Test 2 St 711, 2011, Dickey I. (10 pts.) In the term “Balanced Incomplete Block Design” what does the word “balanced” mean, that is, what makes an incomplete block design balanced? II.(15 pts.) I randomly assign 3 feeds to sets of 30 cows (10 per feed). I measure Y= butterfat content in their milk for 4 consecutive weeks in each cow thus having a repeated measures design with 120 observations. My covariance matrix R is 20 12 12 12 (1) 12 12 12 20 12 12 12 20 12 12 12 20 Find, if possible the variance of Y ________ (2) Find, if possible, the cow variance component __________ (3) Find, if possible the variance ________of the difference two Y values for different weeks within the same cow. III. (8 pts.) In the discussion of balanced incomplete block designs, we mentioned lattice designs in which sets of incomplete blocks (like fields) with =1 combine or “resolve” into larger sets or “superblocks” like farms with certain properties. (A) What characterizes these larger sets? (B) In comparing two treatments, what (if anything) is the advantage of such a lattice versus just having the incomplete blocks? IV. A split plot in a completely randomized design has two levels of the whole plot factor (A) and three levels of the split plot factor (B). (40 pts.) Here is a table of treatment totals, each being a total of 12 observations. Write down in the ANOVA table, as many degrees of freedom and sums of squares as you can from the given data. B low B medium B high A low 24 60 36 A high 36 48 12 Source df Sum of Squares A ______ ________________ Error 1 ______ ________________ B ______ _________________ AB ______ _________________ Error 2 _______ ________________ (12 pts.) Assuming sums of squares 100 for any entries you could not compute, give the calculated F test for A main effects (F=__________ ) and for B main effects (F = _________) (9 pts.) Using the linear orthogonal polynomial coefficients -1, 0, 1 and assuming the levels of B are equally spaced, compute the sum of squares for the linear effect of B __________________________V. (6 pts.) Is it possible to construct a Youden square design with 6 rows, 10 columns, and 10 treatments? If so, how would you construct it (assuming you start with a 10*10 Latin Square)? If not, explain how you know. *******************Answers***************** (1) Every pair of treatments appears together in blocks equally often. (2) 20, 12, 2(8)=16 (3) The sets of incomplete blocks that form the larger superblocks contain one of each treatment, that is, all treatments appear in each superblock (farm for example) In the lattice case, the superblock effects, farm effects for example, drop out so we have a smaller variance and higher power. (4) B low B medium B high A low 24 60 36 sum 120 SS(A)= 1202/36 + 962 /36 - 2162/72=8 A high 36 48 12 sum 96 Sums 60 108 48 (216) overall mean is 3, 72 obs, 3 per whole plot SS(B) = (602 +1082+482)/24-2162/72 = 84 SS(table) =( 242 + … + 122)/12-2162/72=120 Source df Sum of Squares A __1_____ _______8__________ Error 1 __22_____ ______(100)_______ B ___2____ _______84__________ AB ___2____ _______28___________(120-8-84) Error 2 ___44____ ______(100)_________ F for A is (8/1)/(100/22)=1.76 and for B it is 42/(100/44)=18.48 For linear, 48-60=-12 so 144 when squared. The denominator is (1+0+1)(24) so sum of squares is 144/48=3 for linear B effect (1 df). Note (not asked for) that F is then (144/48)/(100/44)=1.32. (5) Columns of a Youden square form a balanced incomplete block design so we must have r(k-1)(t-1). With t=10 treatments and 60 observations r is 6 and so 6(5)= (9). There is no integer that satisfies this. It cannot be done.