any 7 WOR Hund A take fluid B and those who tock fluid B take called a crossover design. It is analyzed as a set of 10 Latin squares the 10 squares correspond to ‘The layout of this design is shown in Figure 4.7. Notice that the rows in the Latin square fepresent the time periods and the columns represent the subjects. The 10 subjects who received fluid A first (1.4. 6.7.9. 12.13, 15. 17, and 19) are randomly determined. of variance is summarized in Table 4.17. The subject sum of Corrected sum of squares among the 20 subject totals, the period sum of squares is the corrected sum of sq uares among the rows, and the fluid sum of squares is computed as the corrected sum of squares among the letter totals. For further details of the statistical analysis of these designs see Cochran and Cox (1957), John (1971), and Anderson An abbreviated nalysi squares is computed as the and McLean (1974). It is also possible to employ Latin square type designs for experiments in which the (reatments have a residual effect—that is, for example. if the data for uid B im period 2 still reflected some effect of fluid A taken in period 1. Designs balanced for residual effects are discussed in detail by Cochran and Cox (1957) and John (1971). 43 __ The Graeco-Latin Square Design SN ed Consider ap p Latin squ: e. and superimpose on it a second p X p Latin square in which treatments are denoted by Greek letters. If the two squares when superimposed have the property that each Greek letter appears once and only once with each Latin letter, the two Latin squares are said to be orthogonal. and the design obtained is called a Graeco-Latin square. An example ¥ 4 Gracco-Latin-squate is shown in Table 4.18.RAMPLE a STABLE 4.20 SraccoLattn Square Desig fnantes 4 Mf Experiments with Blocking Factors “The concept of orthogonal pairs of Latin squares forming a Graeco-Latin square can te extended somewhat. A p % p hypersquare is a design in which three or more orthogocal p *r Latin squares are superimposed. In general, up to p + 1 factors could be studied if comple st of p — 1 orthogonal Latin squares is available. Such a design would utilize all (P IXp-D= p' — 1 degrees of freedom, so an independent estimate of the error variance is necessary. Of ourse, there must be no interactions between the factors when using hypersquares-the ysis of Variance onparametric Methods -31Ld The Kruskal-Wallis Test in situations where the normality assumption is unjustified, the experimenter may wish to use an alternative procedure to the F test analysis of variance that does not depend on this assump- tion, Such a procedure has been developed by Kruskal and Wallis (1952). The Kruskal-Walii test is used to test the null hypothesis that the @ treatments are identical against the altemativ: hypothesis that some of the ireatments generate observations that are Jarger than others. Because the procedure is designed to be sensitive for testing differences in means. it is sometimes con- venient to think of the Kruskal-Wallis test as a test for equality of treatment means. The Kruskal-Wallis test is a nonparametric alternative to the usual analysis of variance. ‘Fo perform a Kruskal-Wallis test, first rank the observations ¥, in ascending order and replace each observation by its rank, say R,, with the smallest observation having rank | hiLets inte Araneae 429) the exe of te same value), assign the averape ask cach ot Several et eo thease en The tet a tf a aa jin : mumbo oF ebscrsity he fh eaten 5 the ttl gabe of ver get [ege men san v JSR ‘ ne ‘Noe that Sus the variance he ean, HE there are mois, = NIM + IYER ad he test statistic sins to He gihowen When the numberof tes is muse, dene will he Ith iy lege ain! 3683, and the sips Con Eton 14934 may eased H ke ma ee he nl yeuies. TR say n> 5. His dintbuted aproninarely the null hyphens i ejected. The i edhe es aac [EXAMPLE 3. peda fom Example 31 te ctexning mks ae ube 3a Thee ne i use Ege gat sect st ro Eon 3.65 a As 2 oe fame BE] ser TABLE 9.20 Rank fr the Paso Kiching Experiment ib re 1 ” * 10 Tey ae 2 » : aa i a a mo is > 5 oe us a ¥ 4 & 0 ws 7 : nt psn of variance PRES. ts em HF yugy = HEL ne mld ee te tl ng ee ‘Nahr axe by he wna (ess an che that he Hennes f(T