You are on page 1of 101
& Chapter . 4 ay DESIGN OF EXPERIMEN \ 4g Introduction ‘An experiment is conducted with an Objective thesis. The sequence of steps taken to ensure OF to verify certain ae @ Scientific analysi . he ae ing to valid inferences about the hypothesis is called "Design of my ag ferences Potheicg experiment”. Tt had its origin from agricultural research and it is credited 5 test Of gre toprol. R.A. Fisher. one nomenclature in many concepts retain its rigia, For example, to verify the claim that a particular manure causes increase in the yield of paddy, we may conduct an agricultural experiment, i In this experiment the quantity of manure used and the quantity of Wyfe yeld are two variables involved directly. These variables are called aperimental variables. There may be other variables such as the fertility asoil, the amount of rainfall, the inherent quality of seed etc. which aso affect the yield. These are called extraneous variables as far as the hypothesis is concerned. i jecti\ is i is the extraneous The prime objective of design of experiment is to control i i a ly to the experimental $0 that the results could be attributed onl called factors- Factors may le principles of Desi three basic principle andomization plication control 1. Randomisation Random assignment of treatment to the experimental un hits j most effective way of eliminating any unknown bias in the , ll For example, if we select plots for use of manure at randon ton ¢ ; ee the are experimental units and application of manure jg treatment é en, plots-selected for not using the manure are called Control up, its control group. The control group provide a standard for conta iti it 2. Replication Replication is the process of repeating the same tre; more than one of the experimental units. Two identicalh (or units) will not give identical results. The differences to uncontrollable random causes. Such differences experimental errors. As the number of replication increase atment Y treated pia are attribug are cal this ery is reduced. So, replication is necessary to increase the accurag estimates of the treatment effects. 3. Local control Another way of controlling the effects of extrancous variabls i by employing the principle of local control. It consists of techigts of grouping, blocking and balancing of the experimental units. Grouping means combining sets of homogeneous a ito groups so that different groups may be subjected to reatments. Each group can have different number of exper rine ing means assigning the same number of plots or Pe a different groups, called blocks. The plots in ee sie ively similar or homogencous. We may use al 1a" different plots in a block. a] a” he ext" g means act of equalising total effect of the © oc fi jn the xP" the elements in the control group and in iI ts Ra og Epes Sy ; f 5 43 asi designs of experiments 1 Tim, 1 number of extrane bi Men, 4 nding OF the extraneous variaby an as r i os ies ating S in th yj De controlled, various design Procedures are Hee effects clop cd in the 1 zi a eth i eo at oY e 10 ¢ experimental design. We shall consider here th ‘ ‘hte important 1, Completely Randomised Design (CRD) », Randomised Block Design (RBD) 3, Latin Square Design (LSD) AM reg A i mi D ‘0 'denticaty teat t Completely Randomised Design ee Ic differences a0 ich differences a replication increases this Eo / tO increase the aca; Suppose there are 5 manures and 20 plots we shall give a random t allocation of treatments as below. Write the numbers of the plots 1, 2,3,.. 20 in identical cards and shuffle well. Name the manures or tieaments as A, B, C, D, E. jn a completely randomized design the treatments are given to the experimental units by a procedure of random allocation. It is used wien the units are homogencous. Choose 4 cards at random. The plots bearing these numbers may ‘egien manure A. Shuffle the remaining cards well and choose another “ards at random. The plots having these numbers may be shes Be B and so on. This is called completely randomied SS ANS design there is only one factor namely ‘treatment’ a Rendomised Block Design P'PPose we want to test the effect of r fe We divide the plots into f blocks, ach 7% ack the BIEOUS and each block contains 7 plots: Within CT cos) are Ma eted at random and the r eatmen eer This is in cach block only one plot rece a * andomised blo OF all the f blocks, This design i ) ze -3) i 2 Total sum of squares SST_ = > Dey os) qt Wt SSW + SSB Sof * There are r samples, so the number of degree freedom=r — 1 ing degress If these sum of squares is divided by the corresponding lom, we get the mean sum of squares. fhe computations : r) he entire set hte f variance © there are several steps involved in thin" iances, " and "within" sample varian ca into a compact table called analysis o We ge Tey “hy y — below. “ : By ~ a table way classification ~ ANOVA table one" Mean square (MS) Variance ratio | MSB [ toal_[_ SST Since value of F should be greater than 1, we take the larger & S= > G HP | MSBor MSW to the numerator. J Conclusion : If the calculated value of F < the table value of = then H, is accepted at 5% level of significance. Short cut-formulae To construct the ANOVA table we ado} Procedure, Se 1: Find the total number of observations N S92: Find the total value of all the observations pt the followin —— 3: Find the correction factor r os Probability ang ee Step 6 : ssw = ss _ sop Ny Step 7 : ‘, MSB . Find F=77eq if MSB>MSW or F=MSW Msp MSW Step 8 : Find the table value of F for CN SHY dtp F for (N—r, r—1) df at 5% level of significance Conclusion : If the computed value of F < the table . i Value og We accept Hp, otherwise reject Hy y Example 1 : As head of a department of a consumers’ esean organisation you have the responsibility of testing and ‘comparing jg times of four brands of electric bulbs. Suppose you test the lieting of three electric bulbs each of 4 brands, the data is given below, aq entry representing the lifetime of an electric bulb, measured in hundres of hours. Brand 2 ands only one factor-lifetime of bulbs. The “ yant to test mean life times. Null hypo! re ed! i ar 5 = My = Mg (ie. population means N= total number of observations 4x3 = 12 7 = sum of all the values Se You test te oT= 60 + 69 + 6 + 63 = 258 ata is given belo (258) D, measured sso " Correction factor = Pr SST = 24-3 egerg+zdt+td-F — 5547 1202 + 1595 + 1460 + 1529 = 5586 — $547 = 39 cal 2 ew _F Ny 2 2 paces, 2, ny ny 5547 Probability a dS Ww La SST ~ ssp +3515 = 24 We 4 hall now form the ANOVA Table. S. oe of | Sum of | Degrees of | Mean ariation | squares SS | freedom df | square MS Between | SSB = 15 | 4-1 = 3 samples Within ssw = 24 |N-1r=12-4 samples ag (or error) Total SST_ = 39. Wu From the F table, F (3, 8) at 5% level = 4.07 the calculated value of F < table value of F Hence Hy is accepted. So, the mean lifetime of all the brands ¢ bulbs are equal. le 2 ; Four machines A, B, C, D are used to produce a cert ze 4 with each unit as 100 squat of the machines at random al founted, with Examp! kind of cotton fabric. Samples of si metres are selected from the outputs the number of fl jaws in each 100 square metres are ¢! Bxpenments in O pst. yf 4.11 ‘ think that there is a significant difference in the perf machines ? ormanice 0 of oe four solution : Here only one factor is involved, namely performa : nce. we want to test with 4 samples for cach. The null hypothesis is qi tam He (ie.) the machines do not differ significantly in their performance. not all are equal in performance. = 4x4 = 16 + 28 + 53 + 90 = 21 uy? 5 44521 = 2782.56 16 16 2 eubeae Te sxazdreared N — 2782.56 410 + 216 + 745 + mass = 7 3409 — 2782.56 626.44 tl 2 pasion, Cx,7 ny Ry 2 2 2 2 40) (28) 53} (90) - COE, C8? , 6, OO? ao = 400 + 196 + 702.25 + 2025 — 2782.56 = 3323.25 — 2782.56 = 540.69 “. SSW = SST - SSB = 626.44 — 540.69 = 85.75 We shall now form the ANOVA table. _ The table value of F (3,12) at 5% = 3.49 and at 1% = 5.95 ‘ ie! calculated value of F > the table value of F ted ee ines differ in their performance 9 t exper pes . 3 . a company wishes to test whether ; its three 3: po tend 10 make sales of the same size op es salesmen ‘Ue velit ability as measured by the avernge her they fitter i : : ¢ last week, out of 14 sales, A made 5 os Of their sales, °, B made 4 and ¢ pore i the weekl; ‘i following are ekly sales (in R s, Th s. thousand) recorg = salesmen. oot the three whether the three salesmen’s average sales differ in size. : Coding of data In this problem the values of the samples are big and 50 the uation would become tedious. So, we shall use the method of wting of data to reduce the magnitude of the sample values. ps 4 fied number to all observations, subtract @ fixed a = / 4 multiply all by a fixed number or divide all by a fied me ‘ data is known as coded data. By coding, the ® d «lution : We shall assume the null wrt in the average sales of the 3 salesme? Here N= 14; T = 15 + 16 + 25 = 56 2 ; Correction factor a = or = 224 j a Ey tPQt2Q-y SsT = = 59+ 70 + 135 - 224 = 264-224 = 40 : 2. 2 : Gx Cx’ Cx Pr = tt oN Sigs ta, ap oOAN A 2, f = sy? , 6? , 2" _ 994 ) Biase tanga? = 45 + 64+ 125 — 224 = 34-224 = 10 Mean Vip aquare MA va msp=2! | 5 2 ~The table value of F(2, 11) at 5% level = 3.98 | The calculated value of F < the table value of F. significant Ea _ Hence Hy is accepted. That is there is no t differ senical. ‘heir average sales. So, their selling ability do not le for 2 completely 4: A particular completed ANOVA tab! design is shown next : square Source | ar | Sum of squares = a eae) or Sompwe the ANOVA table Ae k E panne \ a8 How many treatments are savolved i Ih Source a # squares Treatment 24.7 Error 62.7 = 24.7 | 3 = 38 Total ' 62.7 : ; ince the number of degrees of freedom for r tre; r-1=43r=5 ae ~. Number of treatments = 5 Example 5 : There are three main brands of a certain Powder, i . fiiienteeninetcfecaminetiana (sara to be atc om four groups (A, B, C and D) and a i )) and three brands (I, II, II) as shan, Is there any significant difference in brands preferenct 2 anor 5% level. pu Note There is a mistak 120. We shall solve with 110 as total sal nce. it brands a @ eh is th e in the data because the til® mples ¥ ues. n : We want to test the differe hall use one-way ANOVA. Hp : #1 diffe in the brands preference. east one is different from Gihey, : al T = 27 + 32 + S51 = 110 . 2. Cart factor T = ey = 1008.33 “Ss Di Soke bee zl = 305 + 294 + 715 — 1008.33 1314 — 1008.33 = 305.67 Probability ang ie ivi — 1008.33 = 1088.5 — 1008.33 = 80.17 * SSW = SST-SSB = 305.67 = 80.17 = 225.5 ANOVA Table Source of | Sum of Mean square variations | squares SS SSB = 80.17 SSW = 225.5 SST =305.67 The table value of F (2,9) at 5% level = 4.26 The calculated value of F < the table value of F +. Hp is accepted. Hence there is significant difference in te preference of brands. Example 6 : As part of the investigation of the collapse of a of a building, a testing laboratory is given all the available bolts onnected the steel structure at three different positions 0” ks, “ he forces required to shear each of these bolts (Coded values q =m 5 iments a ay i LEE “uy, %, foe act, 419 ; an analysis 0} lance to test at the for™ 0.05 er the differences among the sample ae ale of Sigpiticang at th . re significant: ae Positions . ‘ i ‘< ‘ A.U-2093 Hy, > Ha = Hy (Le) the means ied 3, 2005} ion PH © colution * 0 si he 3 position ye same: _ pot all means same. wH, : not We shall reduce the values subtracting 90 from each of the vab ues, The coded data is Position 2 | Position 3). x2 3 oy 2 2 15 =7 0 25 9 1 100 16 6+7+4= 1 = 117+31-14 = 9 ee W = 299 + 473 + 166 — 0 = 938 2 a a N SSB = ae = on a, 14? 6 a 4 = 48.167 + 137.286 + 49 = 234.453 «SSW = SST—SsB = 938 — 234.453 = 703.547 ANOVA Table value of F(2, 14) at 5% level = 3.74 ated value of F < the table value of F. i ce ed. So there is no significant differen 3 positions. Re eriment 0 ducing by esi es pros i bolts, 4 botts t the ain, pi at random and found Oh Produceg b rence we conclude that there jg ie, lated. MY gs OE hy in production 5. Sleniti a flog, Me cnines i” Pl 50 far as the 4 ant trea 0 dlarmetery ** bage Machines the Dro» _ Solution : Hy = Hy = My = Hy = wy =p, = M6 ie.) We assume that there is no difference between the machines with regard to diameters. H, : not all are equal. subtracting Since the values are big, we shall form the coded data § from each value. Probability and Sia, Sty 4x6 = 24 sek Ak Sit one a? _ ou . BB = 160.167 apse SRR Se Sepa ya 04 50.16 2 Bl got = Exit 2A te + 2% 7 ON See yee to gs > 7. 160.167 = 412 — 160.167 = 251.833 2. 2 poo Oe iT ny ig -N =0 @ 8,825.7 - 160.167 = 49 + 16 + 306.25 + 2.25 + 1 — 160.167 = 3745 — 160.167 = 214.333 = SST — SSB 251.833 — 214.333 = 37.5 ANOVA Table ‘ yf Even po pt Z 16 Fee ble vane of FS, 18) at 5% Je, $33 “Igy 1% level the table value of p — 4 a R = 425 ‘The calculated value of F > the table vat, Ie of p is rejected. Infact, the difference jg highly k i si +9 ye cannot consider that the machines produce i io = 3 Olts of s, 16.16 ajmeter- oe 44 Two-way classification (RBD) In two factor analysis of variance we consider one classifi slong column-Wise and the other row-wise. For example, poninp crop in several plots of land may be classified according to aoe yaieties of seeds and different varieties of fertilizers. So, seeds et fedilisers are the two factors. Let the N values {xij} represent the yield according to the to fadors. Let there be r rows (or blocks) representing one factor of dassification (say different varieties of seeds) and c columns representing tle other factor (say different fertilisers) so that N =~. We wish to tet the null hypothesis that there is no difference i= yield between ‘Taious rows and between various columns. gsr, SSE The total variation SST consists of three pars Ss, SSC = Sum of squares between columns SSR Sum of squares between rows 8 canal (or atte SE = sum of squares for the residual (0 gsc - SSR W Paks find SSE using others (ie) SS* ag rod 10 » of variations : testi ng Iny -< the meas in ¥©-Way classification ‘residual’ is ? " ynitd tor Of differences. It represen ; Ces called ‘chance’. Probability ang i ay Ti ae Nt he two-way classification ANOVA table is given jy, : low, Source of |” Sum of : Variation |squares ss} irs Between columns Between rows Residual (Error) F, and Fp should be calculated in such a way that Fo > 1 and Fp > 1 as in the case of one-way classification, If calculated value of F < the table value of F, then His accepted, otherwise rejected and the conclusions is made. cut formulae as in one-way analysis. 1: The following data represent the number of units per day turned out by different workers using 4 differs i tS emer en, 4s y five men differ with respect ty , > me the her ce ee, set re crether the mean Productivity is the sim, a _ hy i es. i Me Soy pt machine ‘YP : 1A-U-2006, 2019, 2944 ie . The null hypothesis ; the workers do not differ with respect tg Productivity ang AWity and Th be préductivity of the machines are same there is significant difference in productivity of workers ju * ‘i eT5 01 the machines. since the values are large, we shall code the data. We shall subtract 4 from each value. The coded data is given here. ne-way classification Machines ble value of F, tha ii Bejsp [ec | D | elusions is made. % [% | | |) oF at fe |% yi I esieaeii=2 fe. | -4 [5 [36] 4 | ® = | 0 [14| 9 ojo) mi| 3s | a|% | a] a ] Pmieorj4:| 4c) -8 |-14| * Pte 71% = =7)|2.05| 2 } M% 3 De| 26 | +47 a | 1| Peeiberri@2° | 9-|-1| § 5 eT | | Tor 5} <6 | 38 | 17 joi | a ca Pee i) N= 29 os T= 2 , Correction factor N 20° = 20 at 2 SST = Sxt+ExeEe rag - t = 101 + 28 + 326 + 139 — 20 = 594—20 = S74 seu Ss, Ea Cyt Cx). 90 5 ., a) Bee eon 38s | (217)7 = oa 5 ae 5 =-20 = 54+7.2 + 288.8 + 57.8 — 20 = 338.8 Gy)? . ey? Gy) 2 Sh Geeeqem et go #22 (=u? & ce Seryep aes Sa 7m ay tang tts 25+ 441+ 196 +64 _ 9g ms 8 _ 2 = 181.520 = 161.5 SST - SSR ~ SSC 574 — 161.5 — 338.8 4 — 5003 v iments, \ et pe i Two-way ANOVA Table y. Sum of | Mean Square Mg squares SS (- 1-1) nag (ee 12 The table value for Fe (3, 12) at 5% level = 3.49 The table value for Fp (4, 12) at 5% level = 3.26 In both cases the calculated value of F > the table value of F + the null hypothesis Hy is rejected. ctivity of Hence there is significant difference between the productivity men as well as machines. é given data Pample 2 : Perform two-way analysis of variance (°F the 4.28 Probal ily ana t ss Solution : H, : There is no significant difference between cotymy means (i.e. treatments) and between row meang (i ; * Dog H, : There is significant difference between column, Ley Means oy the row means. We shall form the coded data by subtracting 40 from each Vale Treatment Total Se Braet | tan |X, otal 52] 2 | 2 eae ie, || 21 8 al y J=2] 0 | t[-1)-2] 4 Posi cee mame ge |e 4:| 12 | 25.) 4.) erbg pees 22 ola Total of | 3 0 12 ‘column. BN =3x4= 12; 7 12 122 the table ern rim — : bs rejected. So, the dif = erence is high) eae difference b s Peon etWeen the : e Bh regard to cleanness, meee cc hee th eo : ot 5; Four different, though supposedly we : ly equiy; gomtl reading achievement test were given too, ent "MS of s 0 each of § at following are the scores which they obtaineg, Nt Br (es s50 el nR | 56 61 53 67 62 Y Slenitcant Hep ce Perform a two-way analysis of variance to test at the level of significance 0.01 whether it is reasonable to treat the 4 forms as equivalent, [A.U-2004] Solution + the student’s achievement are same and the forms are equivalent. the student’s achievements are different or the forms are not equivalent. To simplify the computations, we shi iting 70 from each value all form the coded data by 436 Probability ang alia N =20;T = 69 . ie 69? Nt op 7 23805 2 SST = Exe + Ede Eh t+ Exp + Bd - q % 459 + 103 + 670 + 86 + 1629 — 238.05 = 2947 — 238.05 = 2708.85 @ne Cy) G4) C1) c/ ssc = 4+ 4 + 1 OP 4 4 4 4 4 Y 2 72 592 2 342.25 + 12.25 + 625 + 25 + 1560.25 — 238.05 I = 2564.75 — 238.05 = 2326.7 ey? ey) Sr” ey? 2 SSR = 5 e. 5 a 5 oF 5 N 10% 4232 0%: 262 eet oe = 20+ 105.8 + 20 + 135.2 — 238.05 = 281 — 238.05 = 42.95 : SSE = SST — SSC - SSR = 2708.85 ~ 2326.7 — 42.95 = 2708.85 — 2369.65 = 339.2 Of. Experiments psig 4.37 ANOVA Table Sum of oa squares Mean squares| Y@tiance N SSC= 2326.7 eo Msc = 2326.7 238 4 5 = 581.675 SSR = 42.95 Msn = 4295 3 14.32 339.2 2 28.26 5.41 " The table value of F,, (4, 12) at 1% level 27.05 The table value of Fp (12, 3) at 1% level | Since the calculated value of Fe > table value of F at 1% level ihere is significant difference in the student's achievement levels Since the calculated value of Fp < table value of F at 1% leveh there is no significant difference between the forms ‘is made in 5 Fxample 6 : The following are the number of mistakes J oe ang for a photographic laboratory ive days by 4 technicians working Pee meeey st at a level of significance « = oi. Test whether ° shance. the four sample means can be attribated (9 C [aU 2011] Probability ang Here x only one factor is involved, th - the num b made by chnici Y 4 technicians (4 samples) Of ny ~ one way classification. 3 We we | ‘ant to test whether there j betes ere is any difference m the Mag ies ] bee PCE’ iy 1 tees td Hy Not all cauy al Total N_ = Total number of observations = 4x5 T = Sum of all the values = 49 + 59 +55 +50 = 213 2 Correction factor = - = ex = 2268.45 id gst = EG+EG+EG+E4- = 517 + 717 + 639 + 510 — 2268.45 = 7393 — 2268.45 = 114.55 : P ? 2 xy Ex)? ex! _F 3 OP at pe N _ 1 jay? + 507 + 552 4 50} — 95, 5 2268.45 = + [2401 + 3481 + 3025 a 25 + 2500) — 2349 4 = 2281.4 - 2268.45 = 12.95 ssW= SST - SSB = 114.55 ~ 1795 _ 1016 ANOVA TABLE Sum of Mean square df squares Ms SSB = 12.95 Ree ples ia = 4317 SSW = 1016 | 16 |yygy . 1018 ~ 16 = 6.35 Total SST = 114.95 19 The table value of F (16,3) at 5% level is 8694 the calculated value of F < the table value of F * Hy is accepted at 5% level of significance. So there is no significant difference between the 3 Three factor classification (or) Latin Square Design (LSD) ia ~— have seen data from a Latin squat types of Sac (roy) Slassification say (i) variety of i. treatment, So AI : “lay M4 (itl) the letters for different ma" 88 below. results in a of Squares ~ [sso n= 7 (FR = Mey SSK n= IMsk = SSK HS 1,4 |Fk = Mee Ms (o—1y(n-2) In? —1 There is no difference between columns, between treatments. not all are equal. Example 1: An agricultural experiment on the latin square dei gave the following results for the yield of wheat per acre, the kite corresponding to varieties, columns to treatments and rows to bas Discuss the variation of yield with each of these factors betwee? there is no difference between rows columns and between varieties. iments of Bape 2 ie shall form the coded data by Subtracting 19 f, rom ac Total | Ssr =Stersesdergeed-o = 14 + 22 + 30 + 23 + 86 — 38-44 * 235 ~ 38,44 = 196,56 2 ex! ae : 2 (ha) oe ee, ea, Sere 5 N oi 2) — 38.44 Pe get! 38. i = 23 — 38.44 = 105 — 38.44 ey) Sy? | Ey) @yy? yy? SSR 5 See hor sme R a (2+ P4974 F 45% — 38.44 ! = 7B 39.44 = 406-3844 = 216 To find SSK : We arrange the data according to the letters row wise Total han 222 gees? (19? 7 SSK = it “sweet Ot 5 N = 805 5 — 38.44 161-3844 = 12256 = SST - SSC ~ SSR - ssk 196.56 ~ 66.56 — 2.16 — 122.56 ANOVA Table 4.43 Sum of squares SS SSC = 66.56 ah Between columns : Fo = 38.82 The table value of F (4, 12) at 5% level = 326 + caleulated value of Fo > the table value of Fc * the difference between treatments are - ast yie : on $0, the different treatments have effect etveen rows : Fae 123 Me. = 306 ‘able value of F (4, 2 at 5% Tew! ave oF v * te calculated value of Ree the table oe en blocks a -s © is no difference betwe 444 Probably ang Between letters (or varieties) : oy Fx = 6964 The table value of F (4, 12) at 5% level = 3.26 *. calculated value of Fx > the table value of Fy Hence the difference between letters or varicties ig highly iit m1 Example 2 : A farmer wishes to test the effects of four ai fertilizers A, B, C, D on the yield of wheat. In order 10 ang Sources of error due to variability in soil fertility, he uses thy. in a Latin square arrangement as indicated with following tale the numbers indicate yields in bushels per unit area, perform an analysis of variance to determine if there is signiticul difference between the fertilizers at « = 0.05 levels of a [A.U-2006, 2007, een. Solution : H, : there is no difference between rows eww : columns and between treatments. H,: not all equal. : : alu. We shall code the data by subtracting 15 from cach ¥ ee = 107 + 106 + 189 + 117 — 189.06 = 519 ~ 189,06 = 30994 %. SH? ex)? Ex)? Gu fo oa £2) + 2 a ak Pe SSR = 873 | | 4 ~ 189.06 = 218.25 — 189.06 = 29.19 To find SSK, we arrange the data according to the letters TOW sg ear 473.25 — 189.06 = 284.19 = SST — SSC — SSR — SSK 329.94 — 4.69 — 29.19 — 284.19 329.94 — 318.07 = 11.87 my squares ISSC = 4.69 Hetween columns : F = 1.27 The table value for F (6, 3) at 5% level = 8:94 * calculated value of F < table value of F So, there is no significant difference between columns so far as otlty is concerned. Tons : Fa = 4.92 The table value of F (3, 6) at 5% level = 476 5 Z evel * Calcilated value of F > table value of F at 3% © ‘ “ row t0 Fo. “there is significant difference in fertility from Weatments : Fy = 47.89 = 476 le value of F (3, 6) at 5% level . the calculated value of F > table value of p, Hence there is significant difference between the eit eer, Example 3 ; Analyse the variance in the latin square of of paddy where P, Q, R, S denote the different methoas Held ig ‘y) Of cUltag R 123 7 P 121 $ 122 Q 12 Q 124] R 123 | P 122 | s 125, P 120 | Q 119 | S 120 | R 121 S 123 | Q 121 P 122 Examine whether the different methods of cultivation have 9 significantly different yields. [A.U-2006, 2009, 2014, ais Solution : There is no significant difference between Tows, between columns and between the methods of cultivation. Not all equal. We shall use the method of coding to reduce the valuts T2 N é 302 16 ~ 56.25 2 2 q Bap Sa 24+ x2 = N = A+ 20+ 14+ 34 _ 694 = 92— 56.25 = 35.75 (2x, (x,)? Gx) ; ek ee oP N ee) Beoue 16-10 Hila tea 4: 7.56.25 1 = 7 736) ~ 56.25 = 59~ 5625 = 275 En? Ey ey)? ey? 72 SSR = ata tt act oe 8 142 92 igh gat 0+ = 56.25 fe 424) — 56.25 = 81 - 56.25 = 24.75 i ow wise. To find SSK, we arrange the data according to the letters 92 Hi SSK = 3 qadts 72 4 N = $ (242) = 56.25 = 60.5 — 56.25 = 425 + SSE =" ssp _ SSC — SSR — ssk = 35.75 — 31.75 = 4 ANOVA Table Source of Sum of Mean square oe df variation Squares SS MS SSC = 275 | 3 275 lcolumns. Leiter = 0.917 ‘Between rows ISSR = 24.75 3 =" 24.75 Between letters /SSK = 4.25 a MSK (or methods) SSE = 4 SST = 35.75 Between columns : Bees 137, # = a6 The table value of F (3, 6) at 5% level is Fys = 47 ah =<" F) 05 ; : there #® Since the calculated value of FQ < table value of Fo Significant difference between columns of plots in yield. 452 Probabitiyy oh Solution : : e Hy there no difference between columns, between r between treatments, eq Hy : there is significant difference in atleast ong We shall use the method of coding to teduce the Valu es, Subtract 12 from each value 2 ean Seeman Pele 273 AY Desiraunits a Pel Mme ees | to lo |. ¢ 6 hy DY FACS peel ee Peale cc calomel 9 | a6 | a | ‘ fle veal « lial Bags ae eae ge (0) ate Cc B A D wla1lola Pe tee ilo gsm | 16 Total} 10 | 12 | 13 | 9 i#]2[lsie Tape eos at a = : Te SST = Sx tEGtTg+Iy-_ B= 52 78'+ 75 + 73 — 121 = 278-121 = 157 2 ey? ? ex? ex) ee ee ee Zz = 121 = (494) 2 aS 2135-121 = 25 ey, Sr @y)? ayy Regie ti 4 gt + 477121 flee eo 11? at gt 1 = 7 (498) — 121 TAS— 121 = 3.5 Tofind SSK, we arrange the data according the letters, row wise Total KF (4? 13 ww 7 th em N ed. 4 (1062) — 121 * %5S— 191 = 1445 Se © SST ~ gsc — ssp - SSK “Siaa505 = 65 Source of Sum of varieties | squares | “| Mean square mg SSC = 25 SSR = 3.5 SSK SSE Between columns : The table value for F (6, 3) at 5% level = 8.94 Since the calculated value < the table value, there is no significa difference between the columns Between rows: Fp = 1.08 The table value of F (3, 6) at 5% level = 4.76 Since the calculated value of Fa < the table value of Fy is no significant difference between rows. Between letters or treatments : Fy = 446 table value of F (3, 6) at 5% level is = 4.76 : +. calculated value of Fy > the table value of Fx. ie significant Hence the difference between treatments is sigeif ani < "i Foy . ween rows Since there is no significant difference between icates columns but highly significant between treatments 1" ‘Square arrangement has not been advantageous. ve \ pesigtt of Experiments prample 5: Three y, plock design with four ro Plicat “3 gs below. og a A6 C5 4 al iz the table Value of FR so, there is significant difference between the 4 fertilisers, Between the letters (or varieties of paddy) Fx = 202 The table value F (3, 6) at 5% level = 4.76 * the calculated value of Fy < the table value of Fy * there is no significant difference between the varieties of paddy 4 Yield, 4.02 _ 2 Probabitey ong Solution ; Ho : There is no difference between rows, betw between fertilizers a 1, : — There is difference in atleast one of them We shall form the coded data by subtracting 25 from Cach o} -7] 2] 5 4 a. [pes HRM ESS ~ 26 | 1 JU laeelee eras tel eentne pees pts ‘ (ae lA |e | D | Sts | Soe ea |e Bile =z lane | Sa he | | SIE. ICMEIAC eit ID 4 Filia |cctitiets| (oll edo? | flog ] Dies |Ea Baa Ce OR | 4“ | ale io] 2lecapr sli? So Ss N =5x5=25,T = 18 ’ Brig 118% Correction factor = N Deas 12.96 aia SST = Bx +3G+EY+ svt Ee- = 103 + 95 + 113 + 84 + 101 — 12.96 = 496 - 12.96 = 483.04 : i Sg =5(2 + (Bx)? + x, + @x,)*+ ex] a) n = ements am a * be, i & f “ee ‘ a Adres ic Son p+ (-3)8 + 7 + 2? + 5° é 1 [ 12,96 tthe thy ae Stig inc re \ } ~ 8 25 5 136 _ 12.96 = 27.2- 12.96 = 1424 to, 25 i ~ : x Li sy,)? + Qe)? + ys)? + (Ly? 4) 7) 72 ° es e “SA, ue x 2] r= 5[ y ar 0 ‘ ae ‘ 22+ +9+7 +3] ~ 1295 eS 2 4 3 | 3 a = 16- 12.96 = 3.04 1 to find SSK, we arrange the data according to the letters, row wise 9 9 | 1 lel os | 4 9/8 a4 | 11 Pr SK = 3 [(-30)24 (7)? + 132+ 092+ her 2 $1900 +49 + 169 + 1156 + 64] — 12.96 * 467.6 — 12.96 = 454.64 = SST ~ ssc — ssp - ssK : os 483.04 — sa04 = 3.04 — 454.64 ~ 483.04 — 471.92 = 11.12 Sum of squares S§ << SSC= 14.24 ANOVA TABLE Probabini ang it Sati, SSR=3.04 SSK = 454.64 SSE=11.12 Between columns : Table value of F, (4, 12) at 5% level = 326 “. the calculated value of F, > the table value of Ey ~. there is significant difference between columns or plots. Between rows : Calculated Fp = Table value of Fp (2, 4) at 5% level = 5.91 “© the calculated value of Fx < the table value of Fx So there is no significant difference eee rows or yell Between letters : Calculated value of Fy = 12261 Calculated F, = 3.84 1,22 of Expenments pesigt le value of 465 Cae aad nk (4, 12)p at 5% level «5 = 426 the calculated value of K > the table Value of Fy there is significant difference between the lett, ler or fertilizer , 46 Merits and Demerits of the common : CRD, RBD and Lsp 'Y Used designs 46.1 Completely Randomized Design (CRD) Merits [AU 2011) 1 It has a simple lay out. 2 There is complete flexibility as the number of replication is not 4 The analysis of the design is simple as it results in a one-way Classification analysis of variance. 4 Analysis can be performed if some observation are missing other designs The experimental error is large as compared bs sopteeaey homogeneity of the units is not taken “$2. Randomized Block Design (RB) i fficien! ‘i has simple layout but it is more ©! Teduction of experimental orf" of is exible and so any number Teplication may be uscd: init results gn is simple + than CRD because y number ¢ treatments uae in a two-way The analysis of the desi anak °sSification analysis of V4" Probaty itty nd lati, é © even if some observation ay, ot is itm he most popular design with experime . iments 4, simplicity, flexibility and validity. Peasy of Demerits 1. If the number of treatments is large, then the the c , size 9 Will increase this may cause heterogeneity within ‘ 1 Ha locks, 2. The shape of the experimental material Should be ,, ectangy, la 3. If the interactions are large, the experiment may results. me mie 4.6.3. Latin Square Design (LSD) Merits (or advantages) TAU apy 1. Latin square design controls variation in’ two directions of experimental material as rows and columns resulting in the of experimental error. 2. The analysis of the design results in a three-way classification analysis of variance. 3. The analysis remains relatively simple even with missing data Demerits The number of treatments should be equal to the number of L and columns as the area should be in the form of a squat 2. The process of randomisation is not as simple as RBD. bel 3. It is suitable only for a smaller number of treatments, S*Y S’and 12. 4, 2X2 Latin square design is not possible. experiments pst gf 407 pom 5 a and contrast LSD ang RBp. [A.u. U. 2003, 2013, 2014) ; —— ___ Bp . fhe number of rows an ere is no 4 columns are equal and hence |It can h the number of replication is lreplicat lequal to the number of uch Testriction, ‘ave any number ons and treatments, {reatments. itis suitable for small lumber of treatments, between 5 and 12. No such Testrictions suitable for upto 24 treatments. Experimental error is reduced to a large extent, because Wariation is controlled in two directions. Experimental area must be a square LSD is preferred over RBD because of (3) Variations is controlled in Jone direction only. Suitable if it is a rectangle or square. IRBD is the most popular one for its simplicity, flexibility and validity. io | "ple 8 Why a 2x2 Latin square is not-possible ? Explain : : > [A.U-2006, 2015, 2016] 1 the degrees of feedom fy giier an Xn Latin square design, the Ly = @—-1)~(-1)-@-)- 0-9 Bet 1— 343 non 4+ 2 Os gp i not defined. co MSE 8 sign err a sce 0 and hen 2 Latin Square Des sons are not possible. Hence * z frovability ang Stari, istieg Part - A 1. Name the basic principles of experimental design, (A2y 2. Define analysis of variance. i When do you apply analysis of variance ? TAU. What are the assumptions in analysis of variance ? Explain one-way classification technique of analysis of Variance, 3. 4 5. 6. Describe F-test with regard to analysis of variance. 7. Write the ANOVA table for one way classification, 8. What is a completely randomized design ? Define the term completely randomized design. [A.U-2005] it is a Latin Square Design ? type of analysis of variance results from the data of a Latin Design ? is a randomized block design ? (MUS6 two differences between completely randomized design ad ized block design, [avo and contrast the Latin square design with the randomize! ign. A fau-200s) 2% 2 Latin square not Possible ? sown in 12 plots. of Experiments ee Plots A 14 16 18 4 desi BH ie i ign, ' Bcc, Mt 19 swing ) ‘ : : ae Beret 2 ference in tig Production of the three varieties ? sre re the munber: of mistakes! mage in 5 successive 3 eects Working Yor 4 Photographic laboratory, T analysis of vari E % | Technician 1 | Technician [1 | Testis 1 Variance, 14 10 5 sification, PaO Pata the level of significance a = 0.01 whether the diftrence ibuted to chance. the 4 sample means can be attribute + eg f plot under I following table gives the yields of 15 samples of pl Varieties of seed. | 4.70 Probability and sy, latist Using analysis of variance, test whether there js 2 signi, ; ; ify difference in the average yield of seeds. ican 19. The following table shows the yields per acre of a certain ‘ . . Varigg of paddy grown in a particular type of soil treated with 3, i Varietiog of manures A, B, C. cs Test whether there is significant difference in the between the manures. a The R and D manager of an automobile company wishes to study the effect of tyre brand on the tread loss (in milli metre) of tyres. ‘our tyres from each of 4 different brands (A, B, C, D) are fitted four ‘different cars using the completely randomized design. data as per this design are presented below. Tyre brand : 8 whether the tyre brand has effect on the tread loss of is Significant level of 5%. on of Experiments esis 47] 2: Tw ation (RED) The following table gives the i. imber “, in three months. OF ACS sold by 4 salesmen Salesmen Determine () whether there is any significant difference in the average sales made by the four salesmen, (i) whether the sales differ with respect to different months. ) ofS | 2 To study the performance of three detergents and three different aft! | water temperatures, the following ‘whiteness? readings were obtained desig With specially designed equipment. Detergent | Detergent A B r + i @ 68 49 e iS 54 is ‘ssing 5% level of a fiance, Petform a two-way analysis of aH £2 Probability and \Stiey 23. A laboratory 1 ician measures the breaking strength of 5 kinds of line threads by using four different m instruments I, L o f cach ai CaSUrin, >» 1, and I, and obtain the following Tesults owners. A doctors cach test treatments for a certain disease and obs wumber ‘of days each takes to recover. The results are as ys (recovery time in days). Treatment iments. d discuss the dith aus = octors and (ii) treatinents TEC bene a4 igs By, B,, . our breeds of pigs mae ie Be were fed on three ee ‘ , R,, Ry gains Weight in kgs oye. ent rations Ry Ry ‘BS Over g VEN Dering oa recorded. Is there significant difference between (i) rations, (ii) breeds 7 The following is a Latin 5 of seeds are being tested slate your conclusion. quare lay out of a design when 4 varieties set up the analysis of variance table and . ‘colour ' fects of 5 colou ‘ Gictring fiem wants to investigate the ix Variations in ities On the setting time of a new ea changes 2 i i om who tp times can be ae different workers © and humidity an yf neous sources Of "ae the test moulds. To eliminate these - y Probability and Stas variation a 5X 5 latin square design was used in which the erg, A, B, C, D, E represent the 5 additives. The setting time ; hours, for the 25 moulds ate shown in the table. (Can we say that the colour additives have any effect on the setting of the concrete mix at the 0.05 level of significance ? Fo = 145%, Fogs (2, 9) = 4.26 ; Fe < Fos Fo = 147; Foo, (16, 3) = og Fee Foor No difference Fo°= 814; Foos (% 12) = 3.89, FL > Bigs There is significant difference in the average yield of the 3 varie ies, 8 Fo = 95 Fos (2 9) = 426; Fo > Fyig There is significant difference between manures 1 (i) Fe = 1.02, No significant difference (i) Fp = 3.33, No significant difference 985, Foos (2, 4) = 6.94, significant difference in the three detergents. F = 238, Foos (2, 4) = 6.94. No significant difference. Water temperatures do not make a difference A Fo = 124 3 Fogs (12, 3) = 8.74 no difference RE, +S difference» i A = 198; Foos (12, 4) = 5.91 significant Rg we : ; ie e i Fp = 299 between treatments is ,, -Sificant difference (ii) difference % Nie % ‘ Sifference 3 no difference ee agp icant difference between 1 "N varieties of seeds id yn columns ani d betwee! ‘ows al EE AL DESIGN 4.7 22 FACTORI 4.7.0 Introduction | design is one : logical foundations of experiment it tora design, the experiment can ga alg ‘or more factors when used simultane, 0 Factorial of the most fruitful advancemeny; endevour to improve the In experiments based on the combined effect of two Mate Nish, A special type of experimental design is the 2" factorial des hhere the effect on the response of 7 factors, each with two lee studied. The factors are independent experimental variables, The a levels may be quantitative or qualitative. For example two y, alas temperature, say 25°C and 30°C or hot and cold, two machines, «i But generally we denote the levels as ‘high’ and ‘low, though it nay be arbitrary in the case. of qualitative variablés. The complete factors design requires that each level of every factor occur with each level of every other factor and thus giving a total of 2" treatment combinations, of variance. 2" factor design ctor has 2 levels and so it is impossible to judge whet ts produced by variations in a factor as linear or P arabolt ential. ber of factors is more, then the number of exper iment may large. For example in a 2° factorial dest” Combinations are required, which is large. wel igl the large number of treatment combi nations “a R we use only a fraction of the experime™ ou gn gn is called fractional factorial design. 1" syperiments al . e- Ra he js to reduce the size of the tot ls ay al sae a er ations regarding the fact, Petiment ,, mv ornations ae taining the { ? factorial design we when there are two factors A,B ang tae 2. have a 2” factorial desi : each factor We Be . ‘orial design, Inspite of tg ae Bn i. eo, ign is a powerful tool to impr, % 87 is the an S| je design PE BNE 05 sees ig, n for combination of factors ‘or example co ty | The treatment combinations or experimental cond a iy small letters as 1, a,b,ab. In this notation y chines gnesponding to a factor indicates a low level of t gh’ and ‘tow, though it ay | mnee indicates high level of the factor. Thus ab indicates high level ables, The Complete favor | ihe factors A and B in this treatment. The symbol ‘1’ is used to factor occur with each kx | tate the combination of low level of both A and B. a denotes low 2 treatment combination, | el of B and high level of A. b denotes low level of A and high litions are denoted absence of fetter the factor and the i: bel of B, 10 study the effect i ; dependently by | Note: We can also use ‘1’ for high and ‘0’ for low or: + for high . ‘81 ~ for low, a These can be shown in the following table. Level of factor . Level of factor ae B a - OR Sf + oy . a totals ote the treatment (Ol ans combination a treat 8 to experimental conditions oF ld ment = : | | i | | | Probability ang coe ey 4.7.2 Analysis of variance and the calculation of eff Pets . 2 * . in 2° factorial experiment Consider a 2? factorial design in which ther Observations per treatment combination (or ex ie. n replications. Let (1), a, b, ab denote the ti for each of the four treatment combinations. table gives these total yields. € aren XDetimeny ‘perimental Condition otal yields (or ee The following Wo-vay We define the following contrasts among the treatment totals. A contrast. = a +ab—b ~(1) B contrast b+ab-a-(1) og Experiments ie putation of sum of squares . Com) ote ‘ j th ? factorial design, each main effect and interaction effect h, as es of freedom 1 under the assumptions of independence and ejiy in the experimental design, we test the variations due to ppity io: ne ? . therwise. We find the sum of the squares for each contrast. ce oF 0 ro (Acontrast)” _ (a+ ab-b~ (yy? Pa 4n Rep a Bicontnaie P _ (bab a ~ (1? 2n 4n : ‘AB contrast)’ _ (ab + (1) — a - b)? AB = (ABcontrasty” _ (ab + (1) Y ei SS Pa an Now SSE = SST — SSA — SSB — SSAB Since each contrast has 1 degree of freedom, the degrees of inedom for SSE is = 4(n — 1) [v for SST d.f = 4n—1, others 1 each. “df of SSE = 4n-1-1-1-1 = 4(2-1)] ANOVA TABLE Mean square yariation ratio F 4.80 Probability and Statist, ey sie Note : +ab ! It can be seen that main effect of A = (eae) a 4 2n 2n s = Mean response at high level of 4 — Mean response at low level of ; = V1 Yo, Main effect of B = Mean response at high level of i, — mean response at low level of B Example 1. In a process of growing crystals it is believed that temperature-and ph influence yield of crystals, a 2” design is attempted and the following results are obtained for yield. Experiments io The two levels of A are 300, 350 (low, hi 1) , hig ‘The two levels of B are 2, 3 (low, high) So it is a 2” factorial design with n fg Tofind the mean square. we code the data by subtracting 20 from every valu e. Treatment Replication combination 4.82 Probability ang Sai ; Fa i leg A (Acontrast)? _ (28)? :: ie ssa = cot = “47 = 8% h® 2 2 _, were SSB = (B contrast)” _ 12" _ 18 ie eff 4n 8 od 2) 2 The = (ABcontrasty? _ 4? _ f SSAB a 772 a 1. SSE. = SST - SSA - SSB.- SSAB the cal = 138 - 98°. 18 - 2-2 a9 1 Hy 15 | eriments 2 483 el of significance, the table yay ue of FA(1, 4) All, )=7.7) of Fy 5% Fe the calculated value of F, > the table vay Hy is rejected, 42 je. there is significant sence a w the effect of factor A is significant, gp: The calculated value of Fy = 36 : > SSAB At 5% level, the table value of Fy (1,8) = 771 1 29 «the caleulated value of Fy, < the table value of Fy «1. Hy is accepted. ie. the effect of factor B is not significant. iieation AB: The calculated value of Fag = 25 At 5% level, the table value of Fan (4, 1) = 2.25 + the calculated value of Fan > the table value of Fag * Interaction of AB has significant effect. +b 3% a tab = 8 a *j => yim at =) 1G; Probability ang Statstcg y; 11 S., cease = 40 Yo = > = 12, Yio = 5 = 20 Ya = 500+); : = 19 Ya = 700 +) = 7 [16 +22] = ee cee : 40 = 70m +0) = 5[12+ 20) = 16 Yi -Vo = 19-16 = @) 2n Hence y,—y, = i) 301 + Yog) — 5619 +o) 1 1 i 2 (22 +12] ~> [20 + 16] eniments 9 oi owing data represents the results of a 2? lactorial fol ‘ 3 : Sy a h Pic and 2 levels each with four replications, Analyse wah 2 {2 Ny 4 og wit ¢ : ana Replications 4 i il Vv. 18 16 13.7 11.5 119 126 118 14.4 Solution : Let A and B be the two factors. Lat high ‘and low be the levels. n-= number of replications = 4 code the data by subtracting 12 from each value. The coded is Replications slelels 0.2 17 Probability and Statistics AB contrast (1) +ab-a—b = -03+10.1—5.1—(-02) = 49 _ (Acontrast)” _ Cae 15.41 SSA = rm 4x4 : tp 2. contrast)? _ 4n 2 2 SSAB = AB contrast)” _ (4.9)" _ 15 4n 4x4 2 sst = 24-F = 573467149054 132-135 34.7 — 13.5 =, 21.2 SST — SSA — SSB — SSAB 21.2 — 15.41 - 163-15 = 2.66 ANOVA TABLE MSB = SSB = 1.63 eriments on of Experi esis! tw 1 4 Ho : All the mean | ‘AL H, : Not all equal 163 jactor At The calculated value of Fy miiggig At 5% level of Significance, table Value of “iis Fa (ly12) = 475 the calculated value of Fy > the table value of F, 9.05 +13 4 Hg is rejected at 5% leyey, = 212 ie. the effect of A is significant, factor B : At 5% level, the table value of F, (1,12) = 475 = 266 “. the calculated value of F ‘A > the table value of F 4 oH, p is rejected. ie. the effect of B is significant. Variation ri) | Interaction AB : The calculated value of Fan = 6.76 At 5% level the table value of Fap (1,12) = 4.75 * the calculated value of Fan > the table value of F,, * Hy is rejected. ie. the effect of interaction AB is significant. 3. The following data are obtained from a 2° factorial ted three times, Evaluate the sum of the squares by the contrast method. Draw conclusions. Replicate 3 10 16 ue wey Solution: First we code the data by substracting 20 from every yaj,, . The coded data is ‘Treatment Replication combination n = number of replications = 3 N =4x3 = 12 2 2 Correction factor = = = oe = 36.75 A contrast = a+ab—b~(1) Fe ¢10 = 3) - (19) = 3 in Bsa Be 075 = 14134 MSA= 44.08 gg=4408 | 1 sso=10208| 1 |MSB= 10208 MSAB = ssae = 0.75 SSAB=0.75 | 1 4(n-1) mse = SSE SSE= SE=141.34| = 4x2 ; 41.34 8 ag |= 217,66, Sst=204.25 of B® © - yo of Ba table value he table v8" of ignition’: ants 4.90 ater sce Probability — Y and Sia Stati, Factor B : The calculated value of Fy = 5.78 At 5% level, the table value of Fp (1,8) = 5.32 the calculated value of Fy > the table value of Fy Hy is rejected at 5% level of significance ie. the mean effect of B is significant. Interaction AB : The calculated value of Fan = 23.54 ‘At 5% level, the table value of Fan (8, 1) = 161 the calculated value of Fag < the table value of Fay Hy is accepted at 5% level of significance. ic. the mean effect of interaction AB is not significant. Example 4. In an experiment conducted by a mining department, to study a particular filtering system for coal, a coagulant was added to a solution in a tank containing coal and sludge, which was then placed in a recirculation system in order that the coal could be washed. Two factors were varied in the experimental process : Factor A : Percent solids circulated initially in the overflow. Factor B : Flow rate of the polymer. stem determines The amount of solids in the under flow of cleaning sys! vd ‘Two levels of each factor were yach of the 2” = 4 com Mt, in the wncertiow of * p table. f a Experiments igh e analysis of the da complete analy data, Use 5 We effects : “Hee of sini mine nce to pion : Let A and B be (wo factor, ‘ sil First we code the data by Subtracting 20 from Om every value, n = number of replications = 2 ircsimes x % Total 2 combination a 4 0) 15.35 | -14.19 | -29.54 | 235623 | 201356 @ 1.42 135 2016 | 18% b -734 | -7.44 53876 | 55354 2993 | 11.424 294.508 | 269.957 N= 4x2 = 8,T = 46.66 A contrast = a + ab —b - (1) = 2.77 = 5a — (-14.78) — (29.54) = 4188 = b+ab—a-(l). B contrast s 2.71 - (2958) = 68 = -14.78 - 5.11 - (1) +ab-a~> 277 + 14. B _ (= 46.66)". 272.14 big = 292525 ose Probabiti wea. (Acontrast)® _ {aubey : SSA = m7 = 25 ssp - (Beontrasty’ _ (6.88)" _ SS = an 8 = 3.92 ‘AB contrast 22.6 SSAB = (AB contrast)? ~ BS = 6385 4n SSE = SST — SSA — SSB - SSAB = 292.325 — 219.24 — 5.92 - 63.85 = 3.315 ANOVA TABLE Mean square SS MSA = SSA = 219.24 SSA =219.24 SSB =5.92 SSAB =63.85 wriments \ VS ae” | cl of significance, the tab : op Nev value a5 MAY {value of Fy > the table yah eulated ke calculat ue of is rejected. ie, the effect of A ig WPih i som rence Ho actor A has effect on the filtering of epal ator BE THE calculated value of Fy = 7.13 At 5% level, the table value of Fy (1, 4) = 7.71 «the calculated value of Fy < the table value of F :. H, is accepted. ie. the effect of B is not significant Factor B has no effect on the filtering of coal. Interaction AB : The calculated value of Fay = 7693 Tm At 5% of level, table value of Fa, (1,4) © the calculated value of Fay > the table value of Fag © Hy is rejected. Ineraction AB has an effect on the filtering of coah es A contrast _ 22 - 13085 2n 4 B contrast 2n AB contrast BOR Oe Ss i ene f se and i) : Stain, yo" Example 5. Given the following observations for two fae, 1 fie e \ctors é at two levels compute (i) the main effects Gi) make aq A ane wot variance. Maly a} a jot? Solution : Given n = 3, the no. of replications We code the data by subtracting 20 from every observation Replication -10 -6 ce sl 3 =. 0 4 yp Esperiments a a Contrast fects of oe ms 2n Main effect of B contrast ae n 6 ~ 35 effect of interaction AB = AB contrast me wai mt 6 ~15 2 i Ssre st (i) iT N = 110+ 4 78+ 171 ~ 60.75 = 20935 ssa = Asontrasty? (51)? 4n = Sr =-26ds _ Beontrast)? — (21)? SSB = fn = SD = 635 _ (AB See -9 peal | es 3 SSAB = = 615 Es SSE = SST — SSA — SSB - ssaB LO ia Ea = 298.25 — 216.75 - 36.75 - 6.75 = 38 1 av 3 ANOVA TABLE 9 | % 3 a2 6 eG Variation ratio 9 Oo bee ; SS df MS F mn 216.75 _, A SSA=216.75 1 MSA=216.75 Fa= 475 aes Fe= 65 7S =774 SSB=36.75 1 MSB= 36.75 a= 475° Fp = Sees 12 =6.75 1 MSAB=6.75 re 38 mge=$=478 i “yang 8 i . Ste pe Hy : No difference in the mean effects ie Hy: Mean effects are not all equal qh . NATEAVallielok * cep Factor A : The calculated value of F, = 45.63 (od % level of significance, the table vz . At 5% level of significance, the table value of F, (1,8) ‘re com the calculated value of F, > the table value of F, Hp is rejected, ic. the difference is significant. -. the effect of A is significant on the response. B : The calculated value of F, = 7.74 S he table value of Fy (1,8) = 5.32 ed value of Fg > the table value of Fs significant on the response. ated value of Fan = 142 5,32 value of Fay (1,8) = Fyn < the table value of Fas respo B is not significant on the 4 19) oo (EXERCISE 42>) data are obtained from a 2? f \ctorlal experiment Squares for aly ed three times. Evaluate the sum of effects bY contrast method. Draw conclusions, fp ie es Replicate Replicate © ofp : 25 it, 19 C = ans, SSA = 4033, SSB = 85.33, SSAB Fy Probability ang ¢ it Misticg 3, A preliminary experiment is conducted to study the effec oe . cts factors and their interaction on the output of a certain Of fou IN machi, ning operation Two runs are made al cach of the treatment comb Dinat in order to supply a measure of pure experimental ery, oy : or, levels of each factors are used, resulting in the data show : D beloy make tests on all main effects and interactions at the 0.95 of significance. Draw conclusions. Treatment Replicate Replicate conclusion I i 79 9.6 10.2 Sele 58 fects of all the three factors are not significant A eriments: CHAPTER - 4 Design of Experiments (pant A - QUESTIONS AND ANSWERS the basic principles of experimental design, [AU 2003, 2014, 2015, 2016} ans. The three basic Principles of experimental design are pando misation, 2. Replication and (3) Local Control : Name 7 State the Prime objective or aim of design of experiments. [AU 2014, 2015] ‘Ans. The prime objective of design of experiments is to control ‘te extraneous variables so that the results could be attributed only to ‘ie experimental variables. What do you mean by analysis of variance ? Ans, Analaysis of variance is a technique that analyses variances. ‘separates the variance ascribable to one group of causes from the ascribable to other groups. = [AU 2005] ‘When do you apply analysis of variance ? ns of more Ans. When we have to test the differences between moa ‘Wo samples we use analysis of variance. 2 t are the assumptions in using ANOVA ? [AU 202] The assumptions are 'the samples are drawn from normal popula sas the samples are drawn independently from these P ‘ the populations have the same variance « ce A should not be uscd if we cant! make t “ee Ke the term completely randomized ne as oe i ign the cation. It is @ completely randomized er random alloc ntal units by a proce units are homogeneo¥s 4.100 Probability and Say list 8. What do you mean by two-way classification in analysis or variance? Ans. When data are classified according to two factors One classification is taken column wise and the other row wise, gy ’ classification is called a two-way classification. * State 2 differences between completely randomized design and randomized block design. [Au 2004] Ans. 1. Completely randomized design (CRD) analysis resyi ts in one-day classification, whereas randomized block design (RBD) analysis| results in two-way classification. 2. Experimental errors are large in CRD compared to RBD and RBD is more popular. Ii. Describe Latin square design. - Latin square design controls variation in two directions of ntal material as rows and columns resulting in the reduction rimental error. The analysis of the design results in a three-way tion of analysis of variance. mean square. ! [AU 2007, 2011] square is obtained by dividing the corresponding sum s degrees of freedom. ie aim of the design of experiment 2? [AU 2007] 15 Question 2. e on completely randomized design. {AU 2008) itin-square design. [AU 2008, 2m) Muestion 11. d alternative hypotheses for a completely [AU 2 sare equal = Ay = lt ic. the treatment mean: gt of Experiments ye 4.101 js the purpose of blocking i é a locking in a randomized block design? [AU 2009) ans. If the yariation due to heterogeneity in experimental units j nits is lage then the sensitivity of detecting treatment differences is reduced se of yoriation due to W smaller homogeneous blocks. 2 Jarge value of s”. A better idea would be to "block off” these units and thus reduce the extraneous variation fl State the advantages of factorial experiment over simple experiment. [AU 2010] Ans. Factorial design is one of the most fruitful advancement in ‘he endeavour to improve the logical foundations of experimental design. In experiments based on factorial design, the experiment can evaluate d effect of two or more factors when used simultaneously. us to study the effect of each d independently by the usual Analysis lesign (RED) as, ‘he combine The factorial design experiment enables every factor to be estimated and teste nalysis of variance. ; State the advantages and disadvantai ‘hase _ Ans. Refer Page No. 2.65 ges of randomized block {AU 2010, 2014] t are the advantages of a Latin square design ? [AU 2012] The advantages of a Latin square design *1° directions of the tion in two . ng in the reduction lumns resultit square design controls varia erimental material as rows and col ' perimental error. of the design results in a thre of variance, eeway classification of even with missing data. is remains relatively simple ae

You might also like