Transcript Slide 1
Chapter 15 The Analysis of Variance 1 Copyright © 2005 Brooks/Cole, a division of Thomson Learning, Inc. A Problem A study was done on the survival time of patients with advanced cancer of the stomach, bronchus, colon, ovary or breast when treated with ascorbate1. In this study, the authors wanted to determine if the survival times differ based on the affected organ. 1 2 Cameron, E. and Pauling, L. (1978) Supplemental ascorbate in the supportive treatment of cancer: re-evaluation of prolongation of survival time in terminal human cancer. Proceedings of the National Academy of Science, USA, 75, 4538-4542. Copyright © 2005 Brooks/Cole, a division of Thomson Learning, Inc. A Problem A comparative dotplot of the survival times is shown below. Dotplot for Survival Time Cancer Type Stomach Ovary Colon Bronchus Breast 0 3 1000 2000 3000 Survival Time (in days) Copyright © 2005 Brooks/Cole, a division of Thomson Learning, Inc. A Problem The hypotheses used to answer the question of interest are H0: µstomach = µbronchus = µcolon = µovary = µbreast Ha: At least two of the µ’s are different The question is similar to ones encountered in chapter 11 where we looked at tests for the difference of means of two different variables. In this case we are interested in looking a more than two variable. 4 Copyright © 2005 Brooks/Cole, a division of Thomson Learning, Inc. Single-factor Analysis of Variance (ANOVA) A single-factor analysis of variance (ANOVA) problems involves a comparison of k population or treatment means µ1, µ2, … , µk. The objective is to test the hypotheses: H0: µ1 = µ2 = µ3 = … = µk Ha: At least two of the µ’s are different 5 Copyright © 2005 Brooks/Cole, a division of Thomson Learning, Inc. Single-factor Analysis of Variance (ANOVA) The analysis is based on k independently selected samples, one from each population or for each treatment. In the case of populations, a random sample from each population is selected independently of that from any other population. When comparing treatments, the experimental units (subjects or objects) that receive any particular treatment are chosen at random from those available for the experiment. 6 Copyright © 2005 Brooks/Cole, a division of Thomson Learning, Inc. Single-factor Analysis of Variance (ANOVA) A comparison of treatments based on independently selected experimental units is often referred to as a completely randomized design. 7 Copyright © 2005 Brooks/Cole, a division of Thomson Learning, Inc. Single-factor Analysis of Variance (ANOVA) Dotplots of Yield by Fertilizer (group means are indicated by lines) Yield 70 60 50 40 Type 1 Type 2 Type 3 Fertilizer Notice that in the above comparative dotplot, the differences in the treatment means is large relative to the variability within the samples. 8 Copyright © 2005 Brooks/Cole, a division of Thomson Learning, Inc. Single-factor Analysis of Variance (ANOVA) Dotplots of Price by Subject (group means are indicated by lines) Price 85 75 Statistics Psychology Economics Subject Business 65 Notice that in the above comparative dotplot, the differences in the treatment means is not easily understood relative to the sample variability. ANOVA techniques will allow us to determined if those differences are significant. 9 Copyright © 2005 Brooks/Cole, a division of Thomson Learning, Inc. ANOVA Notation k = number of populations or treatments being compared Population or treatment 1 2 … k Population or treatment mean µ1 µ2 … µk 10 Population or treatment variance 12 22 … k2 Sample size n1 n2 … nk Sample mean x1 x2 … xk Sample variance s12 s22 … sk2 Copyright © 2005 Brooks/Cole, a division of Thomson Learning, Inc. ANOVA Notation N = n1 + n2 + … + nk (Total number of observations in the data set) T = grand total = sum of all N observations n1x1 n2 x2 nk xk T x grand mean N 11 Copyright © 2005 Brooks/Cole, a division of Thomson Learning, Inc. Assumptions for ANOVA 1. Each of the k populations or treatments, the response distribution is normal. 2. 1 = 2 = … = k (The k normal distributions have identical standard deviations. 3. The observations in the sample from any particular one of the k populations or treatments are independent of one another. 4. When comparing population means, k random samples are selected independently of one another. When comparing treatment means, treatments are assigned at random to subjects or objects. 12 Copyright © 2005 Brooks/Cole, a division of Thomson Learning, Inc. Definitions A measure of disparity among the sample means is the treatment sum of squares, denoted by SSTr is given by SSTr n1 x1 x n2 x 2 x 2 2 nk xk x A measure of variation within the k samples, called error sum of squares and denoted by SSE is given by SSE n1 1 s12 n2 1 s22 13 nk 1 sk2 Copyright © 2005 Brooks/Cole, a division of Thomson Learning, Inc. 2 Definitions A mean square is a sum of squares divided by its df. In particular, mean square for SSTr treatments = MSTr = k 1 SSE mean square for error = MSE = Nk The error df comes from adding the df’s associated with each of the sample variances: (n1 - 1) + (n2 - 1) + …+ (nk - 1) = n 1 + n 2 … + nk - 1 - 1 - … - 1 = N - k 14 Copyright © 2005 Brooks/Cole, a division of Thomson Learning, Inc. Example Three filling machines are used by a bottler to fill 12 oz cans of soda. In an attempt to determine if the three machines are filling the cans to the same (mean) level, independent samples of cans filled by each were selected and the amounts of soda in the cans measured. The samples are given below. Machine 1 12.033 11.98512.009 12.009 12.033 12.025 12.054 12.050 Machine 2 12.031 11.98511.99811.992 11.98512.027 11.987 15 Machine 3 12.034 12.001 12.021 12.021 12.020 12.038 12.029 12.058 12.011 Copyright © 2005 Brooks/Cole, a division of Thomson Learning, Inc. Example n1 8, x1 12.0248, s1 0.02301 n2 7, x2 12.0007, s2 0.01989 n3 9, x3 12.0259, s3 0.01650 x 12.018167 SSTr n1 x1 x n2 x 2 x 2 2 nk xk x 2 8(0.0065833)2 7(-0.0174524)2 9(0.0077222)2 0.000334672+0.00213210+0.00053669 0.00301552 16 Copyright © 2005 Brooks/Cole, a division of Thomson Learning, Inc. Example n1 8, x1 12.0248, s1 0.02301 n2 7, x2 12.0007, s2 0.01989 n3 9, x3 12.0259, s3 0.01650 x 12.018167 SSE n1 1 s12 n2 1 s22 nk 1 sk2 7(0.0230078)2 6(0.0198890)2 8(0.01649579)2 0.0037055 0.0023734 0.0021769 0.00825582 17 Copyright © 2005 Brooks/Cole, a division of Thomson Learning, Inc. Example n1 8, x1 12.0248, s1 0.02301 n2 7, x2 12.0007, s2 0.01989 n3 9, x3 12.0259, s3 0.01650 x 12.018167 SSTr mean square for treatments = MSTr = k 1 SSTr 0.00301552 MSTr 0.0015078 k 1 3 1 SSE mean square for error = MSE = Nk SSE 0.0082579 MSE 0.00039313 Nk 24 3 18 Copyright © 2005 Brooks/Cole, a division of Thomson Learning, Inc. Comments Both MSTr and MSE are quantities that are calculated from sample data. As such, both MSTr and MSE are statistics and have sampling distributions. More specifically, when H0 is true, µMSTr = µMSE. However, when H0 is false, µMSTr µMSE and the greater the differences among the m’s, the larger µMSTr will be relative to µMSE. 19 Copyright © 2005 Brooks/Cole, a division of Thomson Learning, Inc. The Single-Factor ANOVA F Test Null hypothesis: H0: µ1 = µ2 = µ3 = … = µk Alternate hypothesis: At least two of the µ’s are different Test Statistic: 20 MSTr F MSE Copyright © 2005 Brooks/Cole, a division of Thomson Learning, Inc. The Single-Factor ANOVA F Test When H0 is true and the ANOVA assumptions are reasonable, F has an F distribution with df1 = k - 1 and df2 = N - k. Values of F more contradictory to H0 than what was calculated are values even farther out in the upper tail, so the P-value is the area captured in the upper tail of the corresponding F curve. 21 Copyright © 2005 Brooks/Cole, a division of Thomson Learning, Inc. Example Consider the earlier example involving the three filling machines. Machine 1 12.033 12.025 Machine 2 12.031 12.027 Machine 3 12.034 12.020 22 11.985 12.054 12.009 12.050 12.009 12.033 11.985 11.987 11.998 11.992 11.985 12.021 12.029 12.038 12.011 12.058 12.021 12.001 Copyright © 2005 Brooks/Cole, a division of Thomson Learning, Inc. Example n1 8, x1 12.0248, s1 0.02301 n2 7, x2 12.0007, s2 0.01989 n3 9, x3 12.0259, s3 0.01650 x 12.018167 23 SSTr 0.00301552 SSE 0.00825582 MSTr 0.0015078 MSE 0.00039313 Copyright © 2005 Brooks/Cole, a division of Thomson Learning, Inc. Example 1. Let µ1, µ2 and µ3 denote the true mean amount of soda in the cans filled by machines 1, 2 and 3, respectively. 2. H0: µ1 = µ2 = µ3 3. Ha: At least two among are µ1, µ2 and µ3 different 4. Significance level: = 0.01 5. Test statistic: F 24 MSTr MSE Copyright © 2005 Brooks/Cole, a division of Thomson Learning, Inc. Example 6. Looking at the comparative dotplot, it seems reasonable to assume that the distributions have the same ’s. We shall look at the normality assumption on the next slide.* Dotplot for Fill Machine Machine 3 Machine 2 Machine 1 11.99 12.00 12.01 12.02 12.03 12.04 12.05 12.06 Fill *When the sample sizes are large, we can make judgments about both the equality of the standard deviations and the normality of the underlying populations with a comparative boxplot. 25 Copyright © 2005 Brooks/Cole, a division of Thomson Learning, Inc. Example 6. Looking at normal plots for the samples, it certainly appears reasonable to assume that the samples from Machine’s 1 and 2 are samples from normal distributions. Unfortunately, the normal plot for the sample from Machine 2 does not appear to be a sample from a normal population. So as to have a computational example, we shall continue and finish the test, treating the result with a “grain of salt.” 26 Copyright © 2005 Brooks/Cole, a division of Thomson Learning, Inc. Example 7. Computation: n1 8, x1 12.0248, s1 0.02301 n2 7, x2 12.0007, s2 0.01989 n3 9, x3 12.0259, s3 0.01650 x 12.018167 SSTr 0.00301552 SSE 0.00825582 MSTr 0.0015078 MSE 0.00039313 N n1 n2 n3 8 7 9 24, k 3 MSTr 0.0015078 F 3.835 MSE 0.00039313 df1 treatment df k 1 3 1 2 27 df2 error df N k 24 3 21 Copyright © 2005 Brooks/Cole, a division of Thomson Learning, Inc. Example 8. P-value: Recall MSTr 0.0015078 3.835 MSE 0.00039313 df1 treatment df k 1 3 1 2 F df2 error df N k 24 3 21 From the F table with numerator df1 = 2 and denominator df2 = 21 we can see that 0.025 < P-value < 0.05 (Minitab reports this value to be 0.038 28 dfden / dfnum 21 2 0.100 0.050 0.025 0.010 0.001 2.57 3.47 4.42 5.78 9.77 3.835 Copyright © 2005 Brooks/Cole, a division of Thomson Learning, Inc. Example 9. Conclusion: Since P-value > = 0.01, we fail to reject H0. We are unable to show that the mean fills are different and conclude that the differences in the mean fills of the machines show no statistically significant differences at the 1% level of significance. 29 Copyright © 2005 Brooks/Cole, a division of Thomson Learning, Inc. Total Sum of Squares Total sum of squares, denoted by SSTo, is given by SSTo (x x)2 all N obs. with associated df = N - 1. The relationship between the three sums of squares is SSTo = SSTr + SSE which is often called the fundamental identity for single-factor ANOVA. Informally this relation is expressed as Total variation = Explained variation + Unexplained variation 30 Copyright © 2005 Brooks/Cole, a division of Thomson Learning, Inc. Single-factor ANOVA Table The following is a fairly standard way of presenting the important calculations from an single-factor ANOVA. The output from most statistical packages will contain an additional column giving the P-value. 31 Copyright © 2005 Brooks/Cole, a division of Thomson Learning, Inc. Single-factor ANOVA Table The ANOVA table supplied by Minitab One-way ANOVA: Fills versus Machine Analysis of Variance for Fills Source DF SS MS Machine 2 0.003016 0.001508 Error 21 0.008256 0.000393 Total 23 0.011271 32 F 3.84 P 0.038 Copyright © 2005 Brooks/Cole, a division of Thomson Learning, Inc. Another Example A food company produces 4 different brands of salsa. In order to determine if the four brands had the same sodium levels, 10 bottles of each Brand were randomly (and independently) obtained and the sodium content in milligrams (mg) per tablespoon serving was measured. The sample data are given on the next slide. Use the data to perform an appropriate hypothesis test at the 0.05 level of significance. 33 Copyright © 2005 Brooks/Cole, a division of Thomson Learning, Inc. Another Example Brand A 43.85 44.30 45.69 47.13 43.35 45.59 45.92 44.89 43.69 44.59 Brand B 42.50 45.63 44.98 43.74 44.95 42.99 44.95 45.93 45.54 44.70 Brand C 45.84 48.74 49.25 47.30 46.41 46.35 46.31 46.93 48.30 45.13 Brand D 43.81 44.77 43.52 44.63 44.84 46.30 46.68 47.55 44.24 45.46 34 Copyright © 2005 Brooks/Cole, a division of Thomson Learning, Inc. Another Example 1. Let µ1, µ2 , µ3 and µ4 denote the true mean sodium content per tablespoon in each of the brands respectively. 2. H0: µ1 = µ2 = µ3 = µ4 3. Ha: At least two among are µ1, µ2, µ3 and µ4 are different 4. Significance level: = 0.05 MSTr 5. Test statistic: F MSE 35 Copyright © 2005 Brooks/Cole, a division of Thomson Learning, Inc. Another Example 6. Looking at the following comparative boxplot, it seems reasonable to assume that the distributions have the equal ’s as well as the samples being samples from normal distributions. Boxplots of Brand A - Brand D (means are indicated by solid circles) 49 48 47 46 45 44 43 36 Brand D Brand C Brand B Brand A 42 Copyright © 2005 Brooks/Cole, a division of Thomson Learning, Inc. Example 7. Computation: Brand k Brand A 10 Brand B 10 Brand C 10 Brand D 10 xi 44.900 44.591 47.056 45.180 si 1.180 1.148 1.331 1.304 x 45.432 SSTr n1(x1 x)2 n2 (x 2 x)2 n3 (x 3 x)2 n4 (x 4 x)2 10(44.900 45.432)2 10(44.591 45.432)2 10(47.056 45.432)2 10(45.180 45.432)2 36.912 Treatment df = k - 1 = 4 - 1 = 3 37 Copyright © 2005 Brooks/Cole, a division of Thomson Learning, Inc. Example 7. Computation (continued): SSE n1 1 s12 n2 1 s22 n3 1 s32 n4 1 s 42 9(1.180)2 9(1.148)2 9(1.331)2 9(1.304)2 55.627 Error df = N - k = 40 - 4 = 36 SSTr F 38 36.912 MSTr dfSSTr 3 12.304 7.963 SSE 55.627 MSE 1.5452 dfSSE 36 Copyright © 2005 Brooks/Cole, a division of Thomson Learning, Inc. Example 8. P-value: F = 7.96 with dfnumerator= 3 and dfdenominator= 36 7.96 Using df = 30 we find P-value < 0.001 39 Copyright © 2005 Brooks/Cole, a division of Thomson Learning, Inc. Example 9. Conclusion: Since P-value < = 0.001, we reject H0. We can conclude that the mean sodium content is different for at least two of the Brands. We need to learn how to interpret the results and will spend some time on developing techniques to describe the differences among the µ’s. 40 Copyright © 2005 Brooks/Cole, a division of Thomson Learning, Inc. Multiple Comparisons A multiple comparison procedure is a method for identifying differences among the µ’s once the hypothesis of overall equality (H0) has been rejected. The technique we will present is based on computing confidence intervals for difference of means for the pairs. Specifically, if k populations or treatments are studied, we would create k(k-1)/2 differences. (i.e., with 3 treatments one would generate confidence intervals for µ1 - µ2, µ1 - µ3 and µ2 - µ3.) Notice that it is only necessary to look at a confidence interval for µ1 - µ2 to see if µ1 and µ2 differ. 41 Copyright © 2005 Brooks/Cole, a division of Thomson Learning, Inc. The Tukey-Kramer Multiple Comparison Procedure When there are k populations or treatments being compared, k(k-1)/2 confidence intervals must be computed. If we denote the relevant Studentized range critical value by q, the intervals are as follows: MSE 1 1 For mi - mj: (mi m j ) q 2 ni n j Two means are judged to differ significantly if the corresponding interval does not include zero. 42 Copyright © 2005 Brooks/Cole, a division of Thomson Learning, Inc. The Tukey-Kramer Multiple Comparison Procedure When all of the sample sizes are the same, we denote n by n = n1 = n2 = n3 = … = nk, and the confidence intervals (for µi - µj) simplify to MSE ( mi m j ) q n 43 Copyright © 2005 Brooks/Cole, a division of Thomson Learning, Inc. Example (continued) Continuing with example dealing with the sodium content for the four Brands of salsa we shall compute the Tukey-Kramer 95% TukeyKramer confidence intervals for µA - µB, µA - µC, µA - µD, µB - µC, µB - µD and µC - µD. 55.627 1.5452, n nA nB nC nD 10 36 Interpolating from the table q 3.81 i.e. 60% of the way from 3.85 to 3.79 MSE MSE 1.5452 q 3.81 1.498 n 10 44 Copyright © 2005 Brooks/Cole, a division of Thomson Learning, Inc. Example (continued) 95% Confidence 95% Confidence Interval Limits Difference mA - mB mA - mC mA - mD mB - mC mB - mD mC - mD 45 0.309 ± 1.498 (-1.189, 1.807) -2.156 ± 1.498 (-3.654, -0.658) -0.280 ± 1.498 (-1.778, 1.218) -2.465 ± 1.498 (-3.963, -0.967) -0.589 ± 1.498 (-2.087, 0.909) 1.876 ± 1.498 (0.378, 3.374) Notice that the confidence intervals for µA – µB, µA – µC and µC – µD do not contain 0 so we can infer that the mean sodium content for Brands C is different from Brands A, B and D. Copyright © 2005 Brooks/Cole, a division of Thomson Learning, Inc. Example (continued) We also illustrate the differences with the following listing of the sample means in increasing order with lines underneath those blocks of means that are indistinguishable. Brand B Brand A Brand D Brand C 44.591 44.900 45.180 47.056 Notice that the confidence interval for µA – µC, µB – µC, and µC – µD do not contain 0 so we can infer that the mean sodium content for Brand C and all others differ. 46 Copyright © 2005 Brooks/Cole, a division of Thomson Learning, Inc. Minitab Output for Example One-way ANOVA: Sodium versus Brand Analysis of Variance for Sodium Source DF SS MS Brand 3 36.91 12.30 Error 36 55.63 1.55 Total 39 92.54 Level Brand Brand Brand Brand A B C D N 10 10 10 10 Pooled StDev = 47 Mean 44.900 44.591 47.056 45.180 1.243 StDev 1.180 1.148 1.331 1.304 F 7.96 P 0.000 Individual 95% CIs For Mean Based on Pooled StDev ------+---------+---------+---------+ (-----*------) (------*-----) (------*------) (------*-----) ------+---------+---------+---------+ 44.4 45.6 46.8 48.0 Copyright © 2005 Brooks/Cole, a division of Thomson Learning, Inc. Minitab Output for Example Tukey's pairwise comparisons Family error rate = 0.0500 Individual error rate = 0.0107 Critical value = 3.81 Intervals for (column level mean) - (row level mean) Brand A Brand B Brand B -1.189 1.807 Brand C -3.654 -0.658 -3.963 -0.967 Brand D -1.778 1.218 -2.087 0.909 48 Brand C 0.378 3.374 Copyright © 2005 Brooks/Cole, a division of Thomson Learning, Inc. Simultaneous Confidence Level The Tukey-Kramer intervals are created in a manner that controls the simultaneous confidence level. For example at the 95% level, if the procedure is used repeatedly on many different data sets, in the long run only about 5% of the time would at least one of the intervals not include that value of what it is estimating. We then talk about the family error rate being 5% which is the maximum probability of one or more of the confidence intervals of the differences of mean not containing the true difference of mean. 49 Copyright © 2005 Brooks/Cole, a division of Thomson Learning, Inc.