5 ANALYSIS OF VARIANCE GD INTRODUCTION Analysis of Variance has been defined as the statistical technique for the “separation of variation due to a group of causes from the variation due to other groups”. Here we shall discuss the simplest use of this technique, namely testing whether the means of anumber of populations are equal. The method is based upon an unusual result that the equality of several population means can be tested by comparing the sample variances using F distribution. It may be recalled that the t statistic (14.8.9) is used for testing whehter Mvo population means are equal. The analysis of variance test may there fore be taken as an extension of this test for the case of more than two means. DIFFERENT SOURCES OF VARIATION Consider the data of Example 15.1 (page 619) which show the life-time of motor-car tyres manufactured by 3 companies. We are required to test on the basis of these sample observations whether the means of there populations are equal or not. It will be noticed that there are differences in the observations ‘within’ each sample or group. This variability occurs due to change, and it reflects the natural differences that occur With each of the populations from each the samples have been taken. If we examine the Means of three samples, we see that. another kind of variability is present. This variability is due to the combined effect of the natural differences and possible real differences ‘between’ the di rOUpS. i Our filer ec itetber the differences among sample means is only due chance or whether the differences occur because the means of three populations ™m which the samples have been drawn) are actually different. Hence eee Concerned with the problem to testing the hypothesis that the — 2 ae a) Populations are equal. Rejection of this hypothesis will lead to be conclusior S23 one of the population means is different from others. necessary assumptions in the analysis of variance are: |. The samples are independently drawn; 2 The population are normally distributed, with 3. The effects of various components are additive. with a common variance.Analysis of Variance «asa. (1532) F due fo many unspecified causes del, since xy i assumed 10 be made up of the “sum’ of ' components. tis assumed that e, are independent and identical variates with mean 0) and variance @. This model may also be (1533) is equivalent to stating that there is no special =0) (1534) are not 2er0). the samples = En, tal ofall observations = E 7; EE.xy 1 Mean of the th sample = Total SS = SSW + $58 The expression onthe left of (1535) the Sum of Squares (abbreviated $5) ofa {aml observations about te grand mean ands clld “Tera Su ecrgey ns andi called "Sum of Squares Seth arup about he especie geup eas nis called ‘Simo Sat ro NG | Mh aroun (SSW), oF Sum af Square due to Eror (SSB). The second term isthe in GD reowmgve ow ome. Wetavek indpenen ¢22h ot Nomal pop ‘WAY CLASSIFIED DATA om samples (or groups of observations). 00 £ ws with means J . Hand a common v4statistical Methods os as means about the grand m 619 weighed Sum of Sguszes of? jam and scala : (SSB! en, oa Sess Berwcen groupe (SSB) " 4 "Te degrees freedom an sila sumofsmas = © See MSB = SSB (k~1) Te Mean Sques 9 cles 99 SNE the Sum of Squares by Gi) MSW = SSW (NR) coesponding ezes of te 6 Obtain the observed vale of Fon dividing MSB by MSW Mean Square Within oe ‘Mean Square Between 7005 rma disibution inthe pop ith a common v Seco! ‘. won Variance (g EMSH (7). eject the null hypothesis, and conclude thal the population means are imespectv of whether the population means are equal or not, But not equal. Otherwise, they may be taken tobe equa ESB) TING where the sign of equality holds only when the population means are equal GD woo UNEQUAL PAIRS OF MEANS Inihe Analysis of Variance, ithe observed of Fis found wo be significant the conclusion s uve, both the mean squares MSB al. This is however provide independent unbiased es wetiveetstsmisic sa = Msw inci sums of squares, degrees of red, mean squares corespondag ere and alsothe F values (bth observed andthe sown as “Analysis of Variance Table’ i eee yown as “Analysis of Variance Table’ a ine considered tobe equal, at say. 5% leave of the upper 2.5% value ofr distribution with GED srevs zu compvarzon (owe-way CLASSIFIED DAS) the df. of ErrorSS, ie. the divisor used 0 Comeetion Factor (cr) = 7 = Total $5 ~ $58statistical Methods 620 ae 7 . = 52 tans CD. Ifthe difference berween tele pale ee ! act ne Cie te poplatn meats 1 2 CONE oe gh er apie of raf ares n=-¥Ey claige ember of obsersatons in each sample which WSs te ya discussed 50 far rands of tyres are equal or not. Tifesime of Tyres (000 boc [T a8 a 2 u 2 8 a oy 2B 382 wD 3 Also test which pairs of tyres differ in quality: Solution Each observation is reduced by 30, and shown below: fable 15.1 Calculations for Analysis of Variance ‘Samples $2190 554 = 120-964=336 SSE = Toal S- $58 =51.6-33.6= 180 yy Analysis of variance gable 15-2 Analysis of variance abl enweer CFOUPS fre) a ee Saceiie Oal numberof cates inal np 5 fr Toa is 4 Since there are 3 groups Of observations, the fr, SSB ory Son vale o Fs MSB + MSE 168 vale of Fa ohne fm Sis Tables ompnding ied hae ihe dior edn ner anand =2.Thed for 2)excedsthe 1% tabulated vale (vz 6 93), we eect tion means, and conclude thatthe average lifetimes of (ssi). Hee s= VES .0=5,taes= Citic Ditference = TS x BX «218 = i x28 = 37218 =344 ‘The sample ovals (ofthe reduced observations) eT; #2073 = 2,73 = 14, Wehave the Cita Difeence 8.44 we find hat brands A and Bae of 14 Cmay betakentobe ofthe same quality. as rimenter wished to study the effect of 4 fertilisers on the to 24 plots and assigned each feriliser at Comparing these Figures ferent quality and so ane brands Band C. Brands Aa Example 15.2 ‘Ss MS Fou po | - | - | He eae (4) Complete the above table by. in the values marked by — lete the above table als rig (b) Tear ate 46, vel to see wheter te frtlsers ier sie gy abe, one from exch plot esa eng compre. is @3-3)=20 Solutio “Toar Hoe (a) Working Note: ‘degrees of freedom (df)is 24~ 1) =23. Again Fenilsrs is = 1) = 3. Hen, the d/10°statistical See '"™ 622 i Sums ques 5) ate ae Therefore ‘within Group SS = ‘ 3272 + 20= 1636 “MS between Fertilisers 980 F= ys within Group ~ 163.6 =5991 “The completed table iss gable 15.3 Analysis of Variance Table zw 980—«S399 3272 7 c 22, 7, Table 15.4 Calculations for Analysis of Variance cre Sy lias Toul $5 = 2020-1125 2495 ssp = [OM # at |e (20+ 100 +784 + 1008 : +1008) 1125 = SSE = 895-787 = 108 od she various Sums of Squares SS) along wih fotowing table the dere of freedom (dare shown inthe table 15.5 Analysis of variance Tabl: ‘Source of Variation ios | G&D vecuwtque mm TWO-WAY CLASSIFIED DATA \Jom sample consisting of hk observations, which ae clasiied according into h classes according to factor A, and also into k classes according, ‘one observation in each of line hk cells comesponding 10a class weously aclass of factor B. These hk observations may therefore n of a two-way table with rows and keolumns. that any observation is made up of the sum 4 (156.1) model i ays weet Bey Jass according 0 factor A, Jas according © factor B, servation xy is supposed tobe made up Tris assumed tat are independently rn variance 0 (unknown). ors led a Linear Model since 0 age, tof eect of the various omPonern ted normal variables each with mean 0= data, the varie nay be Inthe two-way clasified data, of variance may be se i two types of problems simultane ether tee any ete ‘due to classification by factor ‘A. Null hypothesis: ti Hy (y= 0) =.= 0 =0) andor (i whether there is any tfeental effect veto lassifcationby fay hypothesis, i say, number of columns) N = hk = Total number of observations. 7, = Total of the é-th row (i= 1,2, 7, = Total ofthe jhcolumn = Grand total of all obse ‘Mean ofthe j-th column Fe zt = Grand mean of all observations. Algebraicall IE ive, Total Sum of Square = (SS berween factor A) + (S: + (SS due to En or Total SS = SSM SSB + SSE ‘The degrees of freedom (dt) for the various can also be spilt up lated as before, the corresponding degrees of freedom, 7 ves em between classes by factor A (MSA) = SSA+(h-1) Senate bane lasses by factor B(MSB) = SSB (k= 1 Assuming normal distribution list in the TF wit 20F at variance | be shown that the expected Value of MSE ie oo EWMSE) = g? at itespecive of whet general = Analysis of va Hance i 2s 9 EMSA) = o and Buse» 92 om ut EMSA) = 0° Hist nd 55) oy ey oF only ists Thus when panto esa F, = MSA se salows F dstibution with the appropriate depres freedom, Sp flee aie ve 1. Sinilaly, when Hes Pe 2° MSE follows F distribution withthe appropriate depres of reedom, G&D SvEPS IH COMPUTATION (s¥o-War CASstFTED pam.) T, Ree the observation by subeatng x cons om each 2. From the reduced figures obain reach group clasifed according factor i.e, foreachrom foreach group casi acoingt far efor cash (ip grand total (= 7; = E7/) for al figures Gv) total ofthe squares ofall figure @E x?) . 4. Caleulate the CF and Sums of Squares Correction Factor (CF) = zc Gi) Tonal 8 = 3g? cF i S94 7 +{F} CF 2 (888 . (E}- i) SSE = Tol $$ ~ SSA ~S8B r UNote: In the valeulate of various SS, we have expression like >, “p.j- in each of hich the divisor is the number of observations included in the comesponingttals 775) 4 Write down the Degrees of freedom (D.F): DE for Total $5 = hk= 1 DE forSSA=h-1 DE for SSB=k-1 Gv) DE for SSE = (DF. for Total SS) minus Ds. for SSA: 5. Calculate the Mean Squares = MSA = SSA+DF. Gi) MsB = $5B+DF. (Gil) MSE = SSE+DF. and SSB.——_ & bia the observed values of Fon dividing MSA and MSB by ays, =MSA+MSE = MSB+MSE | 1. Cnt Fagbesand sin he eoretical values OFF at (59) 5% * ppoprate degrees of freedom fen, te Degest Fria or Fae: Divs sor MS in MUMErIO one ie sed for MS in denominator] 1 value of F exceeds the theoreti sccordngt factor A asa differe the means of classes by ‘there is no differential effect, can of Fzexceeds the theoretical value, we cone ion acording to factor B hs a differential effect onthesge differential effect. as, Consignment 7 Wt W. A 9 10 9 10 B 2 un 9 u c a 2 10 2 ‘Also, test at 5% level which pairs of experiments differ signi-fieantly, fant (Given Fas 514 for df (2.6), Fas = 4.76 for df (3,6), and tes = 245 or 64 Solution Each observation is reduced by 10, and shown below: Table 18.6 Calculations for Analysis of Variance —_—_—_ Consignment Experiment oral (7) (Comection Factor (CF) = r a Toal $8 = Eos? CF =18-3= 15 Analyste of vg Of vars ‘ance ol {5 between Experimenters = T*E +1) + ales pat eos “ OP sr ere ere prcen Consignments « ET? $17 +1) 5 between Consignments SHOT) oy 2 LPH OF oy S85 due to Ertor = Tonal $$ ~ (SS between expermmentrs) ~ (SS between coasigaments) = 15-65-S67=283 Table 15.7 Analysis of Variance Table Sou SSMS Fob) Fla) eee 2 65 325 691 Pagal 3 507 | 1m 502 Pgh. 6 1 Zig - a ovat uo Be = = @ Since the obser ablated value ford. (2,6) it is significant at 5% level, We therfore concase hat the mean ‘mostre content as determined by the 3 experimenters are not equa: i. there are significant than the coesponding tabulated 1, We coos that the moisture another. thee ao signifiint ) Critical Difference (see 15.5.1) berween foals for experimenters CD. = VOaT x JTRS x 248 = 48 Wehave 7, =-2, Ty =3, Ty=5. The iferenes between 7 and Ts Me respectively 5, 7 and 2. Since S and 7 are larger than C.D. ‘Aland B and A & Cditfer T, and Tyand Ts exp aos __.._._._aae—— asefulnessin ests of significance. variance. Explain the “linear “Analysis of Variance’? Deseribe its a ss made in the analysis 0statistical Methods 628 ese the ectnigu of analysis of variance in One-way clay 3 esr cany te maternal model and the ssSumptions ed gg siving analysis of variance table iy 4, Write notes on: i {@ Berween Group Sum of Squares, 6G) Within Group Sum of Squares. 5 towing bleshows theta prices (Rs perks) of comma shops selected at random in four cit Vins a a eis D 4] 2 | 2 3 a | a | 2 % @ | a | 2 8 a | uw | 2B 35 Carry out the analysis of variance to test the significance of the difeeny ‘between prices of the commodity in the four cities. (Given, Fy; =3.49 oe.) degrees of freedom). 6. Prepare the analysis of variance table forthe following one-way classifeddss and comment: Weight of Be ‘Machine | | Machine 2 | 3 20 18 30 2 2 28 7 20 32 Total 59 60 90 7wBHS 8 fo mpletely at random significant (Tabulated value of F for df. we] ofa | 2 | w|i» | » es of electric lam he foovet i lamps taken from 4 batches show ieee om hes 1660, | 1700, [168 160, | 1730 1s, | 1s70, | 15 1500, | 1s40,_| 1550 Analysis of Variance e29 Analyse the data and tet whether he avenge lie ' c= fatches are equal ornot. (Valu of Fa frmmatis 0 HPS inthe our 4 To est the effect ofa small percentage of bbe sea 2 snaking toys, Sever blches were mixed under acca eae ts txcept for the variation in the percentage of rbber. Pom each batch four pieces were tested for strength The resus in suiable nie wes nn Percentage of Rubber wo [or [a 8 0 2 9 9 8 7 8 9 9 2 u Examine from an analysis of variance, whether the addition of rubber has any effect on the strength. Given Fs=4.26 ford. (2,9). ‘Twenty observations on the yield ofa cern crop grown in plots of equal size ‘are given below. The data are split into five groups of four observations each, according to the fertiliser used, or Used Yield (in suitable units) A WoFenilisey | 67 | 63] 55 | 58 B | 1 | % | 70 3 60 | 69 | 60 | 65 D 69 | 64] 71 | 70 E so | 70 | 79 | 78 (@) Make an analysis of variance ofthe data stating clearly the appropriate ind the necessary assumptions. 'B and C differ significantly at 5% level. Given 5), and fgas=2.13 for 15 Af data show the production of 4 workmen on three machines. Test vee workmen difer in respect of mean productivity: (Given Fas >, 4) degree of freedom) : Workmen Machine a A , | #|3 2 | % 3 [| ele ‘of variance of the 180 each from 3 bl al or not. Also "2 Apply the technique of analysis the yields of 4 varieties of a eroP ‘ean yields of the varieties a °9 block means.630 as Varieties A B c (nen Fag= 51 Fy 1095 Fr 2.98 208 Fes =4.57. For yy 66. : iu pe 1a, Me following table gives the estimates of acreage of cultivable by cultivated land out of 100 acres of total On ineach of three districts. Perform an Investigator oe ae T[a | m A zB | 28 | 26 B m4 | 25 | 27 c uw | 2 | 2% ‘ANSWERS g SMS Ffobs.) Fab 3 Is aa 92 Fys=30 DS 5 op — Prices inthe four cities are not equal. Gs us tines 320814 WiinMoane ¢ oa Total oe Cen: Mean weigh race bythe 3 Masinesa ot egih rtsab.) cs MS — Fiobs.) Fil Between s 2 om 12 me Far a ‘Total W 4 Conclusi . end aie wet ie Variatior “ _ es Frobs.) Fil protien SSMS, ra Tout 16 a 1,687.5, » Conclusion: Ay is ‘Average life isnot for all batches. Analysis of Variance 632 9, Source of - a Variation Go perween groups 2717 ag UBS) lth) ‘Within groups 9 2150 239 ‘as ‘Total u 2867 Conclusion: No effect due to addition of rubber. 10. Source of variation g ss MS Flobs.) —Fltab.) Beweenferilisers 4 7230 1806 97 Fg, =4.89 i 15 meas as itooss save dfn effecton yl ) Fertilisers B and C diffe significantly. (C.D.=259) I. Source of Mode ofS MS.Flale) Fick) penwcen Machine "2 286143 SS FygsOd fawenWormes 2 06 53 OB Fga654 Bror 2 1266 067 Tal 8 mo Conlon: Workmen dono erin respect of mean productvigy 12 Source of Var df SS MS F(obs.) —_F(tab.) pemmeenVareies 2 0500S. FygnLG3 PaweenBlocks = 3967322 Fyn AS7 Ener ess 281 "ou 700 Conclusion ields of varieties may be equal. of blocks may be equal SSMS Flobs.) tab.) 467233 063 Fos=6.94 Baweendigtes 2 (Ws 5 TS Parte Error 4 466367 Total 8 3000 a Conclusion: No significant difference between investigators, or between dist,