4.4, Introduction - 7 2. Definition —— eae ont Tome vs. Sample Random Sampling _ — ou fing vs. Straified Random Sampling __ ip oa i ts of Systematic Sampling _ ee UNIT I-ESSAY TYPE QUESTIONS ___ = 5.4. 5.6. Gauss-Markoff Linear Model 5.7. Fixed Effects Model... 5.8. Random Effects Model .. 5.9. ANOVA—One Way Classification. 5.10. Critical Difference (C.D. 5.11. ANOVA—Two Way Classification (with one observation Per cell) diy 6. DESIGN Md 6.1. Introduction . 6.2. Definition... 6.3. Terminology in Experimental Design ~6.4. Importance of Design of Experiments........ 6.5. _ Applications of Design of Experiments. al 6.6. Principles of an Experimental Design 101) 6.7. Replication........... 102 7.3. Advantages, Disadvantages and Applications... soe 104 7.4. Statistical Analysis Of CRD ........c:.-snmmmnnnnnnnnn nes a 105 7.5. Least Square Estimates........... assim -- ~ 107 7.6. Expectation of Sum of Squares ... 107 8.1. Definition _ “ . _— 112 8.2. Advantages, Disadvantages ‘and Applications - inher . 112 B.3. Layout of RBD.....cnnnes —— a os 113 8.4. Statistical Analysis of RBD .. — annie - 113 8.5. Least Square Estimates....... fete r soon 116 8.6. Expectation of Various Sum of ‘Squares... os - 116 8.7. Efficiency of RBD Relative to CRD. aston - 119 8.8. Missing Plot Technique in RBD . . soe 2A 9.1. Definition : 128 9.2. Comparison with RBD....... oo 128 9.3. Advantages and Disadvantages 128 9.4. Layout of LED... amen 129 95. Statistical Analysis of LSD........- . = 129 9.6. Least Square Estimates. i 192 9.7. Expectation of Various Sum of Squares...... : —— 133 9.8. Missing Plot Technique in LSD = = —¥ : 137 9.9. Efficiency of LSD . . 138 UNIT II-ESSAY TYPE QUESTIONS eee = 14744: 3. Various poblems involved in the Construction of Index Numbers 44.4, Notation 14,5, Classification of Index Numbers 14,6, Formulae of index Numbers .... 44.7, Simple (or Unweighted) Index Numbers 44.8. Weighted Aggregate Index Numbers 14,9. Value Index Numbers 44.10. The Criteria of a Good Index Number 14.11, Fisher Index—An Ideal Index... 21 15.1. Definition... re 45.2. Main Steps in the Construction of 45.3. Uses of Cost of Living Index NUMDESS «0. eeonenen 16.1. ABrief History in Undid... enqeneent 46.2. The Construction of Wholesale Price Index Numbers (WP))..... . Chain Based Index Numbers .. 47.2. Construction of Chain based index Number 17.3. Procedure of Convert ~ Hise ere "ing Fixed Base Index Numbers into 9 Chain Based 4 17.4. Procedure of Converting C Chi ie ean ig Chain Based Index Numbers into Fixed Base 17.5, Fixed 7s. hed ses Numbers vs. Chain Based Index: Numbers 17.7. Splicing of Two Index Number Seri 17.8. Deflation of Index Numbers. “ 17.9. Uses of Index Numbers. _ 17.10. Limitations of index Numbers : NWNHND DP wp19.12. Measurement of Fertility 19.13. Measurement of Population GrowthAnalysis of Variance 5.1. Introduction To test the significance difference between the two samples means of small samples usually conduced by using ¢-distribution. But if we want test the significance difference of three or more samples means, an alternative procedure is required for testing the hypothesis that all the samples are drawn from the same mean i.e. from the same population. The analysis of variance technique provides answer to this problem. The main purpose of analysis of variance is to test the homogeneity of several means. The term analysis of variance as introduced by Prof. R.A. Fisher in 1920 in solving the problem of analysis of agronomical data. In fact the analysis of variance is a powerful statistical tool which separate the controlled (assignable) and uncontrolled (chance) variations in the data. 5.2. Definition According to Prof. R.A. Fisher Analysis of variance termed as ANOVA is the “Separation of variance ascribable to one group of causes from the variance ascribable to the group.” By using ANOVA technique, the total variation in the sample data can be expressed as sum of variations due to some specific independent factors (non-negative components). The technique ANOVA is most powerful in estimating the amount of variation due to each independent factors separately. This technique is also helpful in comparing the estimates due to various factors. Example : If your fertilizes are applied to six yield of paddy in kilos is observed. Now the pro Significance difference among the effects of fertil paddy is significantly different due to fertilizer ‘isher, the ANOVA technique separate the tot Controlled and uncontrolle: fertilizer and uncontrolled plots each on cultivating paddy. The blem is to test whether there is any lizes i.e., to test whether the yield of effects. To examine this, according to al variation into mainly of two factors, d factors of variation, In this example controlled factor is factor is called chance or error.2s ¥* 5.3, Causes of Variation ‘The variation in any experiment is inherent in nature, The total Variation in Numerical data is due to a number of actors or causes which may be categorised into the following two ways; 1, Assignable causes of variation 2. Chance causes of variation, ‘The variation due to ‘assignable causes can be detected and measured, hence Ae called controlled variation, The variation due to chance causes is beyond the oon of human hand and they cannot be traced Separately, hence these are called uncon: variation 5.4, Assumptions of ANOVA - ‘The technique of ANOVA is based on Fes, forthe validity ofthe F-test in ANOV ‘he following assumptions are required, 1, The observations are independent, 2. The parent population from whi ich the sample observations are taken is normal 3. Various ‘treatments and envis ronmental effects are additive in nature, Let us suppose X,, Xp, vy Xq denote a random mean 0 and variance o ie, N(0, 2). Let us cone variables as 1 DXA where Q is « quadratic form inX), Xp, ...,X, with Tank (means df.) p, Td=1,2,..,k Then the random variables Qs, Qo, . are mutually independent ang tise chi-square variates with rdf, ifand only if nen, i 5.6. Gauss-Markoff Linear Model : of nis ‘and Markoff were developed a linear model fora set. vail 7498 + Yn With n! unknown parameters By, By, By a8 described ‘codim the matrix form yeABte 7 ivan nL vector we i nx coofficient matrix pismx1 unknown parameter vector rors cis n x1 random vector ofer5.7. is called fixed effects model. ANALYSIS OF VARIANCE 73 21 ay Con @22 Gam a ay Qin Qng nm |nxm a B= a ex dnxa Assumptions of the model 1, ©~N(,o7D ie, &y & ... . Tis an identity matrix of order ‘n’, 2 Ele)=0 ie, Ee) =0 V i=1,2,. 3. E(ee’)= 0%! . 0, Vi=j n ie, Elge)= {o ea which means o, Vi=j Cov (e;&)) = {6 Viej 4. Y~N(AB, oD) i.e., 1) Yay ++.) Yn are independent random variables with means and variances E(Y) = AB, D(Y) = o°I DCY) is known as dispension matrix contains both variances and covariances matrix of Y. ie, EQ) = air Bi + iz Be + --- + Gim Brn Var (y;) = 0%, i= 1,2,.. Cov (yy) = 0, #5 Under these assumptions, the linear model Y = Af + eis known as Gauss- Markoff linear model. ‘ixed Effects Model Let us consider Gauss-Markoff linear model y =AB +e with EQ) = AB, Do) = 0%! In this model, if all the parameters f,, By, ..., B,, are fixed constants, then the modelB.Sc. STATISTICS 1 74 oe San 5.8. Random Effects Model Lot us consider Gauss-Markoff linear model y =AB +e with EY) = AB, Do) = 0° In this model, if all the parameters B,, Bz, Bn are assigned by some probay; law, then the model is called random effects model. 5.9. ANOVA —One Way Classification Let us consider a random variable Y consists of N observations yj (i = 1, 2, ..., h 1, 2, ..., n;) is divided into k classes of sizes nj, no, ..., m4 respectively on the basis of so hk criterion such that Yn; = N. The one way classified data can be explained int A | following table. Class Values of Variables Totals | Means i yu Me yey Ty, i | 2 Yar Yor Yay oe Yang Te, Ye, vi ve Yin, T, yi G 7 Lt G =Grand Total ¥.,= Overall mean, Ifny=nz=...=n, k= Nn (say), then the dati equal number of observations, Otherwise, ‘a is called one way classified data wit! with unequal number ofebecry one the data is known as one way classified dat The total variation in th i a © observations y,, can be divided into the following t™ 1 ee between the classes (ie, treatments) This is due to assignablé ca Heh can be detected and controlled by human endeavour. Jausee whidh canna beet than hang, iAtion. This i de oct? ‘The sources of variation in the data are : 1. Effect of the treatment a, i= 1,2, ... p. 2. Error due to chance named ey follows a normal distribution. wa ;\ ereeerneee — mathematical Mode! In one way classification, the linoar mathematical model will be ww rw tey some =P +O Woey mA snsaeey . where Yo is the yield from the jth. ‘unit by applying ith treatment, (j= 1, 2, soy My i = 1,2, : pris general mean effect a . the basig <4) nal $3 daw “Plaine ig Naawe & =H ie, gis an amount of decrease or increase due to ith treatment. tyis error effect due to chance . * Note : Zaw=Zmo-w 2 = ny Nn=Nn-Nu=0 , Assumptions of the Model 1. All the observations (y,;) are independent. 2. Various effects are additive in nature. 3. gyareiid N03 fied data wit {Null Hypothesis classified dat® We have to test the equality of the population means ie., we are of di treatments. following t¥° + Null hypothesis becomes ve testing homogeneity to assignable sue to cnane Hoa) =ay =, Statistical analysis of the Model Let us consider ¥, = mean of the ith class76 = BSc. Stan ny mR x se nj j,, = Overall mean ifm ore. syd Lowey wi MAN We have to minimise the error ey so that the parameters p and a, ary Hence by using principle of least squares minimising the error (or) residual, Bad Sat islj=l kom =Y Y oy-n-a? as ‘The normal equations for estimating p and o, are aE . 2 5. =0 2) ¥ Oy-n-a)(-=0 on dslj=l kon k = LY LY wy-Nw- Lnia=0 ielj=l jel k => Ny,-Np=0 (: > no, = Fst Bey, ie, h=5, 0 > 2Y Gy-n-wn=0 jel 28 a = XY yy-nin—nio, = 0 iat = ny.-nin-nja; =0 => O=¥,-p o &y = Yy- W-H = yy Now the mathematical model becomes Jy = Wt Ot ey + Vi -F.) + Wy-Fwd_ . OO al _guaysis OF VARIANCE 7 Wr-In =Vi-F.+ yy Fi, Squaring on both side and wumming over ‘andj we gt y z Oy -y.P = zy z (Gi-F.) Hy -F OP mus hon “2B G3. wed z wu-5e+2 EH oy-F0U-F7) ietjet Since L Qy-¥.) = DL yy-niyi. yet ja = pa —niy.=0 3 So FP = x nGi-J. wed w-F.” TSS = SST +SSE where TSS =Total Sum of Squares = 5 Lov-7P Ain SST = Sum of Squares due to treatments i =X G.-¥P aA SSE = Sum of squares due to error kon, =X YL ow-7? faa jet Degrees of freedom of sum of squares 1. The total sum of squares (TSS) has N — 1. degrees of freedom (d.f.), one d.f, lost since the wneer constraint xs Oy-7.)=0 feet 2. The sum of the squares due to treatments (SST) carries k — 1 d.f. k Since Y njG,-7,)=0 A . The sum of the squares due to error (SSE) carfies N— K df. Since N quantities Y) (yy-¥i.)?= 0 of k linear constraints. intof 78 B.Sc. STATIS, Mean Sum of Squares (MSS) Mean sum of squares is sum of the squares 1. MSS for treatments is Soe 2. MSS for Error is 2 _ SSE St =N-K Expectation of Various Sum of Squares Consider the mathematical model as Jy TRH GTS. ‘Summing over j and dividing by n;, we get divided by its degrees of freed, Fi =Mt+ ate, Now summing over i and j and divide by N for equation (1), we get 1A & itm 1ehe hd Lwawtyd Dory dD Ley ae izljel i=lj=1 ‘ J. seen d mare, _ ry Ey ‘ E(&-)- Eni Fone, [nine 3 na=o int Since ey~iid. N(0, 0,2) ie, Eley) = 0, Var (eyJ= 0,2 v i,j a) Var (ey) = Bley~ Eley)]? =E(e,2) te =o E@) =0, EE)=0 tle. -&iis Var (@) = wld by | wi y Var (ey) del debjed spat =f eet Wo can easily observe the following formula ; hon igh bas Nie} dy ietjed From the above results, we can calculate iaaacad of various sum sof ‘squares as ‘allows : (1) Expectation of Sum of Squares due to Treatments (SST) ‘ ESST) als ngi-i3| in ‘ =ELD nidencica-iy| int (from equations (2), (3)) ‘ =E/D niast-er| fet b v als nlet-tremG-0| - int ist ‘ ‘ -a(3 nats z n@- areal niu &,-, | fer A s ‘ 2D navel near 0} n@B@,-é) in it int ‘ ‘ sy need wien in iB.Sc. STATIST¢y o Since EG) = 0, BE,)=0 Yr © 8 cay r hk k t and ¥ mG EP= Y mi (6h + El - ME) i=l isl k k 4 2D nen 2-2, D mk ms A h zi 2 oS meen @-wNE =D mee jel = ° k E(SST) = i= Ah na2+ Ym EGD-NECE) A oF? hk k o2 2S nate Eng Naw % njo2+ KoP-o? k (SST) = > nj 0,2+ (K-1) 6? ist ‘ 1 nya2+ 02 SST) _pig%y= B (5) -ESD=K- Under null hypothesis Hy: 01 = G2 = 0% =0, E(S}) =02 iie., Mean sum of squares due to treatments provides an unbiased the null hypothesis Hy. (2) Expectation of Sum of Squares due to Error (SSE) estimator for o/' E(SSE) -#|3 s oi] ieij=d iy B/D ins eey—tne +80] from equations 0) i=tjel re ; “PIX «-a]-8 yy bits #2043] feljer int jad ae ES aged aa intial int [3d Gs3 nat mi| qari OF VARIANCE pay int -»|3 XeeRndsE nc] eo A kon & “EDD -D mats DD wef) Y nBted) istjet ist oy y 2-3 pee feljed int MH E(SSE) =(N-K)o2 No?-Ko2=(N-K)o2 SSE * BNR] 262 ie, E(S)=02 ie., mean sum of squares due to error is an unbiased estimator of 02. +. Under the null hypothesis Hy: 0) = 02=... = 04 = 0, E(S}) #02, E(S})=0, ie. E(S;)=E(Sp), otherwise (Sq) > ES) Test a Statistic By using Cochran’s theorem, under null hypothesis Sr , SSP are two independent chi-square variates with K- 1 and N-K df, respectively. Hence by the definition of F- statistic, under Hy sst on K-D SSMh-1) : FE ——- Fann SBN) se rs a (N-k) 2 Fege Faun 7 Conclusion : If calculated value of F is greater than the significant (tabulated) value of F at specified level of significance and at (K - 1, N-K) df, we may reject the null hypothesis Hy. Otherwise we may accept Ho. ANOVA TABLE ‘The entire statistical analysis of the given data can be easily understand with the help simple presentation of the table is known as analysis of variance (ANOVA) table.e B.Sc. STATISTIcg ANOVA Table for One Way Classified Data Se Source of Sum of af. ‘Treatments Sst k-1 Error SSE Nek Total TSS N-1 Simplified Formula for calculations In the practical situations, we can use the following formula for easy of calculating kom Let G =Grandtotal= ) } yy i=1j=2 zo RSS = Raw Sum of Squares= ) yj, i=aja. Ge. WN *S called Correction Factor (C.F) TSS = RSS_cr a $g-spet ‘ SST = 2 NGF FD m+ D nz? = . ~2¥_ 2D nye Aa Tr Sst = Zor. and SSE = TSS - SST is 5.10. Critical Difference (C.D.) Ufthere is a significant difference in between the treatments (i.e., if Hy is rejected), thea we would be interested to find out which pair of treatments differ significantly. For this, instead of calculating student's ¢ for different pairs of treatment means, we calculate the least significant difference at the given level of significance. This least difference is known as the critical difference (C.D.). CD. at cc level of significance is given by CD. = (GB. of difference between two treatment means) x (tax for error df) For example Let Ho: 4)= 1) ie, two treatmensts means do not differ significantly. If Ho is rejected, then Hy is accepted ie., Hy : ni # 1) may be accepted. Now we have to determine which pair of treatment differ significantly. To obtain this we have to calculate a student ¢ statistic for every pair. So instead of this we compute C.D. as follows : yy) = C.D, = (SE. of difference of means) X ta for error d.f. Si provides an unbiased estimator for 6,? and if each treatment replicated » times ie, ny =n,i= 1,2, A, then CD. =Ss £ io © (ae X tag for error df. X tag for error df. S} = MSS due to error.source of Vartation | S.S ay aa Sun Pr — (Variation ratio) etween Process petyeatments) 7 |h-123-1) 9 ee ee Within process (error) | 51. 16 (N-k=19~3) Total 58 7 (N-1=19~1) Tabulated value of F at 5% level is Foon, k-1,N-&) = Figq,, 9, 16)) = 3.63 Fea = 1.098 < Fyay = 3.63 Hence Hy may be accepted at 5% level of significance ie,, outputs of three processes , Band C are equal. 11. ANOVA—Two Way Classification (with one observation per cell) Let us consider a random variable Y consists of N observations yj (i = 1, 2, ..., k; = 1,2, ..., ) is divided according to two factors viz., treatments and varieties. Let us uppose that N observations are divided into h classes (due to varieties) and each class onsists of k observations with the effect of k treatments such that N = kh. The values of he variable Y can be explained in the following k x h two-way table. Varieties >| 1 2 Ge RT Totals | Means Treatments ao 1 yn ye oa dy a Din aaaes|uaet at 2 Yar Yee ” 925 oo Yeh Te, Y2. y,, = Overall mean. ‘The source of variation in the data are 1. Effect due to the ith treatment 4, 2. Effect due to the jth variety fj 5. Error effect due to chance ey follows normal distribution,F ” B.Sc. STATs 86 —— ————— TC Mathematical Model / In the two way classification, the linear mathematical model will be ye w+ + By + ey i=1,2, 00% Jel Qewh ty by applying ith treatment whore yyis the yield from the jth varie to be normally distributed as N(y,, yyare independent and assumed (i 1 2 von pj My By vey AD jis general mean effect 1 koh siz Ra x w cy, is effect due to ith treatment = a.-B and 1 h whore w= 5 wy mi {is effect due to jth variety B =ny-P 14 where ny = j, DB im ‘Now we can write the following : eta arena wey Leepg h Dw mi mi k (or) x (or! k x we 1A (or) Dw in Note by ; , ores Xm = Zo wad wks kn-an ao hei x 8 =z, (y-W = hn- mp =0 Assumptions of the mode 1, All the observations (yy) are independent, 2. Various effects are additive in nature, 3. eyare iid. NO, 0.7) null hypothesis for the treatments and vacieties. Sepa" is We have to set UP their effects are homogencou® , ee ee | 4(4nat¥SiS OF VARIANCE (1) For treatments : sr Hy: All the treatment effects ime ie Hi ==... = y=0 do not differ significantly. (2) For varieties : Hy: All the varieties effects do not ie. Hy: Bi=Be=...=8,=0 Statistical Analysis of the Model Let us consider differ significantly. ¥., =Mean yield of ith treatment (or) (or) _ The parameters u, ,, B, are estimated by the principle of least squares for minimizing the error or residual sum of squares. kh E = Y Oy-n-a-BF ietj=t Differentiating w.r.t. p, a; and B;, we get the normal equations. es ko o> 2¥ LY Gy-2-o%-B)-)=0 Ai ho ‘ s = LD LD yw-Ne-h Y w-k Y B=0 delj=l set j=l ; s ” Ny_-Np=0 - EY %=0, Y B=0 i=l jel ~ B=y_dE A aq 79 7 2D Ou B- GB) (-1) =0 00, h => ry yy-hn—ho,- Be jel j=l = hy,—hp-haj=0 = Oj =¥i-B &=F.- 2 &=.-J- s Eg = 2Y, y-v-4- BDO OB; im , * = LD y-mn- ZF =O i=l a b PA apcheteo (° 2 5 B=Fs-B a =5y-B= Yar iB -h-&-B=y-F-- G-F.-Gi- Fd) Now the ae model becomes iy = n+ O +B + ey 7.4 Gi-F.) + Gy I.) + l9g-F_- Gi-F.)-Gyj- FI) -F, =Gi-F.)+Gj- I.) + Wy-Fe- Jt ¥) Sauaring and summing over i and j on both sides, we et EE wv i P= rd G.- 5983 G- 7943 Bou Fi-Fst IP mis i er ‘Since the cross-produet terms vanish i.e. zero ie, TSS =SST+SSV+SSE hk TSS = “2 Zo- 5. is total sum of squares SA k SST hd G.-Y.) is sum of squares due to treatments.-_ watysis OF VARIANCE n SSV = az Gy~ FP is sum of squares due to varieties on SSE = EE Oy-¥.-Fy+F.)in sum of squares due to error. Degrees of freedom of sum of squares nae TSS carries N — 1 or hk — 1 d.f. since subject to one linear constraint © > Oy- ¥. =0 mi * SST carries k — 1 d.f. since subject to one linear constraint ©) G,-7_.) =0 h SSV carries h—14.f. since Y G,-¥.)=0 ima afof SSE =df. of TSS ~ df. of SST — df. of SSE =hk—-1-(k-1)-(h-1) =h-Dk-D Mean of Sum of Squares 1. MSS for treatment is SST St 2. MSS for varieties is ss 2 Sv =A-a 3. MSS for error is SSE h — 1) Ck = 1) Expectation of Various Sum of Squares Consider the mathematical model as yy = H+ Oy + By ey CQ) Let ae & - 1 1 t. =a Vey or) ED it ms A A hE, = DE, AEH DE, om yaSumming over j = 1, 2, ..., h and dividing by h for equation (1), we get 1 if, 1¢ EL =nras, DBR 2% i= = Ye = N+ +E, Summing over i = 1, 2, ...% and dividing by h for equation (1), we get k 2 Lo te 15 a; + B+ Ly & ist Fy = B+ Be, Summing over i andj and divide with N for a (, k h bE Ewch ona d ond 2 mo, janet [: Bano d0-9) 7 y. = Bre, Since ey ~i.i.d. N (0, 0,2) ie. E (ej) =0, Var (ey) =o," > Var (€,;) = E (25) = 0,2 E@)=0, E€)=0, E€)=0 ae 1< Var (&,) = Var (3 o)-mE Var (e,) 1 92 = pp hoft=7-=E€?) 2 Similarly Var €,)=E(@?) = = 2 g2 Var @)=E@2)= $0 = 55 Based on these results, we can calculate expectation of various sum of squares ® ollows : D Expectation of Sum of Squares due to treatments (SST) a E(SST) =| 3 a-z|aia ysis OF VARIANCE 91 k [» L m+a+e,-(m veo [From equations (2), (4)] im ‘ -» [a X la &, isd ond ate ay wey] +an[a 3 LY a €,-=. | isa ima h Y €?-hke? Joa 3, Y o, (E,) - EE.) Zr] Since we have E €,)=0, E(€,)=0 and ‘ k k hY G@-EP =h [3 Ei + ke? - 28, > a| A mA fot E(SST) = " > Ms & + > Ms =|9, oe ene op X o2+ko2-02 mn ‘ E(SST) =h x a? + (k - 1) 6,2 > & st) - iz 3, ats 02 = ms k-1 « Under null hypothesis Hy: 0 = G2 = ... = 0% =0, E(S}) = 6,2 ae Mean Sum of squares due to treatments provide an unbiased estimator for 9? under null hypothesis Hy. © Expectation of Sum of Squares due to Varieties (SSV) As similar to the above, we can easily prove that h E(SSV) =k Y B2+(h-1) 02 gel ”os BSc. STATIST CS y, a SSV k ) -Bsto= pry L B97 jen 2 ag 2. Under the null; hypothesis Hy: By = Br= --- = Bs = 9 E(S4)=6/ (8) Expectation of sum of squares due to Error (SSE) ESSE) = E z Eo -F.- 505 ised =E 5 SB nrqe Bega r ate) O+ Bre) + sis getz=2 -2 [3 z by-8-2 +23] [From equations (2), (3), (4) pat ea _ - HE] YY 2 +82 +83 +82 —2e 58, — Wey Ey + Beye maj +2E;- -2; aa = YL we2+h z nep+4 3 Ee?) +hk EE?) eet j=t we (Sa 3 «|-2e (E25 «) eat j=2 a vee 3 a Sas [ead e]oefead el a 2 2 2 =hk ate hk + bh + a SE . a -28 2, ana]-ae [3 ca] +E (E_hke) + 2E bee he) 2E E hké)}-2E #hE)re OF VARIANCE assis h > Ee?) =hhol+ho2+ho2+o2—2h 5 E@?)- 2% A it = RO? +h 2+ ho? + 02-2hk% — bn = =hhoP—k o2—ho2+ o2 = (hk ~k-h+1)02=(h-(k-1) 62 E(SSE) = (h-D)(k-1) 62 SSE —SSE__| _yge = lacpara| =E(S§)= 07 je, the mean sum of squares due to error is an unbiased estimator of 0,2. Under Hz, E(S}) = E(Sj ), otherwise E ($3) > E(S%). Under Hy, E( Sj) = E(S§), otherwise E(S%) > E(S2). Test a Statistic (Variance Ratio) SST SSV SSE oe? of 884g? are independent chi-square variates with k — 1, h — 1 and (k- 1) (h-1) df. respectively. By the definition of F-statistic, under Hy and Hy, (1) For Treatment By using Cochran’s theorem, under null hypothesis (Hy and Hy), SSP fia -) SSE 02 Fr = ~ Fa-1,a-pade-») [a-axa-v SSTKk-1) __ St = SSEMh - 1Xk- 1) g2 (2) For Varieties SY fav = ~ Fa-1,a-pa- SSE fin - 1Xk-D _ssvah-) __ Sv > SSEMh - 1Xk-D~ g2 Conclusion level! “aleulated value of Fy or Fy is greater than the significant value of F at specified ro of Significance at (k — 1, (h ~ 1) (k= 1)) or (h— 1), (h ~ 1) (k— 1)) df. then we may “ect null hypothesis. Otherwise we may accept null hypothesis (Hr or Hy)4 Sc. STATISTI 94 B.Sc. Ics i ANOVA Table for Two-Way Classified Data tio Source of | Sum of afi Mean Sum of , Variation | Squares (Variance ratio) Treatments | SST ~ Fa-1-a OO _ 2 Varieties SSV # ~ Fa-1-va SSE E = 5 eS ror _ SSE | (h-DA-D/ S8= GSH aETD Total TSS hAk-1 ——__ Simplified Formulae for Calculations k A G =D > ois grand total ie1j=1 a ’ , CF. = N is correction factor N =hk koh Rss = YY y} is Raw sum of squares fei jet TSS = RSS-CF 1 k SST =], > 12 -cF im A SSV =), Pe -CF “1 SSE = TSS - SST - SSV PROBLEM 1. There are four doctors, innate io mediciDesign of Experiments 6.1. Introduction ; js essential to draw meaningful and valid conclusions { jstical study- Design of any experiment involv jon regarding the statistical hypothesis und study and making statistical analysis 0 . Planned and well-design experiments give us more accurate and fruitful results. For example, an experiment is conducted to examine the effect of three differe drugs administered to five patients. To conduct this experiment efficiently, first tt experimentor has to plan the major “concerns like selection of suitable and appropria patients which are relevant to the study, under which climatic conditions the drug has be administered, how much quantity of drug is to be applied, the better way of taking t! measurements etc. Proper care has to be taken to formulate statistical hypothesis and! obtain relevant information to study it. Eventually an efficient statistical analysis has! be made to obtain valid inferences. This type of planning is called design of ‘experiment! 6.2. Definition The design of experim experiment in which the degree well defined 6.3. Terminology in Experimental Design Experiment Experiment m ‘There are two kinds of experiments (1) Absolute experiments and (2) C s and () lompa planning, obtaining ent may be defined as the logical construction of of uncertainty with which the inference is drawn m8) eans getting an ani . eans getting an answer to a specified problem under consideratio” rat experiments.gS10N OF EXPERIMENTS. —_____ re ' qy Absolute Experiments Absolute oe. designed to determine the absolute value or precise value of some characteristi ly. Examples + . 4, Determining the average marks in a subject in a college. 2. Finding the correlation coefficient between income and expenditure of a group of ! selected families. (2) Comparative Experiments Comparative experiments are designed to compare the effect of two or more abjectives of some population characteristics. Examples : 1. Acomparative study of three fertilizers on a crop in different plots of a field. 2. A comparative analysis between two different drugs made by a company. 3, A comparative study of different kinds of cultivation processes. Treatment ‘Treatments are various objects of comparison applied on one or more experimental units in a comparative experiment. Examples : 1. If five fertilizers are applied on various plots in the agricultural land, then fertilizers are treatments. 2. If three cultivation processes are applied on various plots in the field of agriculture, then cultivation processes are called treatments. 3. If three kinds of drugs are applied to five patients to make a comparative study, then drugs are known as treatments. Experimental Unit The smallest division of the experimental material to which we apply the treatments and on which we make observations on the variable under study is termed as an xperimental unit. Examples : 1. In the agricultural field experimentation, the smallest division or part is a plot. ie., plot is an experimental unit. 2. In a medical experiment, the smallest division is patient, hence patient is an experimental unit. 3. The experiment relates with cows, cow is treated as an experimental unit. Blocks _ If the whole experimental material is divided into relatively homogeneous sub- Ups or strata, then these strata or sub-groups are known as blocks. imex#mple : In agricultural experiments, we divided the whole experimental material Telatively homogeneous sub-groups. These sub- groups are called blocks. melON B.Sc. STATISTICg 100_ —_— Yieta A The measurement of the variable under study on different experimenta} un known as yield. Examples : ; es 1. If an agricultural experiment relates to cultivating paddy, for example 7) paddy is obtained from a plot (experimental unit), then 70 kg of paddy jg te ‘as yield from the respective experimental unit i.e., plot. Experimental Error If we allocate the same treatments for all the experimental units of the homogeng experimental material, but the yields still vary from experimental unit to unit, 1 variation (apart from the variation due to known factors). which does not follow , systematic pattern and due to random factors beyond human control is known experimental error or error. It is also called Noise. Example : If we consider an agricultural experiment on which plots (experimen units) are homogeneous of equal shape or size in the field (experimental material) evey the same treatments are applied to all the plots, still the yield vary from plot to plot ¢ to the differences in soil fertility. Such variation from plot to plot due to random facto (soil fertility) is called an experimental error. Note : If the experimental error reduces to a great possible extent, then it leads to m accurate results of the experiment. Peactes The reciprocal of the variance of the sample mean is known as the amount: information of a design or the precision. If an experiment is repeated ‘r’ times, then 8 precision of the experiment is given by 1 lor where 6,” is error variance. Efficiency of a Design Let there are two designs I and II which are replicated r and r; times with the e™ variances &, and o,, respectively. Then the variance of the difference between t¥° treatmente meso the design and I are given respectively“ and 22 Now the efficiency of design I over II is ¥ , ° p= Ulta Ve, 2en/rs rie, iz, the efficiency of design I over Il is the ratio of the precisions of design 1 design Il : a_ DESIGN OF EXPERIMENTS. Unig, interpretation ‘f= 1, then both the designs I and I are equally efficient. exam, If E> 1, then design I is more efficient than design I. of pag a 70, IfE < 1, then design Iis less efficient than design II. "Stra! Note: The efficiency of the design can be increased by (@ Controlling the error variance (6,2). This can be done by dividing the material into small homogeneous blocks. f the h, (ii) Increasing the number of replications (r). unit to gests Uniformity Trials >eS not folloy, ® In field experiments, usually the fertility of the soil does not increase or decrease trol is kngv-*? uniformly in any direction but follows an erratic manner over the entire field. Uniformity “0 ¢ trials give us an idea about fertility variation in the field. Uniformity trials means a trial in which the field is divided into small units (plots) and the same treatment is applied on lots (experimen, each of the units and their yields are recorded. From these yields, we can draw a fertility 1 material) ey; contour map which is a graphical picture represents the variation of the soil fertility and m Plot to plot gy this helps in formulating good idea about the nature of variation of the soil fertility. The to random facie, tility counter map is drawn by joining the points of equal fertility through lines. According to this principle, the entire field can be divided into relatively . homogeneous subgroups or blocks with some shape and size of equal fertility (or more or 'n it leads to min jess fertility is alike) so that error can be reduced. 6.4. Importance of Design of Experiments 1. A proper design of an experiment leads to accurate results, this helps in achieving the objectives of the experiment. 2. Design of experiments is more useful in reducing the effects of errors due to various sources. 3. The design of an experiment increases the precision of the experiment. 4, The principles of design of experiments support in increasing the efficiency of the experiment. 5. The principle of design of experiments are useful in eliminating the personal bias in any form. s the amount ” times, then thy 2s with thee™ 6.5, Applications of De: in of Experiments ce betwee? 1. Design of experiments is extensively used in agriculture, green house studies, 2, etc, 20ey | 2. The design of experiments is applied to study the laboratory techniques, medical re experiments, pharmaceutical experiments, etc, 3. Design of experiments is used in the field of economics and industry. 6.6. Principles of an Experimental Design i According to R.A. Fisher, the basic principles of an experimental design are ’ ee or desis? ! 1. ReplicationB.Sc. ST: 102 2. Randomisation 3. Local control. 6.7. Replication Replication means rej executing the experiment more than once, petition of treatments more than once under investi (or) Replication is @ process so repeating the same treatment on various units under similar conditions. Replication is applied to average experimental units, Thus, the rep! D treatment is replicated ‘r’ times to the experimental units, out the influence of the chance factors on di lication results in more reliable estimates, then the standard error: o mean is = Vr Advantages 1. Replication is useful to reduce experimental error and thus it enables obtain more precise estimates of the treatment effects. 2. The large number of replications reduce the standard error and it increasing the precision. 3. One of the most important purpose of replication is to provide an estimate experimental error. Disadvantages 1. Increasing the number of replications is not possible since it involves more: 2. Replication of treatments may be sometimes subject to bias. 3. Large number of replications may lead to scarcity of the resources. Notes : 1. It is desirable to have as homogeneous as possible withi is homogeneity is not required between the replications, "i ¢ach FePl 2. The adequate number of replications to vari the experimental material, ous (reatments depends on te 3. The number of replications considered in the experimental should provide at degrees of freedom for the error. Hence usuall not less than 4, ly we should consider number of 6.8. Randomization ‘A process of allocating the treatments to various experim ar that ech and every experimental unit has an equal chenee ete ae treatments is called randomization. siving ‘The main objective of randomization is validity of the statistical tes ‘All the statistical tests (e.g. ¢-test, F-test, normal test ete.) Gaede ie a assumption that the observations and drawn randomly ie, independent.— (a an / gn OF EXPERIMENTS 101 nt experimentay nj, interpretation wu IfE= 1, then both the designs I and II are equally efficient. ' If E> 1, then design I is more efficient than design II. Pg of ape % k If B < 1, then design I is less efficient than design If. dy ig trea! Note : The efficiency of the design can be increased by : (@ Controlling the error variance (0,2). This can be done by dividing the material into small homogeneous blocks: ; (i) Increasing the number of replications (r). nits ofthe homa, ty Tria ental unit to yw Uniform ch does not folign In field experiments, usually the fertility of the soil does not increase or decrease n control is kngu, uniformly in any direction but follows an erratic manner over the entire field, Uniformity "¥® & trials give us an idea about fertility variation in the field. Uniformity trials means a trial " in which the field is divided into small units (plots) and the same treatment is applied on ich plots (experimeny each of the units and their yields are recorded. From these yields, we can draw a fertility nental material) eyeq; contour map which is a graphical picture represents the variation of the soil fertility and FY from plot to plot dy this helps in formulating good idea about the nature of variation of the soil fertility. The -due to random facia, tility counter map is drawn by joining the points of equal fertility through lines. According to this principle, the entire field can be divided into relatively . homogeneous subgroups or blocks with some shape and size of ‘equal fertility (or more or it then it leads to mm jess fertility is alike) so that error can be reduced 6.4. Importance of Design of Experiments wn as the amount é 1. A proper design of an experiment leads to accurate results, this helps in achieving the objectives of the experiment. ited ‘r’ times, then g the obj Design of experiments is more useful in reducing the effects of errors due to various sources. ‘The design of an experiment increases the precision of the experiment. The principles of design of experiments support in increasing the efficiency of the experiment. 5. The principle of design of experiments are useful in eliminating the personal bias in any form. 2 times with the *™ 6,5, Applications of Design of Experiments ference betwee? 1, Design of experiments is extensively used in agriculture, green house studies, ; 2. ete, 2a, . = 2. The design of experiments is applied to study the laboratory techniques, medical experiments, pharmaceutical experiments, etc, i 3. Design of experiments is used in the field of economics and industry. 66, Principles of an Experimental Design I oo! According to R.A. Fisher, the basic principles of an experimental design are : sions of desi? 1. Replication2. Randomisation 3. Local control. 6.7. Replication _ Replication means repetition of treatments more than once under investigatig executing the experiment more than once. (or) Replication is a process 80 repeating the same treatment on various Xperin ‘units under similar conditions, the ch hist Replication is applied to average out the influence of the chance factors on dig tal unite T lication results in more reliable estimates experimental units, Thus, the rep] 1 trestment is replicated ‘r’ times to the experimental units, then the standard error, 1. Replication is useful to reduce experimental error and thus it enables; obtain more precise estimates of the treatment effects. 2. The large number of replications reduce the standard error and it resul increasing the precision. 3. One of the most important purpose of replication is to provide an estimate o experimental error. Disadvantages 1. Increasing the number of replications is not possible since it involves more 2. Replication of treatments may be sometimes subject to bias. 3. Large number of replications may lead to scarcity of the resources. Notes 1. It is desirable to have as homo; iti vl ‘ geneous as possible withi th replication ty is not required between the replications. “S 2 The te icati : The sdeauate numberof replications to various treatments depends on the variabil! 3. The number of replications considered in the experimental should provide at Ie degrees of freedom for th not lees than 4. error. Hence usually we should consider number of replis** 6.8. Randomization A process of allocating the treatments to various experi h J in such #" that each and every experimental unit has an equal chase ncine any treatments is called randomization, The main objective of randomization is validity of the statistical test of signif All the statistical tests (c.g., t-test, F-test, normal test ete.) are based on the funda™ assumption that the observations and drawn randomly ie, independent.4 DESIGN OF EXPERIMENTS 103 Advantages 1. Randomization eliminates any kind of human bias. 2. It provides logical basis for conducting various statistical tests of significance. , 3. Itensures the experiment free from any systematic effects of environment. Disadvantages 1, Randomization is difficult to apply of large number of experimental units. 2. Randomization process can be effectively applied by statisticians only, Note : Randomization without replication is not sufficient to apply the various statistical tests * of significance. | 6.9. Local Control If the experimental material is heterogeneous and different treatments are applied randomly over the entire material. If the experimental error is increased due to some uncontrolled factors even the treatments are applied randomly in sufficient number of times. Then it is desirable to reduce the experimental error without increasing the | number of replications and without interfering the principle of randomness. This is achieved by principle of local control. ' Defi The process of reducing the experimental error by dividing the relatively heterogeneous experimental material into homogeneous blocks or subgroups is known as local control. For example, in agricultural experimentation, the heterogeneity is observed in various experimental units, i.e. plots. The experimental error is increased due to fertility gradient of soil (or fluctuations in the field) even if different fertilizers (treatments) have been applied randomly with sufficient replications (number of times). In addition to the principles of replication and randomization, the experimental error can further be reduced by dividing or grouping the agricultural experimental field into relatively homogeneous blocks than those widely spread, i.e., the entire field is divided into homogeneous blocks according to the fertility gradient of soil such that the variations With in each block is minimum and between the blocks is maximum. And then the treatments are allocated randomly with in each block. Advantages 1. Local control reduces the experimental error. 2. Local control ensures the experimental design is more efficient. 3. By reducing the experimental error to a great extent, we can detect even small differences between the treatments. Disadvantages 1. If the experimental material is homogeneous, then dividing into blocks is not necessary i.e, application of local control principle is not much useful : 2. In general, soil fertility does not follow any systematic pattern, hence dividing into homogeneous blocks is a difficult task,7 Completely Randomised Design (CRD) 7.1. Introduction The completely randomised design (CRD) is implest of all the designs and it based on the principles of replication and randomization. This design is suitable which the experimental material is more homogeneous. In this design treatments at allocated randomly to all the experimental units over the entire experimental materia Let us suppose that we have k treatments, ith treatment is being replicated n, time & i= 1,2, ...,k such that the total experimental material is ) n; = N. If all the treatment! ist are replicated equal number of times, then total experimental material is N = nk. 7.2. Definition If the treatments are allocated to all the experimental units randomly over the entire experimental material by using the principles of randomization and replication is called completely randomised design (CRD). 3. Advantages, Disadvantages and Applications Advantages 1. CRD is more flexible since any number of treatments can be used and various treatments can be used unequal number of times without complicating the statistical analysis. 2. Statistical analysis remains simple if some or all the observations for 29¥ treatment are rejected or lost or missing for some random accidental re#*?™5- Moreover the loss of information is smaller than any other design a AYr cg ETE BANDOMED DEBION (CRO) a “s RD results in the maxim: 4 2 perimental material can Seueuane of the experimental units since all the 4. ORD provides the maximum num! ber een error variance, which increasos th; of degrees of freedom for the estimation of © precision of the experiment, 1. CRD can be used only for homogeneous ex; tal i i possil always, Hence CRD in seldom used in field Cpa eneaitant Sonate 2, Since the randomization is not restricted in any direction, the variations among the experimental units is included in thy ‘ance whi i the experi \@ error variance which make the design 4, Since th» principle of local control is not used, it results in increase of experimental error. Applications CRD is most useful in laboratory techniques and methodological studies ¢.g., in physics, chemistry, ete, or in biological, chemical experiments or in green house studies, ete. 74, Statistical Analysis of CRD ‘The statistical analysis of CRD is similar to the ANOVA for one way classified data. The linear mathematical model becomes yysurarey, i=1,2,..,k J=1,2, 00% where yyis the yield from the jth unit by receiving the ith treatment. wis general mean effect ; is effect due to ith treatment tyis error effect due to chance and ey ~ iid N (0, 0,7) E N= > n,is total number of experimental units. = ‘his number of treatments ni8 number of replications of ith treatment. ko G= > > yyis grand total. sia T= y yy is total yield of the units receiving the ith treatment in i fo 5, = © ¥ yyin moan yield due to ith treatment. jelBSc. STATS, 106 a Sy yis overall mean. Then the variation can be split into ny k on s y LY Oy-FH= YL uGi-FF+ LL O- Fu” fajed in inijen TSS = SST +SSE where bow TSS = } Y oy-F_*is total sum of squares i jet k SST = ¥ n,@,- ¥_)*is sum of squares due to treatments. A bon SSE = DY (y-J.%is sum of squares due to error. mh Null Hypothesis Hy : = a)=......= 04 ie. all the treatment effects are same. (or) all the treatment effects do not differ significantly. Test a Statistic (or) Variance Ratio Under the null hypothesis Hy, F statistic becomes $2 Fre oF ~ Fau-an-w FE where S32 = St is MSS due to treatments SSE SE = yj i8 MSS due to error Conclusion ppc pelea of F is greater than the significant value (tabulated value? specified level of significance i.e., if F > Fy jected, Other sear ; k-1,N-4) then Hy may be rejected. ANOVA Table for CRD' . COMPLETELY RUNDOMBEDDESION(CRD) - 107 7.5. Least Square Estimates ‘The parameters p, 0; jdual sum of squares E i. in the mathematical modi » by using principle of le lel are estimated by minimising hon ast squares, Es) Leb=d ¥ oy- pap ielj=l ieij ‘The normal ecuees for estimating and a, are aE on™? ae Oy-n-o) 1) = a 5 x yy-Np- Sra 0 ~ Oa = Ny_-Np=0 Sn-y, ie, fay, ny dq,7° 7 2% Oy-n-awCD= => DX yy- ni w—nj oj =0 jet 3 nF,—mw—n,0;=0 = OH =Y,- aA. A a =y,.-H % =Fi-Y. & = yy B= Iy- 7. -Gi- I) =Ig-Yi. 7.6. Expectation of Sum of Squares We have ty ~ iid N (0, 0,7) He E(ey) = 0, EE) =0, BE) =0 Var (ey) = E(e§) = 9 0.2 Var €;) =EED=, of Var @,) =E@2)=7 > N= Sty ist a, ~N (0,07) ie, E(a,)=0, Var (a) = E(o,?) = 07>| BSe. STATIEn, ee | 108 er Consider linear mathematical model as yee + Oj + &y Vi SMH OY +E, (Refer ANOVA one way classification) hk x ni ¥, =p +e, (Refer ANOVA one way classification) (1) Expectation of Sum of Squares due to Treatments (SST) k E(SST) =E] > nG.-3.9| ina . : <8 [3 ngtenyt—29. 3 nz] inl * k -[3 ngtenst—a.ny.]-2 18 ny2-N ind =i ‘ E(SST) = Y n,EG@?2)-NEG?) eal Consider EG?) =E(n +0; +€,)” [from equation (2)) =E [p?+ 0,248? + 2p 0+ 2ne, + 20,8.) = p+ Ba?) + BE!) +24 B@)+ WEE) 4 way EE) 0? * EG?) =p? +0747 - i 2 y no | ae EG?) =E|p+—~y *® ttrdthequatio! & nPE (a?) Zz, —+E@2)+0- couPLETELY RANDOMISED DESIGN (CRD) 109 , from equation (4), we get Now, fe a L no? o2 i OF BOSD = Dm ves or4 Se) wy aa 2) =Np?+No2+ko2- Np?- ya o2 a a {14 Jersa-ves Us 2) 1 =EQ)=; 4 (2) Expectation of Sum of Squares due to error (SSE) Ay E(SSE) =E] YY oy-y." min =E z Board ng2-2 37S 9y ielj=1 i=. jaa =E y Svi+dn n5t-2 5m n3.| imi kom k -[3, 353-3 033] mip kon A E(SSE) = © Y EG2)- Y 21EG?) 25) msi A Consider E(Q?) =E((a+a;+e,)"] _ {from equation (1)] = p+ E(a,)+E(€})+0=p?+02+02 o2 EG? 2) ants of +e Now from cawation (5), 2 B@sE) => ¥ wt+o2+0%-E n i( +02+%) isij=t =Np?+No2+ No2-Ny_No?-ho? E(SSE) =(N—k)o2B.Sc. 110 = 2) G2 Nog) “ED =o; ie., MSS of error is an unbiased estimator of 6,2. SE e(F wing data to 20 animals and the obs i data using CRD.wo Randomised Block Design (RBD) 8.1. Definition ‘A mothod of dividing the heterogeneous experimental material into relative d the treatments are applied randomly homogeneous subgroups or blocks or strata an ' relatively homogencous experimental units within each block and replicated over all t blocks is known as Randomised Block Design (RBD). For example, in agricultural experimentation, the experimental area i.e. field is n homogeneous and the fertility gradient is only in one direction, then a simple method controlling the variability of the experimental material consists dividing 1! heterogeneous field into homogeneous subgroups or blocks perpendicular to the directiv of the fertility gradient. Then apply the treatments randomly to relatively homogencoi experimental units within each block and replicated over all blocks. 8.2. Advantages, Disadvantages and Applications Advantages 1. RBD provides more accurate results than CRD since the experimental mater is divided into blocks this results in decreasing error variance. 2. In RBD, no restrictions are placed on the number of treatments and 0 number of replications. But atleast two replications are required to test ! significance of treatments or blocks, Statistical analysis of RBD is simple and rapid. Statistical analysis is simple even if the data is missed for any number treatments, The number of replications provide enough degrees of freedom for error 8!” squares,g. The error may be increased i i 7: exist. if the interaction between the treatments and blocks 3. In every type of experiment blocking or grouping is not my RBD is most useful in the field experiments like agriculture, envi ete. , environmental studies 8.3. Layout of RBD The layout of RBD is good enough to explain through an example. In agricultural experiment, ifwe consider six fertilizers (treatments) are applied (replicated) four times. Then we divide the whole experimental area (field) is divided into four relatively homogeneous blocks and each block consists of six units, six fertilizers are to be applied randomly to all six units within each block so that a treatment should be replicated four times. Let the 6 treatments are A, B, C, D, E, F, then layout of RBD can be explained a Shae wo> ts mown To allocate treatments randomly, use any one of the methods, lottery method or random muenber tables method. For example, the selected units of the first block are 2, 3, Jot 2, B to plot 3, C to plot 4, D to plot 1, E to plot 4, 1,5, 6; then apply treatments A to pl 5, F to plot 6. gnilarly the units are selected for all blocks and then treatments are allocated accordingly. Notes: 1. InCRD, randomization within the units of each blocks in removed from vi 2. The number of experimental units 3. Ifwe want additional replications to more than one unit in a bl 4. The number of blocks are equal to number of replications, Each treatment occurs once and only once in each block. 8.4. Statistical Analysis of RBD ‘The statistical analysis of RBD is similar to the ANOVA is not restricted. But in RBD, treatments are allocated at random lock ie., randomisation is restricted. Hence the variation among ariation due to error. Thus RBD is more efficient than CRD. in each block are same as the number of treatments. for some treatments, then each of them must be applied for two way classified data.na B.C. STATIN, ‘The linear mathematical model becomes yy 2h +o + be i212. h Jed 2- where ith treatment: yyis the yield from th block by receiving 1 is general mean effect a is effect due to ith treatment 8, is effect due to jth block - > w is error effect due to chance and €y ~ iid N 0, 0’) = kh is the total number of experimental units. iq = number of treatments h = number of blocks or replicates.” ra G = LY Y yyis grand total aint i D T= Ly WL w int ma Then the total variation can be split into Bdseira§ card 6, 794% E oy- d, iy TSS = SST + SSB + SSE et ‘ re TSS = Total sum of squares Bp Foye ~7 iat jat SST = Sum of 8 " 'quares due to treatments 883 =Sum m of squares due to blocks be ts iS Ot squares due to error Ls. It 7?