252solnF1 11/7/05 (Open this document in 'Page Layout' view!) F. ANALYSIS OF VARIANCE 1. 1-Way Analysis of Variance Text 11.1-11.6, 11.7**, 11.8 [11.1- 11.7, 11.8*] (11.1- 11.7, 11.8* (Same problem, different numbers – both answers will be posted) 2. 2 -Way Analysis of Variance Text 11.15-11.18, 11.23, 11.29-11.32, 11.36 [11.15-11.18, 11.23, 11.28-11.30, 11.34] (11.15-11.18, 11.23, 11.28-11.30, 11.34), F1, F2, F4 3. More than 2-Way analysis of Variance F3 4. Kruskal-Wallis Test Text 12.86-12.87, 12.89 [11.39-11.40, 11.42] (11.39-11.40, 11.42), Downing and Clark 18-12, 18-13 (in chapter 17 in D&C 3rd edition), 5. Friedman Test Text 12.93-12.95 [11.46-11.48] (11.65-11.67 on CD) Downing and Clark 18-4, 18-6 (in chapter 17 in D&C 3rd edition), Graded Assignment 4 (Will be posted) This document includes exercises 11.1 - 11.8 ------------------------------------------------------------------------------------------------------------------------------------------------------------------ 1-Way ANOVA Problems. Exercise 11.1 (11.1 in 8th edition): Solution: The table in the outline reads: Source SS Between SSB DF MS dfB SSB MSB m 1 m 1 Within SSW dfW SSW MSW nm nm Total SST F F MSB MSW dfT n 1 We use B(Between) in dfB for the text’s less-standard A(among) in dfA . m 5 levels(columns), and n1 n 2 n3 n 4 n 5 7 . So n n i 57 35 . dfB m 1 5 1 4. dfT n 1 35 1 34. dfW n m is best found as a residual. If we fill in the table with the numbers above we get: Source SS DF MS Between Within Total . F F.05 H0 4 30 34 1 252solnF1 11/7/05 Exercise 11.2(11.2 in 8th edition): The table now includes the given values. SSB 60 and SST 210 . Source SS DF MS F.05 H0 F Between Within Total 60 . 210 4 30 34 SSW = SST – SSB = 210 – 60 = 150 is the missing number. SSB 60 (b) MSB 15 m 1 5 1 SSW 150 (c) MSW 5.00 n m 30 MSB 15 F 3 (d) = MSW 5 The table now reads as below. Source SS DF MS F.05 F (a) Between Within Total 60 150 210 4 30 34 15 5 3.00 H0 Column means equal Exercise 11.3(11.3 in 8th edition): (a)Take the table above and add the appropriate table value of F and the null hypothesis. Source SS DF MS F.05 H0 F Between Within Total (b) 60 150 210 4,30 F .05 2.69 4 15 30 34 5 3.00 s F 4,30 .05 2.69 Column means equal (c) Decision rule: If F > 2.69, reject H0. (d) Decision: Since Fcalc = 3.00 is above the critical bound F = 2.69, reject H0. The s for ‘significant’ shows regression. Exercise 11.4(11.4 in 8th edition): We use B(Between) in dfB for the text’s less-standard A(among) in dfA . m 3 levels(columns), and n1 n2 n3 7 . So n (b) dfB m 1 3 1 2. dfW n m 21 3. (c) dfT n 1 35 1 34. (a) n i 37 21 . Exercise 11.5(11.5 in 8th edition): The solution is repeated, edited, from the Instructor’s Solution Manual . 11.5 Source df SS MS F Between groups 4 – 1 =3 80 x (3) = 240 80 80 20 = 4 Within groups 32 – 4 = 28 560 560 28 = 20 Total 32 – 1 = 31 240 + 560 = 800 m 3. n1 n2 n3 n4 8 . n 48 32. 2 252solnF1 11/7/05 Exercise 11.6(11.6 in 8th edition): The solution is repeated, edited, from the Instructor’s Solution Manual . 11.6 (a)-(b) F 3,28 .05 2.95 Decision rule: If F > 2.95, reject H0. (c) (d) (e) Decision: Since Fcalc = 4.00 is above the critical bound of F = 2.95, reject H0. There is enough evidence to conclude that the four group means are not all the same. To perform the Tukey-Kramer procedure, we use m 4 in the numerator and n – m = 32 – 4 = 28 degrees of freedom in the denominator. Since Table E.7 or Table 21 in the syllabus supplement does not contain a value for 4 and 28 degrees of freedom, we will use the next larger (and more conservative) value for 4 4, 24 and 24 degrees of freedom as an upper bound. q.05 3.90 (On the other hand, a guess 4, 24 4,30 that the value of the number we want is between q.05 3.90 and q.05 3.85 and (f) thus about 3.875 would work here.). From the outline we get the following confidence intervals.. i. A Single Confidence Interval If we desire a single interval, we use the formula for the difference between two means when the variance is known. For example, if we want the difference between means of column 1 and column 2. 1 2 x1 x2 t n m s 2 1 1 , where s MSW and n m dfW . n1 n 2 ii. Scheffé Confidence Interval. If we desire intervals that will simultaneously be valid For a given confidence level for all possible intervals between column means, 1 1 . 1 2 x1 x2 m 1Fm1,nm s n n 2 1 iii. Bonferroni Confidence Interval If we only need k different intervals, use 1 2 x1 x2 t n m s 1 1 n1 n 2 2k iv. Tukey Confidence Interval This also applies to all possible differences. 1 2 x1 x2 q m,n m s 1 1 This gives rise to Tukey’s HSD n1 n 2 2 (Honestly Significant Difference) procedure. Two sample means x .1 and x .2 are significantly different if x.1 x.2 is greater than 1 2 qm, n m s 2 1 1 n1 n2 All of these could be used, but the Tukey is most popular. dfW n m 28. s MSW 20 , 3,28 2.95 and .05 . s 28 t .025 2.048 , F.05 1 1 1 1 20 5 2.236 n1 n 2 8 8 i. A Single Confidence Interval 1 2 x1 x2 t n m s 2 1 1 x1 x2 2.048 2.236 n1 n 2 x1 x2 4.579 . So the difference between two means is significant if it is above 4.579. 3 252solnF1 11/7/05 ii. Scheffé Confidence Interval. 1 2 x1 x2 m 1Fm1,nm s 1 1 n1 n 2 x1 x2 32.95 2.236 x1 x2 6.652 . So the difference between two means is significant if it is above 6.652. iii. Bonferroni Confidence Interval 1 2 x1 x2 t n m s 2k 1 1 - Similar to the single confidence n1 n 2 interval with a larger and hard-to find t. iv. Tukey Confidence Interval 1 2 x1 x2 q m,n m 1 1 1 2.236 x1 x2 3.90 2 2 n1 n 2 x1 x2 6.166 So the difference between two means is significant if it is above 6.166. This is Tukey’s HSD test and the text calls 6.166 a critical range. s Exercise 11.7 in 10th edition [for 9th and 8th editions, see next problem. The solution is repeated, edited, from the Instructor’s Solution Manual . 11.7 (a) The problem, which involves the Computer Anxiety Rating Scale gives the DF and SS columns below. You will compute MS by dividing SS by DF. You will then get a computed F statistic by dividing MSB by MSW. One-Way ANOVA Source DF SS MS F Statistic Between 5 3172 634.4 4.9567 ? Majors Within 166 21246 127.9880 Majors Total 171 24418 Note that F has 5 and 166 degrees of freedom. The problem says .05 . It also gives means and values of n for different majors as shown on the Tukey-Kramer table below. The solution given has a 5,166 2.2686 . This cannot be found exactly on a table. If you are working reference value of F of F.05 from my table go to page 11 (or 13). Look in the 5 column . The closest denominator df’s that you can 5,125 2.29 and F 5,200 2.26. You can guess that the find are 125 and 200. You should find F.05 .05 number you want is 2.27 or 2.28. Since your computed F is larger than either of these, reject your null hypothesis. (b) H0: A B C D E H1: At least one mean is different. F5,166 = 2.2686. Since F = 4.9567 > 2.2686, reject H0. There is enough evidence to conclude there is a difference in the mean computer anxiety experienced by different majors. To determine which of the means are significantly different from one another, we use the 5, 4.03 Tukey-Kramer procedure to establish the critical range: We use q.m,nm q.05 4 252solnF1 11/7/05 The ‘critical range’ for Marketing and Management is q m, n m s 2 1 1 127 .9880 4.03 n1 n 2 2 1 1 4.03 9.1857 4.933.0308 12 .214 19 11 The absolute difference between the means is 44.37 43.18 1.19 . Since 1.19 is less than the critical range, we cannot say that there is a significant difference between the means for marketing and management. Since there are six majors, the number of possible 6! 6 5 15 . Partial PHStat output for Tukey-Kramer multiple comparisons is C 26 4!2! 2.1 comparison appears below. Note that the absolute difference of 1.19, the standard error of 3.0308 and the critical value of 12.21 appear in the first row on the right side of the table. (c) Tukey Kramer Multiple Comparisons Sample Sample Group Mean Size Comparison Marketing 1 44.37 19 Group 1 to Group 2 Management 43.18 11 Group 1 to 2 Group 3 Other 3 42.21 14 Group 1 to Group 4 Finance 4 41.8 45 Group 1 to Group 5 Accounting 5 37.56 36 Group 1 to Group 6 MIS 6 42.21 47 Group 2 to Group 3 Group 2 to Group 4 Group 2 to Other Data Group 5 Group 2 to Level of 0.05 Group 6 significance Group 3 to Numerator d.f. 6 Group 4 Group 3 to Denominator 166 Group 5 d.f. Group 3 to MSW 127.988 Group 6 Group 4 to Q Statistic 4.03 Group 5 Group 4 to Group 6 Group 5 to Group 6 Absolute Std. Error Critical Difference of Difference Range Results 1.19 3.03079827 12.21 Means are not different 2.16 2.81764126 11.36 Means are not different 2.57 2.18865081 8.82 Means are not different 6.81 2.26841672 9.142 Means are not different 2.16 2.17478228 8.764 Means are not different 0.97 3.22314015 12.99 Means are not different 1.38 2.69067325 10.84 Means are not different 5.62 2.75594714 11.11 Means are not different 0.97 2.67940444 10.8 Means are not different 0.41 2.44807815 9.866 Means are not different 4.65 2.51964456 10.15 Means are not different 0 2.43568722 9.816 Means are not different 4.24 1.78877019 7.209 Means are not different 0.41 1.66843109 6.724 Means are not different 4.65 1.77177436 7.14 Means are not different The Tukey-Kramer procedure does not show any pair-wise significant difference. The inconsistency of the results obtained in (a) and (c) could possibly be due to the violations of any of the assumptions needed. Exercise [11.7 in 9th edition] (11.7 in 8th edition): The solution is repeated, heavily edited, from the Instructor’s Solution Manual . ni 520 100 . dfB m 1 5 1 4. n1 n 2 n3 n 4 n 5 20 . So n (a) dfT n 1 100 1 99. 5 252solnF1 11/7/05 H 0 : 1 2 3 4 5 where 1 = Wendy, 2 = McDonald, 3 = Checkers, 4 = Burger King, 5 = Long John Silver H1 : Not all j are equal where j = 1, 2, 3, 4, 5 Decision Rule: If p-value < 0.05, reject H0. One-Way ANOVA Source DF SS MS F Statistic p Value Between 4 6536 1634 12.51148545 3.24067E-08 Within 95 12407 130.6 Total 99 18943 4,95 2.70 . There is Since p-value is virtually 0, reject H0. Or note that Fcalc F.05 (b) sufficient evidence to conclude that there is a difference in the average drive-through time of the 5 chains. To determine which of the means are significantly different from one another, we use the 5,95 3.92 Tukey-Kramer procedure to establish the critical range: We use q.m,nm q.05 1 1 130 .6 1 1 3.92 10 .017 n n 2 20 20 2 1 2 Note that the means in this table are given by the problem statement. Partial PHStat output for Tukey-Kramer multiple comparison follows. The ‘critical range’ is q m,n m s Sample Sample Absolute Mean Size Comparison Difference Results 1 150 20 Group 1 to 17 Means are Group 2 different 2 167 20 Group 1 to 19 Means are Group 3 different 3 169 20 Group 1 to 21 Means are Group 4 different 4 171 20 Group 1 to 22 Means are Group 5 different 5 172 20 Group 2 to 2 Means are not Group 3 different Group 2 to 4 Means are not Group 4 different Group 2 to 5 Means are not Group 5 different Group 3 to 2 Means are not Group 4 different Group 3 to 3 Means are not Group 5 different Group 4 to 1 Means are not Group 5 different From the above PHStat output, the drive-through times are different in the pairs between Wendy and each of the other 4 chains. From the sample averages, Wendy has the shortest drive-through time at 150 seconds. Group (c) 6 252solnF1 11/7/05 Exercise 11.8(11.8 in 8th edition): The solution is repeated, edited almost beyond recognition, from the Instructor’s Solution Manual . We start with the problem as stated in the 9th edition. It is much more valuable to study that the answer in the 8th edition. H 0 : 1 2 3 H1 : Not all j are equal (a) where 1 = Experts, 2 = Readers, 3 = Darts where j = 1, 2, 3 You should be able to do the calculations below. Only the three columns of numbers were given to us. Sum 1 39.5 -1.1 -4.5 -8.0 Sample 2 -31.0 -20.7 -45.0 -73.3 3 39.0 31.9 14.1 5.4 25.9 + -170.0 + 90.4 nj 4+ x j Sum 4+ 4 6.475 -42.500 22.600 SS 1645.7 + 8787.4 + 2766.6 x 2j 41.925625 1806.250000 + 53 .7 2 ij 2 j .j 53 .7 4.475 x 12 13199 .7 x ij2 2358 .935625 x 510.760000 2 j 2 2 2 2 ij 12 n x nx 13199.7 12 4.475 12959.39 SSB n x nx 46.475 4 42.5 422.60 12 4.475 SST x 2 2 2 42358.935625 124.4752 = 9435.74253 – 240.3075 = 9195.44 Source Between Within Total SS DF 9195.44 2 3763.95 12959.39 9 11 MS 4597.72 F F.05 10.99 s F 2,9 4.26 H0 Column means equal 418.2167 Minitab output follows. Decision Rule: If p-value < 0.05, reject H0. Results for: 252CONTEST2001.mtw MTB > Retrieve "C:\Berenson\Data_Files-9th\Minitab\CONTEST2001.mtw". Retrieving worksheet from file: C:\Berenson\Data_Files9th\Minitab\CONTEST2001.mtw # Worksheet was saved on Fri Jan 18 2002 Results for: 252CONTEST2001.mtw MTB > AOVOneway c1 c2 c3. One-way ANOVA: Experts, Readers, Darts Analysis of Variance Source DF SS Factor 2 9195 Error 9 3764 Total 11 12959 MS 4598 418 F 10.99 P 0.004 7 252solnF1 11/7/05 Level Experts Readers Darts N 4 4 4 Pooled StDev = Mean 6.48 -42.50 22.60 StDev 22.20 22.82 15.53 20.45 Individual 95% CIs For Mean Based on Pooled StDev ---------+---------+---------+------(------*-----) (------*-----) (-----*------) ---------+---------+---------+-------35 0 35 Since p-value = 0.004 < 0.05, reject the null hypothesis. There is enough evidence to conclude that there is a significant difference in the average returns for the three categories. (b) To determine which of the means are significantly different from one another, we use the Tukey-Kramer procedure to establish the critical range: QU(c, n – c) = QU(3, 9) = 3.95 Data was rerun using Minitab. To use comparison procedures, the data had to be stacked with all data in column 5 and identifiers in column 6. The data is shown in the first data display. The command to start ANOVA is now ‘Oneway c5 c6;’ with the subcommand ‘Tukey 5.’ to do a Tukey procedure with a 5% significance level. ————— 11/5/2003 8:17:19 PM ———————————————————— Welcome to Minitab, press F1 for help. MTB > Retrieve "C:\Documents and Settings\RBOVE.WCUPANET\My Documents\Drive D\MINITAB\252CONTEST2001.mtw". Retrieving worksheet from file: C:\Documents and Settings\RBOVE.WCUPANET\My Documents\Drive D\MINITAB\252CONTEST2001.mtw # Worksheet was saved on Wed Nov 05 2003 Results for: 252CONTEST2001.mtw MTB > Stack c1 c2 c3 c5; SUBC> Subscripts c6; SUBC> UseNames. MTB > print c5 c6 #This stacks the data in column 5 with column names #in C6. Data Display #These are the original numbers given above. Row 1 2 3 4 5 6 7 8 9 10 11 12 C5 39.5 -1.1 -4.5 -8.0 -31.0 -20.7 -45.0 -73.3 39.0 31.9 14.1 5.4 C6 Experts Experts Experts Experts Readers Readers Readers Readers Darts Darts Darts Darts MTB > Oneway c5 c6; SUBC> Tukey 5. One-way ANOVA: C5 versus C6 #This is the same ANOVA as above. Analysis of Variance for C5 Source DF SS C6 2 9195 Error 9 3764 Total 11 12959 Level Darts Experts Readers N 4 4 4 Mean 22.60 6.48 -42.50 MS 4598 418 StDev 15.53 22.20 22.82 F 10.99 P 0.004 Individual 95% CIs For Mean Based on Pooled StDev ---------+---------+---------+------(-----*------) (------*-----) (------*-----) 8 252solnF1 11/7/05 Pooled StDev = ---------+---------+---------+-------35 0 35 20.45 Tukey's pairwise comparisons Family error rate = 0.0500 Individual error rate = 0.0209 Critical value = 3.95 Intervals for (column level mean) - (row level mean) Darts Experts Experts -24.26 56.51 Readers 24.71 105.49 (c) (d) 8.59 89.36 We now have 5% Tukey confidence intervals for the three means note that the interval for mean of Experts – mean of Darts is -24.26 to 56.51, which includes zero. Since the interval includes zero, the difference is not significant. At 5% level of significance, the Tukey Kramer multiple comparison test shows that there is enough evidence to conclude that Experts and Readers, and Readers and Darts differ in average return. The data collected are the returns of the selected stocks by the 3 categories not the amount of drops compared to the previous returns of the stocks. Even if average returns are concerned, the experts have the sample average return of 6.475% while the readers have a sample average return of –42.5%, and the stocks chosen using the darts have a sample average return of 22.6%. However, these differences in sample averages do not lead to the conclusion of significant differences between the Expert and Darts in population averages according to the result of the Tukey-Kramer multiple comparison procedure in part (b). H0: 12 22 32 H1: At least one variance is different. The procedure with the stacked data was continued by running Levene’s and Bartlett’s tests. Levene’s test is described in the text. The important thing to note here is the pvalues. MTB > %Vartest c5 c6; SUBC> Confidence 95.0. Executing from file: W:\wminitab13\MACROS\Vartest.MAC Macro is running ... please wait Test for Equal Variances Response Factors ConfLvl C5 C6 95.0000 Bonferroni confidence intervals for standard deviations Lower Sigma Upper N Factor Levels 7.8509 11.2208 11.5367 15.5300 22.1962 22.8209 84.464 120.720 124.118 4 4 4 Darts Experts Readers 9 252solnF1 11/7/05 Bartlett's Test (normal distribution) Test Statistic: 0.437 P-Value : 0.804 Levene's Test (any continuous distribution) Test Statistic: 0.101 P-Value : 0.905 Test for Equal Variances: C5 vs C6 Test for Equal Variances for C5 95% Confidence Intervals for Sigmas Factor Levels Darts Bartlett's Test Test Statistic: 0.437 P-Value : 0.804 Experts Levene's Test Test Statistic: 0.101 P-Value : 0.905 Readers 0 50 100 Since, in the Levene test, p-value = 0.905 > 0.05, do not reject H0. There is not enough evidence of a significant difference in the variation in the return for the three categories. This validates the use of ANOVA. Now let’s try the version of the problem in the 8 th edition: H 0 : 1 2 3 11.8 (a) H1 : Not all j are equal where 1 = Experts, 2 = Readers, 3 = Darts where j = 1, 2, 3 b) You should be able to do the calculations below. Only the three columns of numbers were given to us. Sample Sum 1 2 3 23.4 54.0 3.9 21.6 2.4 -21.8 -25.3 -45.9 -46.8 -45.9 -79.8 -96.2 256 .4 x ij Sum -26.2 + -69.3 + -160.9 nj 4+ 4+ 4 x j -6.55 -17.325 -40.225 SS 3761.0 + 11397.0 + 11935.0 x 2j 42.9025 300.1556 + 1618.0506 12 n 256 .4 21 .367 x 12 27093 .0 x ij2 1961 .1087 x 2 j 10 252solnF1 11/7/05 x nx 2709312 21.367 21614.59 SSB n x nx 4 6.55 4 17.325 440.225 12 21.350 SST 2 ij 2 2 2 j .j 2 2 2 2 2 4191.1087 1221.3672 = 7844.43 –5478.41 = 2366.02 Source SS Between DF 2366.02 2 MS 1183.01 F F.05 0.55 ns F 2,9 4.26 H0 Column means equal Within 19248.57 9 2138.73 Total 21614.59 11 Note that the differences between the means for this version are not significant. The data were run on Minitab. Decision Rule: If p-value < 0.05, reject H0. Results for: 252CONTEST.MTW MTB > print c1-c3 Data Display Row Experts Readers Darts 1 2 3 4 23.4 21.6 -25.3 -45.9 54.0 2.4 -45.9 -79.8 3.9 -21.8 -46.8 -96.2 One-way ANOVA: Experts, Readers, Darts Analysis of Variance Source DF SS Factor 2 2366 Error 9 19248 Total 11 21614 Level Experts Readers Darts N 4 4 4 Mean -6.55 -17.33 -40.23 Pooled StDev = MTB > Variable Experts Readers Darts Variable Experts Readers Darts (b) MS 1183 2139 StDev 34.59 58.30 42.67 46.25 F 0.55 P 0.594 Individual 95% CIs For Mean Based on Pooled StDev ----+---------+---------+---------+-(------------*------------) (------------*------------) (------------*------------) ----+---------+---------+---------+--80 -40 0 40 N 4 4 4 Mean -6.6 -17.3 -40.2 Median -1.8 -21.8 -34.3 TrMean -6.6 -17.3 -40.2 Minimum -45.9 -79.8 -96.2 Maximum 23.4 54.0 3.9 Q1 -40.8 -71.3 -83.9 Q3 23.0 41.1 -2.5 StDev 34.6 58.3 42.7 SE Mean 17.3 29.1 21.3 Since p-value = 0.594 > 0.05, do not reject the null. There is not enough evidence to conclude that there is a significant difference in the average returns for the three categories. From (a), none of the 3 categories differ in average return. Nevertheless, the whole thing was rerun with stacked data below. 11 252solnF1 11/7/05 (c) (d) The data collected are the returns of the selected stocks by the 3 categories not the amount of drops compared to the previous returns of the stocks. Even if average returns are concerned, the experts have the smallest sample average return of -6.55% while the readers have a sample average return of -17.325%, and the stocks chosen using the darts have a sample average return of -40.225%. However, these differences in sample averages do not lead to the conclusion of significant differences in population averages according to the results of the one-way ANOVA F test and the Tukey-Kramer multiple comparison procedure. To test at the 0.05 level of significance whether the variation within the groups is similar for all groups, we conduct a Levene's test for homogeneity of variances: H 0 : 12 22 32 where 1 = Experts, 2 = Readers, 3 = Darts H1 : Not all 2j are equal where j = 1, 2, 3 Since the version of Minitab available to the Authors of the 8th edition Instructor’s Solution Manual evidently did not have the Levene method available, they did the whole drill mentioned in the text and got the results below. Decision Rule: If p-value < 0.05, reject the null hypothesis. ANOVA Source of Variation Between Groups Within Groups SS df MS F P-value F crit 639.2517 2 319.6258 0.723278 0.511343 4.256492 3977.215 9 441.9128 Total 4616.467 11 Since p-value = 0.5113 > 0.05, do not reject the null hypothesis. There is not enough evidence to conclude that there is any difference in the variances. Minitab output for the unstacked version of the problem follows. ————— 11/5/2003 9:32:18 PM ———————————————————— Welcome to Minitab, press F1 for help. MTB > Retrieve "C:\Documents and Settings\RBOVE.WCUPANET\My Documents\Drive D\MINITAB\252CONTEST.MTW". Retrieving worksheet from file: C:\Documents and Settings\RBOVE.WCUPANET\My Documents\Drive D\MINITAB\252CONTEST.MTW # Worksheet was saved on Wed Nov 05 2003 Results for: 252CONTEST.MTW MTB > Stack c1 c2 c3 c5; SUBC> Subscripts c6; SUBC> UseNames. 12 252solnF1 11/7/05 MTB > print c5 c6 Data Display Row C5 C6 1 2 3 4 5 6 7 8 9 10 11 12 23.4 21.6 -25.3 -45.9 54.0 2.4 -45.9 -79.8 3.9 -21.8 -46.8 -96.2 Experts Experts Experts Experts Readers Readers Readers Readers Darts Darts Darts Darts MTB > Oneway c5 c6; SUBC> Tukey 5. One-way ANOVA: C5 versus C6 Analysis of Variance for C5 Source DF SS C6 2 2366 Error 9 19248 Total 11 21614 Level Darts Experts Readers N 4 4 4 Mean -40.23 -6.55 -17.33 Pooled StDev = MS 1183 2139 StDev 42.67 34.59 58.30 46.25 F 0.55 P 0.594 Individual 95% CIs For Mean Based on Pooled StDev ----+---------+---------+---------+-(------------*------------) (------------*------------) (------------*------------) ----+---------+---------+---------+--80 -40 0 40 Tukey's pairwise comparisons Family error rate = 0.0500 Individual error rate = 0.0209 Critical value = 3.95 Intervals for (column level mean) - (row level mean) Darts Experts -125.0 57.7 Readers -114.2 68.4 Experts -80.6 102.1 MTB > %Vartest c5 c6; SUBC> Confidence 95.0. Executing from file: W:\wminitab13\MACROS\Vartest.MAC Macro is running ... please wait Test for Equal Variances Response Factors C5 C6 13 252solnF1 11/7/05 ConfLvl 95.0000 Bonferroni confidence intervals for standard deviations Lower Sigma Upper N Factor Levels 21.5724 17.4863 29.4714 42.6729 34.5900 58.2980 232.089 188.128 317.071 4 4 4 Darts Experts Readers Bartlett's Test (normal distribution) Test Statistic: 0.728 P-Value : 0.695 Levene's Test (any continuous distribution) Test Statistic: 0.723 P-Value : 0.511 Test for Equal Variances: C5 vs C6 Test for Equal Variances for C5 95% Confidence Intervals for Sigmas Factor Levels Darts Bartlett's Test Test Statistic: 0.728 P-Value : 0.695 Experts Levene's Test Test Statistic: 0.723 P-Value : 0.511 Readers 0 100 200 300 Notice that this time all the confidence intervals include zero, verifying the fact that none of them are significant. Notice that the p-value for the Levine test is identical to the version that appeared in the Instructor’s Solution Manual . Also note that the p-value for the two tests are well out of the ‘reject’ zone. 14 252solnF1 11/7/05 This problem is worth studying since it is carried through to get all types of intervals except the Tukey interval. Exercise 14.22( From the McClave et. al. text): a) Make a diagram with an x axis marked off with numbers between 0 and 800. If you use different symbols for the three samples, you will notice that sample A has a very different distribution from samples B and C. the sample means also differ, so it looks like there is a significant difference between the population means. b) You should be able to do the calculations below. I will use 1, 2, and 3 instead of A, B and C. Only the three columns of numbers were given to us. Sum nj 1 250 150 275 100 475 600 150 800 325 230 Sample 2 100 150 75 200 55 80 110 160 132 233 3 80 125 20 186 52 92 88 141 76 200 Sum 3355 + 1295 + 1060 5710 10 + 10 + 10 30 n 129.5 106.0 x j 335.5 SS 1577275 + 196963 + 141590 x 2j 112560.25 16770.25 + 11236.00 2 ij 2 j .j x 1405665 .5 x 1915828 2 ij 2 j 2 2 2 2 ij 5710 190 .3333 x 30 x nx 1915828 30190.3333 829024.667 SSB n x nx 10335.5 10129.5 10106.0 30190.3333 SST x 2 2 2 101405665.0 30190.33332 =318861.667 Source SS DF MS F Between 318861.667 2 159430.833 8.436 Within Total 510163.000 829124.667 27 29 18894.926 F.05 F 2, 27 3.35 s H0 Column means equal c) We reject the null hypothesis and conclude that the different classes of firms have different audit costs. We use the p-value of .0014 in the printout in the text and we reject the null hypothesis because it is below the 5% significance level given in the problem. d) We assume that we have three random samples taken from Normal populations with similar variances. 15 252solnF1 11/7/05 e) I had asked that you compute all appropriate confidence intervals, using the formulas in the outline. But before we get to these contrasts, try an interval for a single mean. Note for all the below from the ANOVA 27 table. n m 27 , so if .05, tnm t.025 2.052 , n1 n 2 n3 10, and 2 s MSW 18894.926 137.459 . So for a single mean j x. j t n m 2 1 335 .5 2.052 1 106 .0 2.052 137 .459 10 137 .459 335 .5 89 .2 , 2 129 .5 2.052 137 .459 s . nj 129 .5 89 .2 , 10 106 .0 89 .2 . 10 The contrasts from the outline are quoted and filled in below. i. A single Confidence Interval If we desire a single interval, we use the formula for the difference between two means when the variance is known. For example, if we want the difference between means of column 1 and column 2. 1 2 x1 x2 t n m s 2 1 1 . so n1 n 2 1 2 335 .5 129 .5 2.052 137 .459 1 1 206 126 .1 10 10 2 3 129 .5 106 .0 126 .1 , 1 3 335 .5 106 .0 126 .1 ii. Scheffé Confidence Interval If we desire intervals that will simultaneously be valid for a given confidence level for all possible intervals 1 1 between column means, use 1 2 x1 x2 m 1Fm1,n m s . Note that Fm1,nm n n 1 2 is the F used in the ANOVA table F 2, 27 3.35 . So 1 1 1 2 335 .5 129 .5 23.35 137 .459 206 159 .1 , 10 10 2 3 129 .5 106 .0 159 .1 etc. Watch the values of n j , if they are not the same you have to recompute the error part. iii. Bonferroni Confidence Interval If we only need k different intervals, use 1 2 x1 x2 t n m s 2k 1 1 Ordinarily, you need a n1 n 2 computer to do the values of t that this requires. k is the number of intervals you intend to do. Assume that .03 and that you are doing 3 intervals. Then t nm t 27 2.771 . 1 2 335 .5 129 .5 2.771137 .459 2k .005 1 1 206 170 .343 etc. 10 10 16 252solnF1 11/7/05 Exercise, 14.24†: Another problem worth looking at. a) You should be able to do the calculations below. Only the three columns of numbers were given to us. .10 Sample 2 1.58 1.45 0.57 1.16 1.12 0.91 0.83 1 1.06 0.79 0.82 0.89 1.05 0.95 0.65 1.15 Sum Sum 1.12 0.43 3 0.29 0.06 0.44 0.61 0.55 0.43 0.51 0.10 0.34 0.53 0.06 0.09 0.17 0.60 0.17 8.48 + 8.05 + 4.95 9+ 8+ 15 nj 21 .48 x j 0.94222 1.00625 0.3300 21 .48 0.67125 x 32 SS 8.21660 + 9.22500 + 2.23692 19 .6292 x 2j 0.8878 1.0125 0.1089 Sum is not useful. 2 ij 2 j .j ij 32 n x nx 19.6792 320.67125 5.26075 SSB n x nx 90.94222 81.00625 150.3300 320.67125 SST x x 2 ij 2 2 2 2 2 2 2 90.8878 81.0125 150.1089 320.671 =3.30525 Source SS DF MS F F.10 24.51 F 2,29 2.50 s Between 3.30525 2 1.65263 Within Total 1.95550 5.26075 29 31 0.06743 H0 Column means equal Since the value of F we calculated is more than the table value, we reject the null hypothesis and conclude that the different chemicals have significantly different sorption rates. The SPSS output in the text has a pvalue of .000, so we reject the null hypothesis because it is below the 10% significance level given in the problem. b) We assume that we have three random samples taken from Normal populations with similar variances. c) There are a number of ways to check this. Certainly, computation of variances for the data might help. A fast look at standard deviations on my calculator seems to indicate that (Use the column sums of squares and means you already have.) s1 0.1863 , s2 0.4009 and s3 0.2075 . An F-test, using the first two of these made me a bit uncomfortable. A fast test here would be to get a table of the F-max distribution and compare the largest F ratio from the data to the table using a value of n equal to the largest number in an individual column. This is beyond the scope of this course. Note: Since this was written 2 methods for comparing variances were added to the course. 17 252solnF1 11/7/05 d) I had asked that you compute all appropriate confidence intervals, using the formulas in the outline. But before we get to these contrasts, try an interval for a single mean. Note for all the below from the ANOVA 29 table. n m 29, so if .10 , tn m t.005 1.699 , n1 9, n2 8, n3 15, and 2 s MSW 0.06743 . So for a single mean j x. j t n m 2 s . nj 0.06743 0.06743 0.94 0.15 , 2 1.00625 1.699 1.01 0.09 etc. 9 8 The contrasts from the outline are quoted and filled in below. i. A single Confidence Interval If we desire a single interval, we use the formula for the difference between two means when the variance is known. For example, if we want the difference between means of column 1 and column 2. 1 0.94222 1.699 1 2 x1 x2 t n m s 2 1 1 . so n1 n 2 1 9 1 8 1 2 0.94222 1.00625 1.699 0.06743 0.064 0.214 (Note that this shows no 1 1 significant difference), 1 3 0.94222 .33000 1.699 0.06743 0.612 0.186 9 15 1 1 (Significant) and 2 3 1.00625 .33000 1.699 0.06743 0.676 0.193 (Significant). 8 15 ii. Scheffé Confidence Interval If we desire intervals that will simultaneously be valid for a given confidence level for all possible intervals 1 1 between column means, use 1 2 x1 x2 m 1Fm1,n m s . Note that Fm1,nm n n 1 2 is the F used in the ANOVA table F 2,29 2.50 . So all we have to do to get this interval is to replace t n m t 29 1.699 by m 1F m 1, n m 22.50 2.236 in the individual intervals above. So 2 .005 1 9 1 8 1 2 0.94222 1.00625 2.236 0.06743 0.064 0.282 (Note that this shows no 1 1 significant difference), 1 3 0.94222 .33000 2.236 0.06743 0.612 0.245 9 15 1 1 (Significant) and 2 3 1.00625 .33000 2.236 0.06743 0.676 0.248 (Significant). 8 15 iii. Bonferroni Confidence Interval (Enough Already!) 18