words, powér is equal to 1 minus the p: as being We will now examine a few of the more widely used distribution-free of sampie size When samy (ests are easier, quicker, iple sizes ate <10, and only slightly I distribution-free sta; -xperiment—nonparametric nore convenient than their par of » mathematical criterion of si n superior or equal to their parametric ivaranhetric test are met, but the assump. © nol. If both tests are applied when all.” test can be met, distribution: an extremely small sample size, but as sample size increases. ric counterparts.:In he assumptions of th ecome increasingly less effi stical tests are more {ests of significance, together with their application. The studerit who wishes to study this topic in more detail ma Siegel (1956), and Tate and The Sign Test ‘The sign test is amon, discussing this test, we shi in pairs. Individuals are matche Distibuton-Free Siatstical Tests 249) obability of making a Type Il error, 8B. One important factor contributing to the power of a stat testis the sample size, N. As N increases, the probability of making 2 W error decreases and hence the power of the test increases. It wi noted that some of the statistical tests discussed i te powerful than others. in this chapter are stated refer to Bradley (1968), Kraft ig the simplest of the tests for correlated samptes, In refer to Table 17 and assigned to one of ing of the experiment. A treatment i scores for the pairs of individuals ke this it is assumed ed continuously and TABLE 17.1 THE SIGN TESTtical Methods: 1 uputations or by giving one tie a plus sign and the next a ‘and 50 on I hypothesis tested here is that the median change is zero. This is rejected if there are too few of any one sign. Where there ns pairs we apply a test of significance by the use of the binomial with p = 5 and N = the number of pairs. For our data. N Prable 17.1, we note that we have 8 pluses and 2 minuses. By would expect 5.of each: The question is, does this frequency ‘iffer significantly from what we would expect by chance? With’ the binomial expansion we find the probability of obtaining B plus signs is equal to .0009766 + 0097656 + 0489452, which 10547. The process of doubling this probability for a two-tailed 's p of .1054, or .11. This forces us to let the mull hypothesis this test uses only information on the direction of the differences irs, it is not the most efficient. The next test combines the sign ference with the relative size of the difference. Itis a more powerful n’'s Matched-Pairs Signed-Ranks Test . rate Wilcoxen’s matched:pairs signed-ranks test of significance, we ‘se the data presented in Table 17.1, recopied into Table 17.2 pabiain the difference between each pair of scores, then we rank Wolute values of these differences, “absolute” meaning to disregard LE 172. DATA OF TABLE 17.1. USEO TO ILLUSTRATE THE WILCOKON JICHED-PAIRS SIGNED-RANKS TEST ‘Absolute signs in our ranking. For these data the two differences of 2 are the smallest Since these occupy positions one and two of the ranking, we give-the average of the two ranks, this being 1.5. The next smallest difference is —4. We give this a rank of 8. We continue in this manner until all the difference: are ranked. Whenever a difference of zero appears we disregard it in the computati Tn the last two columns of Table 17.2 the ranks have been summec according to the sign of the differences. The smaller of these is taken a: Wilcoxon's T statistic. If there were no difference between the two groups T would equal T, the mean sum of the ranks. This latter statistic can be obtained by the following formula: NOW 1) T= 4 a7 For these dita 10(10 +1) T= dT differs significantly from T, we can reject of ranks, for N's between 7 and to test the significance of T. For our pro! assume a OW test with-alpha equal to .05, From Appendix I, for N =.10 we see that : T of 8 oF less is significant at the 05 level for the two-tailed test. Our 7 is exactly 8; hence we reject hyp we compare our findings obtained on the si coxon test, we find contradictory results. VW to reject the null hypothesis, which we rejected at Wilcoxon's test. Thus the Wilcoxon test is a more pow’ those for the Wil_Mw+h) fant nT NCW VEN FT f or ¥ ta a7 and the'sum of the T's may be treated with the familiar z test with T— T: as the deviation. : ce INDEPENDENT RELATED DATA ‘The Median Test ‘The median test is applied to see if two groups come from populations that have the same median. In‘using this test the size of the two samples. need.not be the same. Table 17.3 shows arithmetic addition scores for 27 the hypothesis of no difference between these! two sets of scores by using the: median test. Group X contains 13 scores! and group ¥. 14. First, for the entire set of 27 scores we compute the . which we find to be If both of these distributions came fi and half of each of the two di table is set up as ions would lie below it. A contingenc ma om viatan | 7 8] 8 i[ 3] % wanow [6] 8 | 1 pol cia | ou fe ma | lw ‘We next make a test of significance, using d {ad ~ be) — N/2}* 2166-7" ( 33,124 = 034 which with 1 degree of freedom is not It frequencies in the « pp. 96-104). Thi Tine 2742.25) y. A more complicated be used. This may be found i ‘ore than (wo groups{74 TWO SMALL INDEPENDENT SAMPLES ILLUS: THE USE OF THE MANNWHITNEY U TEST Ay = 495 1. the sizes of which’ need not bé the same, When the simple sizes Y small—that is, when both M; and Ng are made up of eight’ or Isures—the reader is referred to-Siegel (1956) for a method and fnat apply to such sample sizes. When the larger of the two samples’ more, the procedure described below is used. 4 Table 17-4 we have two samples, Nx with seven cases and Ny with ses, In this case We afe testing the hypothesis that both samples" Irom the same population. First we rank all the scores in one composi sion in an increasing order of size (Considering algebraic signs when «measures are present). In Table 17.4 the score of 3 in column ¥ lowest one: hence we give it a rank of I. The score of 4 in c Second lowest, and we give ita rank of 2. We have a score the X and Y distribution, We handle such ties by giving them is distribution is 8.5. We cont » rank, Then we fe columns of ic check © wo MN + DA is pre tie NN + ELEY — sky azs) daa yt SED ary (1732) For our problem: v= 79 22-495 = 63 +28~495, =415 Equations (17.8) and (17.3a) give different results. To complete our problem we need the smaller of the two. By solving equation (17.32) we have: veri, #20265 “= 63 +45 — 86.5 Ug NiN2~ Uy =10)—415 were making a directional test, p's of .002, .02, 05, suppose that Ny = 10, and Nz = 15. From part (6) of Ay kes it Le to reject the null hypothesis at st or at the 2 percent level for a one equal to.01 for our problem, Examina ws that when Ny = 7 and Nz = 9, C 10, For example, ix J we see tailed test. Suppose tion of part (6) of Appendis175_ ILLUSTRATION OF THE MANN-WHITNEY TEST FOR, 3 ‘BETS OF OATA y 18 125 310 10 4 180 270 2 15 as 200 3 6 16 109 300 8 12 145 215 ‘ 10 65 180 e 8 109 145 “ 13 270 240 2 13 28 240 4 x 65 45 6 10.0 7 125 3 45 5 80 : 25 1 10 13 249 oy 210 ct 200 . 0 180 _e - 160 Niz21 - NMetO EArsa728 EA, 2205 some fictitious data have been assem! re are QI cases and in group ¥ ( ¢ data are treated two sets wres are ranked toget smallest ‘Then these two columns of ranks are summed. Again these can be ed. M(N + 1)/2 equals the sum of the two sets of ranks. Next Us is 1 by formula (17.3) Nai # mun ED say © following equ Nata + Na+ DY i term, which is the mean of where the numerator is U; minus the second term, which is the mean of | WU. and the denominator is the standard deviation of the U- | 168.5 = 21(10)/2 = FRMOVET + 10-F HAZ | _168.5— 105 VeFRO712 85. 560 3s" 723.7 =2.68 ‘Ace value of this size leads t6 the rejection of the null Hypothesis srcent level per This problem could have also heev: suivea ior Us 210-4 55 293.5, =415 If this value of 41.5 is inserted in the the valu of the‘THE KRUSKAL-WALLIS H TEST ble 17.6 shows three sets of data. 7 The scores of the three sets are aod ranked with the lowest séore receiving a rank value of 1 proportionately among the various of sample size. The sums of the rank of the various tested against these proportions of the total sum of ranks, of the following formula: coal — 2 2 warn 2S ]-se+ 75) N= the number in al Ry =sum of ranks an amples combined “the numbers in i samples pe data in Table 17.6 ~ 8 [Be nagls et 5 ]-92) = 22 (030.45 + 1389.03 +7 Sp (098.45 + 1339.08 + 729 ~69 =.0287(8006.48) — 69 =713~69 =23 i the number of cases in the samples ed in the interpret the number is equal to 2, contain five or more cases, H is interpreted as chi-square wi of samples minus 1 degree of freedom. For this problem and this H is not significa ‘The Kruskal-Wallis testis a power nonparametric test. Exercises 1. The following represent the scores of 9 individuals on a pretest and a osttast. Apply tha sign text to the data and draw the appropriate conch sion, Aree. 16 22 8 12 B13 i 30 1 Gear, 22 28 fioleem 71 1 18 ched-pairs si low are the scores of @ groupbelow are three samples of scores obtained on the arithmetic ‘Wechsler-Bellavue. Apply the median test to all three sar- ly the Mann-Whitney U test to the data in problem 4 18 the data in problem 2 to be uncorrelated. Apply the Mann-Whit- .de the following scores on the vocabulary use of the Mann-Whitney U test, test for ‘9. By using the using the Kru 40, The folowing data trom Bossi: shows ‘of three groups on the Edwards Per the differences among these three groups, using any appropriate nonpar- amotric tests. octal Personal contro! —_Destrabitty Desirability Group Group ‘Group Group Group SOUP 47 76 er 38 78 76 34 6 52 32 3 at a2 62 50 30 a a 38 {Journal of Applied Paces, 1958, 42, 25Statistics Made Simple Ie CHIE! Thif'is a test of difference between the observed and ted frequencies. The Chi-Square is considered a que test due to its 3 functions which are as follows The test of goodness-of-fit The test of homogeneity The test of independence HE CHI-SQUARE TEST OF GOODNESS-OF-FIT This is a tesi of difference between the observed frequencies and expected frequencies. Let's take as an example the theory of Mendel regarding crossing of peas According to Mendel’s theory, the yellow color is dominant ver green and in terms of skin, smooth skin is dominant ver wrinkled skin. In this case, the theory (expected) is compared to the actual experiment (observed). The formula for the Chi-Square test is: (O-B)? X= E Where: X? = the Chi-Square test = the observed frequencies E = the expected frequencies Example 1 The Nonparametric Test 155 The theory of Mendel segarding crossingof peas is in the ratio of-9:3:3:1, meaning 9 Parts are smooth yellow, 3 parts smooth green, 3 parts wrinkled yellow and I part wrinkled green. The researcher conducted an experiment and the result. was that out of 560 peas, 310 were smooth yellow, 100 were wrinkled yeliow. 110 were smooth green and 40 were wrinkled green. Is there a significant difference between the observed and the expected? Use X?- tes: at .05 level of significance. Solving by the Stepwise Method: 1. Probiem: Is there a significant difference between the observed (actual experiment) and the expected (theory) frequencies? I. Hypotheses: Ho: There is no significant difference between the observed and the expected frequencies. There is a significant difference between the observed and the-expected frequencies, Iil.Level of Significance: a =.05 df=h-l Ai wAAWALThe Nonparametric Test 157 a2 50-5)? aa (310-315)? 315 %- te81 of géddness-of-fit lon: Add the ratio 9:3:3:1 (100-105)? 105 =.079 + 238 + 238 + 714 (Actual Result) (Theory) Observed Expected Xx? = [1.269 310 315 If the chi-square computed value i greater than the chi square tabular value, reject Hy 3 100 105 Ho 105 The chi-square computed value of 1.269 is lesser than “the chi-square tabular value of 7.815 at .05 level of significance with 3 degrees of freedom, so the null hypothesis is accepted. This means that there is no significant difference between the observed and the expected frequencies In other words, the theory of Mendel holds true because the value of the X*test did not warrant the rejection of the null hypothesis. : Example 2: A-certain machine is supposed-to. mix peanuts, hazelnuts, cashews, _ and pecans in the ratio of 4:3:2:1. A can containing 500 of these mixed nuts was found to have 270 peanuts, 110 hazelnuts, 80 cashews and 40 pecans. ISK 3~ 105 msn1 Made Simple the .05 level of significance, test the hypothesis is mixing the nuts at the ratio of 4:3:2 py the Stepwise Method: Is the machine mixing the nuts at the ratio of 4:3:2:17 The machine is mixing the nuts at the ratio of 4:3:2:1 I: The machine is not mixing the nuts at the ratio of 4:3:2:1 al of Significance: a = 05 df = h-l = 4 Xi, = 7.815 tabular value tistics: Chi-square ‘est, test of goodness-of. The Nonparametric ® Peanuts 110 pesk$07 577 F 3 Hazelnuts cash 2 0 ee 100 fashew Pecans a o Fa 5 ~ 710 500 300 ee « 50 * 10{ 500 For expected" ee) 50x 4“ J59 50x3 ~ 199 o 50x27 59 50x1* X?= 0)? , (80 -100)* , (40-50) Sais : a 9q.5 + 10667444? x faIf the chi-square computed. value Breater than the chi-square tabular vali, rejéct Hy, — The chi-square computed value of 41.167 is greater than the chi- square tabular value of 7.815 at .05 level of significance with 3 degrees of freedom, so the research hypothesis is accepted which means that the machine is not mixing the nuts in the ratio of 4:3:2:1 implies that the machine is not in good order because it does not anymore mix the nyts as expected, s concerned with two or more samples, ly one criterion vat This test is used to Jermine if two or more populations are homogeneous. Its 1 distribution are similar with. respect to a particular = grand total Almere = the product of the rows and columns __ The Nonparametric Test Example 1. To illustrate this, we can evaluate the attitude of a sample of Lakas and Laban -parties.on the issue“of peace and ordér in Mindanao. To carry out such study, a separate random sample of members of each party is drawn from the nationwide population of Lakas and Laban and each individual in both samples responds to the scale. Scores are then classified into . “Favorable” or Unfavorable” categories, The following frequencies are obtained: Se Favorable Unfavorable Total a re eee nee Lakas 65 35 100k a b Laban 50 50 1001 © d Total 15 85 200 m a N ing by the Stepwise Method 1. Problem: ference _b : tudes “of the two political parties sue of peace and order in Mindanao? a AIVORLIIN102_ Statistics Made Simple 11, Hypotheses: Ho: There is no. significant differenge between’ ie ides of the two fi the issue of peace and order Hy: There is a significant difference between t he attitudes of the two political parties on the issue of peace and order in Mindanao, Level of Significance: - ee ose Hf (c-Ne1) =Q-1)Q-1) statistics Chi-square Test of Homogeneity ee Favocchle Unfavorable Total © Lakes “635 TOR +b - 50 d 7 SBS oe a N ’. Decision Rule: The Nonparametric Test_163 = 200 (3250 - 17507 97750000 = 2001500)" ‘97750000 450000000 97750000 = [4.604 square tabular value, reject Ho. VI. Conclusion: Since the chi-square computed value’ of 4.604 is greater than the chi-sq value of 3.481 at .05 level means that ference between the Laban group has on the said issue.164. Statistics Made Simple The Nonparametric Test_165 ! THE CHI-SQUARE TEST OF INDEPENDENCE «(ONE SAMPLE, TWO CRITERION VARIABLES) The one-sample test of independence is different from the} test of homogeneity. The sample used in this test consists of! members randomly drawn from the same population. This testis used to look into whether measures taken on two criterion} variables are either independent or associated with one in a givea' population using such variables as level of education and income, performance in class and 1Q etc. The calculation of this test ig similar to the test of goodness-of-fit and the test of homogeneity ‘The formula is: Male 18 28 46 Female 32 12 - 44. ~ Total 50 40. 90.- Solving Stepwise Method: 1. ‘Problem: Is there a significant relationship between sex : i motor skill? ey (0-5)? and scores in psychomotor skill "E I. Hypetheses: Where: X? = Chi-Square test © = observed frequency E__ = expected frequencies Y= summation Ho: There is no significant relat sex and scores in psychomotor s| Hy: There is a significant rel and score in psychomotor nship between IL jonship between sex H Example 1. Ninety individuals, male and female, were given a test in psychomotor skills and their scores were Y=) d into high and low: Using the x? -test of aye cs — - independence at .05 level of significance, the a table is shown as follows: X? os = 3.841 tabular value : - o166 Statistics Made Simple Computation (25.56) 28 (2044) 46 Female 32 (24.44) 12 (19:56) 44 Se Tol 50 40 90 OO For expected values : Multiply'the column total to the row total and divide the product by the grand total. VI. Conclusion : = Ve Decision Rule : The Nonparametric Test _167 If the X? computed ¥alue:is ereatee-thanabe 2 tabular vali, reject HE The X? computed: value -of “10292 is” greater than the X? tabular value of 3:841 at .05 level of significance with one degree of freedom. This leads-to the confirmation of the research hypothesis which means that a significant relationship ..exists between sex and_scor skill, It implies that scores are more in psychom their male counterpart. Example 2. Two lots of 50x46 = 95 56 40x46" = 20.44 90 90 50x44 = 24.44 40x44 _ 19 56 90 90 xP=d 7 E . Solving by the Stepwise Method: = 18-25-56)", 32-24.44)" | (28-20.44)* | (2-19.56)? 25.56 24.44 20.44 19.56 52.2364 2.338 + 2.796 + 2.922 10.292 x= Recovered = Died mi iPiWi *, AWALLIA