieo jethat when 120 when pre, there ie fe linear acceciction beh cen y and. fi tc becavee ig pre0 then ECY) =p Semplin dictributton of 8, fiom previous section ; ye showed thal Eh) =p and varcpy = of BCq-%)* By De —hen oie nek Kron, then wianse off mee yest Re mst Zea ay = “The ae tr keting pr ts guenky Rp: cave 5 [Smee fits emelly dsb oe browser fer g ; dandard roma variable Kerth | «| Since ©? ok boar; hence vetance: of By ic teat Frown: Se ue estimate var cB) eg st cp d= MSE 5 (1i- Zea Hever, the tre-dbtictce becomee ots beep Mes oe ba seine ies th | SE CR) ath ona agree apeeetio = FB Bund Ves : Fe 2 eee ; \ | | Pig the, the (1-4) nin intewl por pr ie 5 or de teak tige piso Ve thes aie Therteikstatisits 4g given | by teh ihe 6 ert | seehi vin bauer ns : a ny | tapi ‘on Be "he shrpling Eke, feat “7 fot hes itn (lous The te preedvia ab grm “f We enly? a to replece if. se Cho? He gere ot ati se Cf r megetSC Ba) mse [4 > QCA j-R* tte Pe =e tof toe Pty. cego lll Wt Boge secps) fe > Fei ef29 cneay SEF Example! 5; Business stadents wart te prafict the average cuslemers Hp dodhe grom avg customers bill: They have calleded elate qrom s cudemors ane! recully ave'as fellowe- ove : ' Customer | ill Tp : Ra 2 pS By 2 ase arty 2 efGo a8 + Bo: cs Sonate ete FP etopiag Fon gia Predtcts the average cuckomers Hip om the average watercin tal pendent vaiiable Tip ~ >: . ts Inclependont- variable - bill 3) Daw a. plet fer the deta Find the sinple.linear: veoresiton egyation anc rcewplefe thaceordizaly. cin Padiet th then \illoeo T Sar oF (iy Fler the pecicon etmatle ecrrelation coegedent hotucen Giana t2p one! takerpret- ie accortingly 3 ojPae the coegerient og determination and interpret it accerting ty. i Tek > atte 2% the pllesing dypethed arene : poe @ HotPé = 09 15 de He {Prt Oo. Mss Se ste We . th Bete Hy prt MEL ay 1424e-109gsty Constet 454 esngiclence interes yor feondf, 0 bi Let HpeY and lle wie Ne Be 2 pi Lut lose quae 1 439 WOH 9164-3458 2193S ai a3 te {26-142 VUtTsate Tf IRF att 83-00 623-46 OK TE Fe.zaer 3-41 BGR $5] -S 598 Q56a Hoye (175344) 117-33 oT 3 1-8 Go-RORQ tr5gt- 3156 | 200 +3264 zee st-28 BKs 5534] SkZe= $500-93IC| — AXi*= RA02F 2182 |ZRL TR ni 5 (st-34)C 3 53-99) easy ¢€35392)% fe = BC 4s00-7381¢) EC ATOR ASR) ~ jz o-1G40 5at0T ' pi Bhs s eps -paencosa tt) ne & BRSR FIT _ treet tts: 1491 -T07e A i ~0°139Ga ET EE Ye -0-13847 Ho 1ep x, (iy X=Go J 02-7396 + 0-164 (60) x 7-108 @ ws nexy - Sx2y Ore gees Aly Za8 = Cod TRE zy?) * SC4S00+ 4216) - (54:33) (303-92) [sC2t02 ¢ 2168) - (358.42)7] « [5 (102-1596) - (s#39) 87 R2G0F- 653-1924. 0q¢ = 325 44a 14 3ctroqe SSS Cap BUS TTAB = OT 4REE ww), co-egzediont- of deteratinalion RSR* = (©-92286)*. oageseooeis wd HeBr =o ve « Hegrte 4 seth) Saj-&P n & var Cfid= Me mse = 2Cy¥;-4)4 ara5 = 4 |X Fe 0 rueet0-1e¢) G9)" te Ry* Teh | ese tetTTNE oO B48 V4 EE 181-1391 3¢ 93-te |e pent 0 187409 82S 1298-030976 oro gs 885% , 23-60 |4-4 qcqa3 4 SI goqase _ a4 B= SERTR VeTIN3t¢e treats EO Ee veg: MOSSHITS pect ot 8f | 17-82 lest eeeetet 1336-34113% qqrea) Dest at BOe Ds epeperesy | 3472-35198 = BAP OSHA Ht RETO - | 4C4i-9)*% 23-2054 5 1 eescesos | ae 8 i ee VarCBl = 1:96 geese 1-96 96.00% = 9-00031937 i] | BAA 2472-34192 - A | segos eotratiey = are) a f = Oe = Petty + sech oor TsTIot att ‘ lt @ WH) Ter Faytnee) | soos | _ He EL eco, se \ | | 2 | gr ne = teats,s SB18R cision | Teal tah reject tle | bse atnab - apt jee the qta7 B® -pejet Hehenee reject te . “ : 4 tp = Ve valye Conclusion: Since we have rejected the pall hypothesis, tchows thet f, ts diggerent trom 20M - je Wis shy shaticlica ly sgntiteant @ He fore Vs Hy spore a te pe Es) 2 x fe verces) ae. stcfe) = CR -K) Vav(ge) = (8 87966209 le + Cr0-Ta7)* . 3172-34692 = 695 311 e485 ~ (4539 seeped Travefy = SY (36753609 | =0°53 904400 te -0'T3R¢ tryeee SF —ertsttea - 0-53 204900 | Decision fees : Fey 998) ¢5-43 at . sseoper teal < tanh : |-o- a seoppe co132 . | therpere we fail te eject te o-qis,s > 3148 SEG) A get tte abcoktSF eg conclusion Since we sailed J to reject the nall bypthese. i dee ht fe equal * nero ie) 1s statistically inagnipfant emiy @ congidence ‘interval. gor fi Rte $5t F teary eneay” cecp) - ortet thy. ag esa” OT i o- leet [Braz Ko-01T a5] tT orfeq t OOS S198T To tep 005617, o-let t 00519] LT ortotgt, 0 @2o1dd fe: , 8 x = Hy! Beto < “ ‘ esa a Decision Reet the mill hypetheste cor it dacerd row © ij by engidence interval gor f 4 epee bt eratanen’ » veep) -0-732Gt vgeage-) -}asa¢» Fortaee] t [2-198 x ¢-gcreT 207326 £ F.932C/22 [-e° 1326-4. 9396122 -otsae 4 4-32 20/207 | [-Beeseya2, 3-Cecci2ay Ho Pe =o a He pr =e | Dectcton | Exercise + Test at 5% and 1% =X and eonctruct een “Use, ANOWA tect 4 Sinple Tin Uinear regrescron ysing F tek J 3 Pejvesoe Exerdse- : Rand Y are cceres on prtieni nary tinal testo dudentrin ctatidice | ee Ls To Te 72 63 5% BH aR “G4 30 CF Ty 88 7016 43 80 906g Usethe data above to tect at te 562 and C207 the polleoing sypatheces and construct a 959 ad at arpa cecrini front fe@) ‘He = B ch He *pree Ns Ne Hy feFe Hei fete sola 41Te dlotermine the values e+ and fs we hove thet pe cgxiyi-2aidue fl. Seppe 0B — (38 ke Ni | sige [oa Fe ae creo mend Gi-5)* | @G-2)F qo 17 |ss¢0 | 4900 92-56 30°41 2-849 TF 44 | eqse S4HTe 3Ss°53 qe Bk +9 4 88 |ca3e | sia Bt om ‘bs-oe 12-07 68 to | SHO [HERE Boy for DoF oe. a | ane 936 13-65 7.02 toc -07 Bt Te | lof | agg To 64 Qe 3e Roy tg 48 ta lie | eree qe meer 197-89 oF 90 | S120 Hore lo 3°F1 #9 40 qo |TR00 Gtoo 94-88 0-000 4 130.49 et I | #eoq [9721 18-49 41-33 53-29 Zaiess| eewe Dy =Kosy D347 406 |) acgi-4 . nero 1 = G0) (6,097) = Coaz)Ca19) (AMC #1406) - (C43) 2 + GEosqo - 66,279 HT HOSO = EGE ST fee 2¥i-Ber = Ny = BGM Se OT HR) Tse 912~CoT¥ai tes) 1° = 306. = 80-6]4 The pitted line is es il f= fo © pa Y= 30°61 + O° T4210 : “7k hi Pbewehao i @) He = Be=o ve . Ur! Pete we : I be » eG) > Trarepy | vache) = meal 1 F se Cfo) a I = ee L | AeRB = 75 (088) = aa-3 Ps | { 4 a } MSE= 2C¥e-¥)% - 27-7, eT-RIRS . a8 3 : \t VarcBe)= 27-2125 [i + (ers)? £2) 1se-4 ill HL = (10-64 34853 (SEG) = Atro-e34483 = 13-0014 l £2 90-61 2 R-8435 ans s ilk is-8rte | | At &=s7 it 1 Eerste) en-29 # il | | | t aqrs,a z LEDecision : Heal > Hea , ject te teal < thab tt b reject He 9.3495 72-306 hence ve rgjeck He Gnelusion: ince we have afected the nall hypothesis , b shous thet pte shgerent fom 207 jos dis datictcally sigtficant- ~ confidence ‘intewal ‘ fF tt efeyen-a) secpe) = g0°G1 t (8306)-( 13-0619) go:6! t 30-1807 Lo: +293 , Go- T3071] Hence reject the null hypethers ‘At f= 19 Fees ena 2 t cy -o2N)¢r0-% tee -aasyea 23 35c | Decion % Be 3425 23-355 hence we qail t vejedthe nall hypoth ecie- Conclusion Tce we qailed te vejeckthe null hypathests i€ shour that B. te equal bo tere ie} is ctalistically tnsignizicantcongidence tatenal Rett sechey (tL gC 0-2) j gover t Ce 355)C13-0619) go-cr t t3- 8227 Ets-zret, 14s aary Fail to Decision’, Reject the aull hypothesis dh Het Bree tht pte te Be se cB y= Tverch varcpi) = SEC) (Nar CB.) = 27-8195 =o 09599 | 156-1 SECB) = [o-gsrq + o-l84T} | t= ote) = 2 4NITG o 1847) At test | Fey -ettgyeneay 2 Fae Fa) Cto-8) te T1S33 = 2-806 Deciet en 391767 2-306 hence reject-te mall le pothesic Mset a = — = iconclusion Cine we hove ected He onl hyper ,itchowe tha "fs i= digo pon 2090 1", dette Ssaaiglengh internal * BE tepsteyenay se cp.) or TH t 309) Co 8771) oT del toe BTE TIRE To 304¢2%, 19517 Beak Decision Reject the nll bypetheci: Mast Leesigen-e # hy. 2! Yo10.-2) * Vomgc,4 2 3:25 Decision a-qittg> 8355 heme tech the nall hepthasi conclucton congence intereal i £ by -afay cneay SEPA) o-14ai + (355) Cantey ofa t 1312345 Tp a-3 css. (3 -388],FESTNSES aah nr NF ce econ ANOVA Caray of Variance) A Nova = Fert in Regression > Te decomposition of Sins op cqMa TES or FEgrEECION fakes tha form op SS SSR SSE cher; ; . SS = overall or tetal sum of-cquares ssp = tum’ of squares dueto Regreécion: SSE = cum of squares ue bo arr. “ a A és 2 Cai 99? - SSR = 2 2 i me * Deby ton Total. =in-t > regression =! we < . “error = 0-2 Ceror tp ic anp., here p=) ‘The ANOVA —TRALe™ : Ks SOURCE | RR" oor: Pel jpeatentat pe ae’ oR Pe ERROR rR ose 1, ikea == ee - :~The falysis of Vaiance table fe also known acthe #NOvA: fable itt - hour the regresston equation accours for variability inthe responce vaiable- + The column labeled some has thre clusrna roe’. Regression » Resielual ‘and Total - ~The celemn labeled Sum of Squares describes the variability m the > Response variable gts <2: The fetal amount: of vaicbility inthe response tetthe Total Sim of Squares » BC yt-G’. eee al The row labeled Tolet % ‘sometimes labeled Corrected Total where corrected reyes to subtracting the simple mean _ bepore squaring ancf summing. “le a prdictton had te be made uithod ang other injormation , the Vest that codd be done, in a certain sense + ic-to predict every vale to be equal tothe cample mean - The emrHal- tc, the amount o vail on tn the data that-cantt be accounted gor by Hic simple meted 1¢ given bg the Total Sum of Squares i + When the raqrestion mortel 15 dsed pw prectetion the errarCamount uncertainty thet remain) ‘« the saviebility about the regrecsien Vine 20 45-$:)*- This te the Redelral Sim of Squaree Ut ts somelines alled the Emr Sam of Squares: ~The Regression Sunn of Squares 4c the digjerence bebweon the Tetal wm of squares and the Restelaal sun of qua ree - : + Since Khe total som oy sysares tc the total amount of variability ‘ the recponse and the residual cum of cquarer Ahel-ctill eon nat Le laccountee| so apter the regression medlel*%e gitted, the regression cur a squares ts the amount o¢ vviebility ‘in the respon se Tebse aceuned gor by the regression model ‘\ ‘Gch sien of- saaawes has @ conesponding degrees reese OF) | associeled with de Teal 4 tsar, ene fers than the numberof Lobse rations. .= The teqeession DE “rc the number ag irelependent- variables in the-oreilel:« * paral linear regriscion the regression dpitc ts The enor des icthe theta betweon total de cade tegression p= » fer dnl | eens cect sthe residual dp ten-2 ANIFLNSIS OF VARIANCE CRNOWA SOURCE ody 668 el Mes eB tet bueto rescion] Ket Screg) Sscreple-t _sscegy Tier esCredd > fn-R— Due te residual} n-k ——cseresiduaty secresidy [ak Total art 6 ctotgl) bethe aumber siete j » s “| sScfegeisiony = 2958 nf? ape 20a eceredidualy 2 Bes) = ZOVE- 4)? SCtoa) 2 ZH = WY-Y)? = avat- ay’ NOTE | SoCletaL) = se CRegressian) + $= CRecidual) “Exe le Mend of soy Leone ql ants “rege cblained #n feponse ty the indicated lovelsog ovene a ever the groding season beeen 08 | oot] on | ors eld Cam ppt) geo | 037 | agi | ae & Compete ig Fe id ger the linear os scion of yield on intercepted] erone expsure- Wi he regression quatro:hy Place 48% er of by and fo - Interpretthe antenale: 0 2. | C0) Teak the hypothecis “p Hx 4 Vs" Hy Beto using tHtact i vith az o-1 Is gour recultoper the 4etest wadcent-uith the, Lge I “gar Uh) 7 Explain. ‘ ro E re @) Viog ttettetet Ho: po=o agin Hh: foto. tntemrct-the woul. | ©) Gre the ANOVA rte Tegeeccion through the origin and yse-the fi-teck be test Hoe po to + Compare the recalls of the 4-tect and the Fed cees% |) cdo “~ & x Ne oe z | o0R | aHe | #3f 0-004 ) OT | 857 | ie-5q o.004q +0: 0003 625 | ; jon | i | RBH gor gy lp ce see 25 | ols | got 82°15 or oRas Le 6034 6496 i Aeizo9s | Dig Ble tey 2a o-0nT ” BCzt-Z)* o009 era] f tr = N2xy-2x8y = #1699) - c0:38)Can) = ~ {9-24 ! Werte) RTE ~ C0135) CaN) «= ter aq_ next - Gap © (0-03%4)- (0-35)* 0-037) bi > -2948- 53094972 | on rs ? 7 Z= 0-081 4 9 he for Yo Pix Y= 27-957 aut ~ (298-6 30%413 Cos). "aes - Ae 8 227-15 - C25 09 37626) . Pere for @53-4339cR3 > “* Je . \ ow Mes a oe ee ee ne } Yee BB 43346 Q3 = 293653094 42 9G aula ge# Bit yaaa se) | secho= verb) 5 varc Bid = wick S 2044-2) uses 202-2)4 asia cade Leela ja QI Sess eet 30-1501 1366 Q3R-29614 25 | te 14d #78444 5H, 230-2 5312 gal 1H SSSRE \ q1-1o1g3se | #9908" 8553 oq HoH 3127 __ 1 een ae crn | IMce = 21SCUAGO( = toT. 8057301 ae \ a i varcfiy = lovsesizel 623-28081 | 0009215 | \ sect) > \yens-at097 = oT 42207 | ls | 1 |= 903-5308 13 t te 194,948) (104; 137049) \ 4 | Fegis,a * #993 , | | 2 9q-5309973t Cq-308) C1of-4 13801) = 243° 5309973 £ FESTA | (10. | Teter estos, rote 3810151 i, Decision ~ We gail t reject the mull hypethese- fe 1 bey -efay eneay SE cps Fe-efayen-a9 = Ter esp yep-2) » toans,e sece.) = Avarc pod = 3 var Rede me + (07 305 9301 [4 + ears | 0009215 2 Sbessegy ts. FH VLAOT seg Atieqqerzc] = to-1e7¢ #122 953-4339623 f (+-303(10- 716764792) 953-4937¢923f ug. 3a3;a¢ [eot-teortes , 299-167 1493] Decision. Rejeck the ell ypoth cis! diy He:Br= 4 Hrpige Aeot n > $308, a ZOG-X)® 14se CE) = + 84 35209973 - 1 = be q31a1547/ | ee i toy - 81 BOT 2 9 Ba 591 Eqeataneaay tueeycen > nese \ Decision: Leal 7 Hab - ject He Lal ZG "-0 . : piece 2) Bipteny® = 909814 ex ca271s)*) i = Taq 1392 2 | | ee BCai-BY* = (24 3-5209973)* x 0 009215 = 194-138 Mot “ssCRecidual) = Zee = ZCI Ys)? a! RS CUBCoy | - 7s n | go CTobal) = ya BON -¥)* seven? SoCtotal) = FNS GL 174-1388 = lol 8% |. Anlovér TABLE SOURCE s ay st E Due bo regression 79q-e) | NTH Roto readval ateet (hay 2 2 arse] ae 107-305 THT8! 199-2, TY 27-4 TT erage CH [ot organGritial value Fravese = Fresoos = 5} Conclusion Since F> 14135 4 13-54 we il fo refect He That the dela do not prove corcushe evidence of- 4 linear expect of oxone Co-eppicient o¢ Detemfination- / One measare othe contrhtion. es indogendent: vabablece> in the nicdet tethe coepyicient of determination doncted by pe R?= SSC Regression) SS CTotal) «Ths ic the propstion of the som op-cqueres ot y attrtetle te the 5 raration LS = I “The coeggiciont: o-detonminertion ranges from o te 1 ards of He predet-mement- conlation Febueen 5 and fz « «From the above examele, the coegicient oy detenaination fs given by: e The square of. R?- 199-2 = o1815 loupe 15 = The interpretation of RE ic that, 124 @ the varsation an the dependent variable , yield o fae tc, explained by As linear elationchip with the independent vatiable, ozone level: S >Tn citing Acsumptions for, Regression Medel 1 NF sesnormally dicibited | —Normality : 2. Biterm is normaly cretibuted with mean © ard variance a? 2 Consteney in variance € hornesced acticity) tt | Lonwaly BB Eis and i's are intependent € No mutticllinearity « 4. Ther ie no seibal conlation Depatures from the Linear Med el L In this section, we Fghlght Siletypee epelepchies rom dimple liner + regrescion made with neal everc: ~ ie wofele fo conceler Me ace of resceale te onansire these dopatirte “\The 0 departures include |d) Regrestion tancfion 7c net (near £6) The enorterm do net have cenclart vatance Cheterscedactierty) . diy The eter temcare ec-indepenlert-crmplying that the ebcerations | yp'e are rok independent i fw) Octliers lay Eror-terne are ost normally eihibcted. (a) Oreo Several preticter vaiables have been emitted pomthe vetiables, Piagenicte og recidale < «There are some ingormal chagonicte plete of rectdluale heh can be useel to peti init er vath er ieg the Ctypas opeep atures hao eccaef “the Housing plot se recat wil bo abe hea yore argent” |p plot of vesducle agente vara. on < 0 5 ‘& plot of residual aginst itteel valyes- : db plet oF esiduale against -timerer indore} sequence- duypot-og endelaale againct omitted preter varices. iv) bor pet of esiceale- Gy Nomat probity plete iceP Greneraly | the plot-og resicbels® acinct vehi chould fsb reretom| som lyre ote bod ty lets e pattern. — tev ling - Non-linear “ip anefion. - Tite wen led chee ing ap lot op riceuals agaiadt valves. hy the ual pier lel ing Snel eS nen linecnty ie posit le Fe dela! i ite veluee = oh “ns sporthic fs.to: aller the regedit bee I esq ~ J FPO tex o ye pope ie} qt sonething eke: x i consta me of ree CHetescedesicity ) Ths soled faded by the rotate residuals a Disdntade, Ae the cctttering " pia mel fin say sty, Hs te is pasty °F Hons es varhance: i "4Causes a heternscedladticity- 1p ne hove stn 4 ale wh fave atc role ly we lane outa pata plan ie : \ Remediec o beterscedestiaty + - 1+, Speci ging the inodlet cone sored Ve} hee alt tere variables... 2 ‘Use ather sect mation rocedyres alte bh qccount-yor Non: constanby Navienee- ire oi sa apport. a heaton | al The ean le ie asin abex Herter a oe rb i leeds mene ca values J t ; Yat" pelletsRemecties: 1-Use bust ectonton presdies 35 ridge agrstion, m-edimaren ete: + New indlepercone$ op arrarterms.celueverors) Whenever data ic aldcined tn 9 time sequence Wace gece teen Bs ropare sequence slot: of-recidbals. : a intention =k see it there fc any conclition Letueen the ermrteans Hate nearte cach other. \ _ this important cometimes te view {ho prelem of-nenindependonce ofertor ee ‘asrene: in os an-impartnt-va ralle-Cin Ake case time) bes beon omitted'trom the mostet. - era? oe Remedy: Use comeleted errerderm madet 2°, renelom oftect-medete- 5 Non roel oy-error terms- aif eaten: a “Boxe ad bdr of medals be ed eed pe nay. «rile rset pe grey date hen we cagectte Ing le Sym mete al ot 0.eepochedl Values: on \ \ “Bech eda ie pleted ans expected valves under normal ty ~ “Re abated te nenly Vinee egg agreement ith, neimalty acconpbion z | Remely . Avi can be adfioved by bento ebservationis - tale: = panegometions ‘include 2 | a acre ‘05 int of acing g ac you vespomse » use e as ee map iy Lo transpermetion 4 iat 1 | m Deon panes sl omission 5 : 6: beatles, of 1 ‘my anlertiey ie ae . analy oy Pot the wadal ot wi fe ic any pttem, tren He walle ‘efogerare an Sh Rack 4 ind tude the Paicbl loin the rnodel hence 2 mig (near rgfeston-Formal Fes-invelving restdva le 1) Test ef randomness inthe’ eric Cdopendence-anten theenrr<) | We use Durbin Watson Test te check gor aye ceonelation between ratiduals i065 Hes P =o ve Hi to I 1 ~The Datkin- Wiekson Test cticte, given by D= 2 Cer-e3,)? \ a 2 BE a here esi}. : i o Tet yor normality - Geodnede of gt-dect cin be acl do-exairine Hie normality e-em tect. IL Kelmegioes -smiraey test Segeequertly ued. : 3 Teck gor constant veda nce: ny * The Breach Pagan tact i sequently ased te fedleger ‘constant vatiqnee r Chemescedactgiby) math ag + Inthic teas He ascumed thatthe nen concent vofign @ arises qiom the i relehionship between’ the variance efernr terms dened by 7,” and precPcker ik “veriable Hae plllowes ome oni oe se oy tly, Tie Yet + Agricis . The tnphas that 4 increases or decreases with the love | og: 3. * Ir2 r Maltigle Vinear: Reyression “Suppose we have more than one protcorceeraeveille say Pol op them yey Xigka, <2 Xpet > the linear regression AG accomodate ee ! vavalles is called oulfide (inear seyression mnodeleard ‘given bg - Xi = Bet PX tBetnt oe BB Mipet Ob ' att em also be written as, 4 : ps ee ae 7 clgae let tio = 1; He inedel reccee-to LL Yer 2 Bry tee 7 . Tee ‘sa deight-poard extortion af cimgle linear regression mmedet. “Moti grmalition | : | ‘ | Yas pot Be tnaBetit Rp tipitet oF 1yRsye0250 = lichen at | Vee Potptnt Bate ti pata toe Bp Zap + 0,* — ge a Ya spot Bi Xart fa ten t fateat Bip Repel $0 en 3234) Ys>Bo +B Aa) +Parsa th.a%s feeo Hips apen tes \ : ‘ t ‘ ‘ 3205] Yrs Be tfiXan s Barna tfatnst =~ f Py ap-it oe _ ts red fom, - i Ty t Xn Me te oe Kips +. e j ee OS a i ‘Ns Vv Mae M88 apr & L Ve 1 bs : i : Ly re ee eC oe | On one pat| A cae we + In matix form, the general muttiple regression medel 4 a¢-the-porm 5 * Yexp re : , I amp toagl pm om, ; Vrs avécor of responses Cdlepenclent-vaihles) > is a vector of regreccion parametere- e a re ae keene matix made 4p <4 explancttery yentebles ane one of the lf € = tothe vero’ op Sade ponent nermal rence m veal uth E):0 3 To 20+ of and varce) = EC eer) \ il Ce Variance of € i Var Ce) = ECeet) i] = h = ECE, 9 €s - €q) 2 iy 2 HI Sh, Bal Vo egy &n fins apoP age beta ee va “* tn SE( Ga 8 bets - eae, | LGn rn Cnty Eres a gtNote: VarCe) =E(6*;) 20° Elerej=o i 94) > = BS 6 0-0 7 g : Th zidegt g vf og aes. 0 0 ower : cos cnxn) = 1 oo j f H ) ot}o to € ° oo zo Ty Ex Neo, oT.) @~N (roo0---ol ot Tal “Mean and Variance oY N= Xp re ° EoD = exp +695 Ele)s0 EG) = XB avd the vaiance covariance mattix of te the same ac that-ay-the em tem, varcy) Exp +e] AP reonckang = varCxB) tyarc€) " Trcmetitea 4 = oto"T, varQX) = 5°, Yo N Cup 10° Ta)Ap" = . Yee FY Hi eq ‘ Estimetion of Regiescion Coepictenls t Dedinary Leaet Squqres Approach Cots) The cheategy ‘in the leact-cqegies approach ithe same as inthe bivariate tinear (gression mo: lel - ete + Fick wo cokalale the sum of equeres tesiduale and ce; ind a set-opachinatoe Abe anirintize the sony squares due te residuals - “Tac i alee ite «the sum of: spacred reciclals ‘in matt gorm he dwen by: Fy Tes KBte \ eaynap * ; a £*2(-xp)° i eee CY-xB)'CY~xB) =~ yixp “ae + px'xp | PY y= ak'y tpi ” lt o- Vy - apy eatenp r i Then by faking the detivetive ‘wet fas hove 5 ih [P29 = - ax'y + @pr'x if oP He BAe 9 =F - KY F2BKx 20 ae'xsax'y os pxtx ox'y ahd pa ih Pr KY = KYCx'xyt . AE xx : «Thi ithe lak pives cbf the male ineor egret iL model in matix fem : i,~ 2 [big romel}mes alled. erctinayy feat spears Cots) etinnaton: F Unbiqsedness o¢ the OLs estimater , é «Ip Bhtcnbccdion vet houed ubioslness of he oti arte qalloving ascumbonc: 1) Linear ‘tn parameters: \ Ye xpre ~ dh Ze’ condiflonal mean - ECE) =o os ] iy No pergect cellineaiby. ‘ee x bas rank k faa The ob eirnator gen by # ZOUAY HY), we uate chow 7 Ech =p oe co alld sas + From ascumetion d), Y=xe ce b= Cxtxy xly fe Cx'xy pt EL Beto xicp tix xe fs pt ater x's 7 ech = eLerccxy'x'e] af aye ater 3 slo VECE)=0 Ech) = ECB)st en Ete) anes ec) = ECR) =p . + Under the assumptions diy tociii) the ots estimator & anbiased® Variance of ols estimator of © Next we consider the variance of theestimetor@ - show thal 2 ic BLUE Matencereythe-erner-ten-given “var Ci |x) 268. and CY CEB) 20 RY and thu var Cex en* Ty, * Because apthese ascumptions we have, , Peper”! ae ° var cf > varesd + var Doxey! x! e] varep) = var COKx)! xe) Y-ax ' ' ~ | NaKgy) = 9° varee) i | wach: Lowe T® vere €9 Bo a hath a af I = RR CR KY” varee)y |e Ce) n Cx ey barnes 0? T, POT, cxtay | 20° I, xx) FOP GK) ° unbiased «Under the ascumplions (i toi) fictre best tinea restimator-ce Lue) +|Snar’ 0% is ankaoun, toon be estimated bythe use + 2 as Mse = 3 (1}-4) e i = pro ef prrameterr- = Vy -Bixly ner var ch = MSE Cx!x)"! * ad | : |degrees ce-peadin Interval estimation CT tte » This Gs an extention ate Frtenal ectimation jn the Amd near regression « > In single tinea regrescion, and} ver loin adimatiog Beand Bs - “tn mabigle tinear requecston, since de are edtimeting p premeters, (<3 Perf foros * ogie loose (lero con, Hiroleads for ars 1 - ie 7 Bk SECh,) w Ene where Ni | Ke Otay apy Bevede Maximgm Likelihood Estimation : Inthe mall y=AR +E, Wis assumed Abalthe erosare nonrely didvilared 7 with constant vorance 6? or @~NCo,0*T). “Tre normal dencity sundtion por the enmic Ie Ce = be os : f exp [- set] 158s ood oXan Te likelihood nction isthe ink denny oy 1, €45---98n gvenas 5 Lp o*)= Tt tee) ty "pd 1 ys ols opp Ut eth 2 4°04-8 one | ef =, at £ Y, morse OP i gig ef a dat i ye eld af joi egw mire 5 7 Gpoare OF 357 * ef "Lash 5% nota(x07) FP fSyetep! (yrxp2 f ara me log ansfernation temenétenie',: fBitestonde doitheninteretk. termisn the meelel and hencorite) 4 7 prall LAE Not oy Fem sy Suny + Tite 1¢ an ovenlf orglebal tet ay model atequag Rejection oF the aul| hypoth ect veeledee thal af fead-one of the cre et a Keke, o¥K combats ignifeantly fo the msll Thee é called | analyse opvariance . Gre e~NCo,o* I) so g *NO®P ol) | _betxigy-'x! AN Gp B7CKY!) Fh a beh & ctarsd ath ‘i B=" par \ Bs . ‘ + 4 Sreseuev nalts bastiniyes. oo | inp onal amb Zed bar aus BY Ag 299 ye ako ~ a-K | Gince Cx XT Fo ob end o° ae inclepenclently cebtbuted -RR MOESO* = oses yy “PANY 2 SS eectdual | Senor Shee y'ig= ee and A ts 00 idempctent- mobic, co Stee ~ hien-ky, be; cembal 2? dtctibuton aith cn-k) clegees op-recctom . + Padition ys Ek, x87 there the submatite *%, contrins the explanatory varlables XoXo, 00, Xe vand patition pr Lp BI there the subvedtor rs contding the regrescion weg fctonte PeoBo, Be \ = New partition the tel sim op spares duets y's ae : % Sa gihy > Srey t Bes where cseeq = BY x! A xt by tthe si actos cee 40 rearescron XE AN im of Squares eee to 009 andthe sum of square duets reslualsag fs fen! by: . Sees = (g-Xb) | Cy-xb) =y! Ay = Sq - SSreq Farther ¥ OY el th : es ~ en, [t | > 16. 900 combal 6 Adhd on with non centrality parameter fats Br, | & Z ayn tay® pt Pict ~ Bas [ beet ie non antral 7 detbation ath ron-cente ity parameter By Ax", Be rT,«Gree Ka i =0, 0 Seay el Sex ace repender sil ~The mean squares due @ regression fc Msreg> SS reg nok Ther Moe 'p peixtiac ) MSas Fei (a Roe f cohch i « wansecal Fatale ith che tot dec or and aoneentelity pa rameter ate “Under Hs “Pet Be 02 he Fe Msc ~ tee Fe “The clecteton cule te 40 refec ak eC level of stgatptamce ela | F2 Fa Ce ,n-k) +The ctlaletion of Fistalidtc aay be cummartred ta the yorn 4 a jel of variance ‘ebb given as -ollewe’ | Source operation Gin of Squares Degrre ey qreeden Wleancquerec Regression . Sieg a Ket Woge Sh b Error Sires a-k MB re = Cres [6-8 Tas! Sy avi eR 1 ae5 Egle Edie catght eyo orp y ig “thought te be related ergenic: matter i x, and the emeynt of mrstcture Xp in the coil bythe tines eqction: i Vee, oF BiXig FBaXas FEF The jellosiog dats were cbconce on our tigerent plants shee he erp i was groun. i . 1 tt 1" I (%e 3 3 @ 3 & li - Y [oa 1 to 46 = ‘ 7 ea Determine the (epst svintes narescion o, Yon Ke and Xe arel ace your eoults | bs ett he cle begll'g Be cmp hen Be & enka pone 7 matter anol 2 units o¢ mestyre in the seit - cy Deemine fsa led of signtgtcance uhetter the agaafc matter hae ang eggect on atte weight og the erp: ! yr mte 5 te He i 4 r T 8 fe e 7] t OL t 3 pl t [es f i | | fot ! tt 6 fe e3 1 16 ! " efe Cx) RNY ee | { *T ' I 3: : |% 3 go | 848 is, 193 act ony: Te Bo Is det= | 292 FERS a3. es - 30 AS 2g2 123! {a3 63 123) - 80] 30 ts 63 123 °3 hia 30° Is opal values 292 123 it al lee Nee, we C@aeKe3) ~ (1234128) - 30 [| C3exes) -sxi23) | +15 [Gexies) ~ (15x29) L = [306% 1350 - 10350 21368 awatiie of co-gactors a + - + 1 a a +o - ++ fea =) 3s _fse ts 13 63. _ fee 133 ‘= o3 4 [ee - ' 1s 292° 123 30 yas B26T as -6490 = 5 eq - #2 690 =e Ree _Mipit Tenspare the omit oy cope) =[3267 -ts -8I0 ~ fs a7 HR -290 FR 208 Itnese = Aside a" 2 ot 3267 4s | 1368 - ts a7 | -690 =a xy = t rong 4 1 rota 1" 3 3 6 3 tof “690 “#8 248= B43 at Bt fe OxIxy KY BR6T “45 -@90 34S PL 4s a -#R am o Be E40 4a 2c 134 e4 TBoxs\ we 1368 Jeor spot : 5333151895 53-321¢ titpesse3? | = bet OR 4F 41 SETH 249s hee 53: 38k be roan Bas G-8895 a4 és t Yo bot Bite FP akei: 42 s8-38le g MIN He +6 8895 Age Mi=s, Mar® \ . sa-sare ¢ rims) #62995C2) 21: etch Hor Bree Hy prt © Leck shetiche ts Alp Sec 8) varcp) = MSECKIx) 7! wse= Yy~ pixly uni yy: (14 1@ to 26) 4 Tr | Jot ae = 245929 : fis y= Cs33tte rity ge 249s) & 217 att = 29925.997 pes net "Pp = H-3 FL Mse = RdFes: 29904 -2999S- 997 5 B-013 \var cf) = 2-013 x 3807 -te ~#70 1BP8 -4s a7 -t2 | -c9o0 ~ #2 ta qergss | 0° OFF] =4-5197 -0-0997 eo osts = 009Qg5 = 1507 -009@5 + o-S90g \VarCfe) = T1455 i |Var CBr) = o-os4s var ch = 05408 (SE CB = vareBs) = do-osts = 0839 Uitt-o 2 #3018 62437 yn | shoasy =eue gail tormjoct He since tale & 5 pail to reject tre mull hypotheds ant sonclude thet > the edimale te G6 shatiseally Inca feant- Lynn 7, Mei & = os a Hor meanc qe, Titermnt HY Tiptevk fa eset | . T megs are dipperen | pralte for income = 0.000267 st laton a lden Tor the income ° P-valye = 0:0009267 40-05 fs not the cme. I We wgject-the ru(l hypethestc andl conclude that-at leastene oa i the means ic di4terent _ lead | F -vdue = $4.¢09/ | table Fafa 2 Fag og 2 +10 j one Heide i MN S18 34-9094 . * prealie-por Piles o-1a¢98z > Ovo We fail fo rect Ho and condude that the meens ate equal « celealeted | F-value sg. 0ete : Fyyasores = 51g of Mle ded yom tec:| Decision « Conclusion | Giiy Quantity = 83-1694 t 12.4098 (income) + 0-193 C pire) \ if ie negeives it dettenee = Helding ineoene and pice consbantpthe quantity ‘inereares by 33-769 4 ls bs re t iP _ |For { wietnereae in income ,queity Sncreaee by 12-9638 Be ee ‘in price she quanlity ‘mcreaces by ones | A fr Eb esigeep sects) \ 4fiq e005 | 93-1699 tha gXeteIaF Pewee TUS | [ay lets t 2.26ax ag stat {asters £ ss. ete | [a2-reqet seate, aatere tr SE 44a] | fe aversa, eq.artal as He por | H = Hy t pote . 7 | - -valye Boz 0+ 202 eon ee FX ie} 0208 7 0-05 il forefed- the end conclude thet i Z foie ack detidtally gniffemt . i a He piso Her pet o pralve Bre mot E | i Ke0.06 i i ] i Crom co es weet Ha and eondude Hat Br ic ail sgt 2 ii i}to: pare Wr Pa to ' Prdve fe= 0-198 aro0s 0-194 > 008 5 gail to reject Hs ancl eondude that Bs ts retcedt Healy signigicante | N i) He: free x Hi pr to cl g tvalues | tegl = 3-042 citi vue * fata) carp) FFTOS | to-qrs,q = 2° 262 | Dedsion’, | teal rtieb 5 pate rot Hoye, 208 7 eG? | exrditons | Since we have rigoceed He , Bite statistically sigat ant |W’ fa 20 i “Be to late 068 tab {Cited valae ! t yuatgyaep £2005 bogie, q = BRR _Desigion Since tab > teale ,ve fall to wjed-the nell hypothecis tre 82627 0° 24g Conclec'ion Cine we have qailed to mjee fhe, Ba we Stitt aally F ncignifieant:i) (ih Leos il prvalve = 00007 935 i} 800019354 0-06 | the overall model i srtictcally sgntiteont or 74 te git - | OR | | udng the R¥'s 04954 > oS thu ver meet eg T9-S#% > SOG Wi) Y= 33-1099 12-4096 Inceme ¢ 0: 1293 pice . pice=as | Income > 20 Quanity = 33-1699 + 12-9460 90) + 0-1273 C48) = 305.2353 * HMULTI COLLINEA RITY - This occurs when there ic wlaiondp bebueeri the explndey.verablec ives theté ts high -covrelation bebieen tke explanctory: variable . Maticllinearity lends torngleted dandand. emort oy the extort » We con cece or malfcelfnearity by caning ext poeice eoveation between eqlenctery vatablec 5 | Model balding process Model building precese’invelves. phasee \ 1. Data collection’ and preparation \ 2. Reduction of explenatory variables 5 Modet: sctecfion 4 | Model. valid afion 1IDae ealecion ond preperction. The vaties de ending on the type of thd, berg cared ot Tee typer oF dies are commen’ | dy contral experiment |< Hare the exgoiertor eee ath (of explantory vere Gi) Observation of dudes + 9 bang aio p80] |= The eiphtinnerten hte ne canbat hele ; avitthng art whales atawatl set Hal alkge Poy Bll). % «fet the dda caliente dele 1 ele pmpare he ‘enfry errors al Rede o explanatory vaiables "De several simple Linear ceyrstion viedeband! tenth vaPabler thal r related 4» the recponse 3 Me calcon Qeled 4 subset e¢ the remateing variebles ukich highly onplvin the ‘response varable- We ere ap with less complex veiables med" Giteha ger model selection: - . > mod- common one the Akaike: Injonation ectterion ChEC) + The emaller the Atco the betler-the inode AIC: & Rsdignment Cavite neter en Ale) "Medel validation > > we eaye are ible: at ‘ &) Gllect newdata and check themed el: predictive .copalslity Hall 10% ap the Att eng the model stn and ase toto deck the precetne “epdlty opselected models es - , eAMedel Tecting ~ LA natlige teoyesien ts an Arteractive Ctepatse) procedore) ~ Ore certs ith « sctopexganetry vaiatles ananged in Teesely creer! og thyscal importance - 4 i ~ Wether ted-the signiianca ¢ Heese vavinles starting ith the (esbimprtent Triste ealed beckaaid efirination preedare- The chenges are ‘then rrte abere necessary gllowing the rents ap terest 1 so Initial significance fect'on regression ' \ °K fey gtting ‘he medet ue inadyinel evidence for Vina relabonsh Ts L tray be done in tuo days: “G5 Tedting the by pathecis, © +" . - lle Poe Bee Baz “= fp versus | | thy: BE at lead oneragthe FE fe nok equal 4 re - * | + We can decompose the fofal cum op-squeres ac.alfouc i GS, SST = SERTSCE . , “~ | sot > r-2 uy’ n ribo aye ge on 8 ead et | SeRe ATK - (8 ; ope Kt Saves L | A a x : | ests YY

re | fe > Pe agginst = Hy C tet ene of the: Bis to, Boyne Gitireel lve: Feajnxje’ 2 Fajijoos? > l4A'S0 Fal = 103698 ©Deciston ; i Since Feat < Feb G05 lo ac9e < 1974-5 2 we gail t reject He. Conclusion ; Since ee have piled te wiject He , Be PesB es are got strtictically Significant: Gy B® seR 2 ates Oger ST Es conclusion Since RS =e-qse which ve large, thereere the adel fopit- | loterpretetion 4S-4AL agthe variation tn the de, endent-variable ey) ts explatned by tks linear relationship with the 4 ependlent vartables Xpand x54 leaving 46% unegplatned : |Berdce- . : The ‘ellowing daly teletec dry actght, i pleat @) othe percentuei ‘orgente matter 5% Ckg) ard syplementapn se nitregen adeled) pon « (000 cqamCxey . N cg) yas ts 1ot-3 B16 95-9 PO%R JoRq Ar cKg VC) | 7 to " N 1 oo 38 | Xe Cogm) ag aq so af BR Ce TT cD Watts ¢ modet jnvelving 45 mp ante. ci) easy Compete X™ and xTy- -0- 068547 O-foo7T# — e+ 00 ORB CM) Given that Otay" = [7 1:1995973 roocsshy § -0-essygyp |. ~O°2SRIEHP — —y-oooge1 0-0570 739 NIVergy that(etx)” Sg inverse ay XTX ivy Compute : and wite the regresion eqn. ©) Intepet-each regression ceorpicient- Wi) Obtain ANOVA: fable and pergorm F test Gti) Teck the hy potest’ @ te fe ve) He: B= 0 ve ©) He'Baso y, Hy Be #0 Hrs pr eo Hy? Ba fe Ie (it) Construct-qs4 OT ar por and Pe.1 Xy SER in mato. ge ° 1 ie 1 2 3 ) Me ks fo on ad tPd "Ee, N= wate rato" Re eqs ae fF 2 3 3 * J te 5s WIS 12 QQ te 1g ee ee ce) to 1 3ltp hs, 1 37 7B 13, ta “ i+ Moo leit ray EUR S a4 Then silve for proto” KPYs ee he 7 gic Ss 17 Bs 4 a1) | xTYs (er) } zor (3 -7 TEN apy 2) bo = OSG = #6099 (21) P OeR ty cope ys = "Ey a fo- ‘1 (oi) (° 434 ‘form dsing 4 atk medo Reg +3 lin-ry tle? Mt > 5-7 MF Sst 3313-7 Mt oat -7ME pecs high Sh 4 zx? +4 > gs Bxve ny ag" 7 > RE fy R 4 Se Be key ; A preseahiptr 2 > ptf, a = FLOTTSCogs @ a rf higt et oem = OF ZFC 2489 ehoge #vedor efile [ete 40078 ly edor of rectdeate OST FE OT EH£ Ge {ia 8138 S033 q-Gste me ReRe 1 " " ‘ 4.2004 a 13-3182 3 3 1" lot 8¢ ez y- [te 13-4798 & SOD to | -] q-ceee is Ip 2eay ta le 269% aa to-ar8q EQ) pot hx Agron fe hdycted @ Pond inteyretation