11 Comparison of Several Multivariate Means Shyh-Kang Jeng Department of Electrical Engineering/ Graduate Institute of Communication/ Graduate Institute of Networking and Multimedia
22 Paired Comparisons Measurements are recorded under different sets of conditions See if the responses differ significantly over these sets Two or more treatments can be administered to the same or similar experimental units Compare responses to assess the effects of the treatments
33 Example 6.1: Effluent Data from Two Labs
44 Single Response (Univariate) Case
55 Multivariate Extension: Notations
66 Result 6.1
77 Test of Hypotheses and Confidence Regions
88 Example 6.1: Check Measurements from Two Labs
99 Experiment Design for Paired Comparisons n Treatments 1 and 2 assigned at random Treatments 1 and 2 assigned at random Treatments 1 and 2 assigned at random Treatments 1 and 2 assigned at random
1010 Alternative View
1111 Repeated Measures Design for Comparing Measurements q treatments are compared with respect to a single response variable Each subject or experimental unit receives each treatment once over successive periods of time
1212 Example 6.2: Treatments in an Anesthetics Experiment 19 dogs were initially given the drug pentobarbitol followed by four treatments Halothane Present Absent CO 2 pressure LowHigh 12 34
1313 Example 6.2: Sleeping-Dog Data
1414 Contrast Matrix
1515 Test for Equality of Treatments in a Repeated Measures Design
1616 Example 6.2: Contrast Matrix
1717 Example 6.2: Test of Hypotheses
1818 Example 6.2: Simultaneous Confidence Intervals
1919 Comparing Mean Vectors from Two Populations Populations: Sets of experiment settings Without explicitly controlling for unit- to-unit variability, as in the paired comparison case Experimental units are randomly assigned to populations Applicable to a more general collection of experimental units
2020 Assumptions Concerning the Structure of Data
2121 Pooled Estimate of Population Covariance Matrix
2222 Result 6.2
2323 Proof of Result 6.2
2424 Wishart Distribution
2525 Test of Hypothesis
2626 Example 6.3: Comparison of Soaps Manufactured in Two Ways
2727 Example 6.3
2828 Result 6.3: Simultaneous Confidence Intervals
2929 Example 6.4: Electrical Usage of Homeowners with and without ACs
3030 Example 6.4: Electrical Usage of Homeowners with and without ACs
3131 Example 6.4: 95% Confidence Ellipse
3232 Bonferroni Simultaneous Confidence Intervals
3333 Result 6.4
3434 Proof of Result 6.4
3535 Remark
3636 Example 6.5
37 Multivariate Behrens-Fisher Problem Test H 0 : 1 - 2 =0 Population covariance matrices are unequal Sample sizes are not large Populations are multivariate normal Both sizes are greater than the number of variables 37
38 Approximation of T 2 Distribution 38
39 Confidence Region 39
40 Example 6.6 Example 6.4 data 40
4141 Example 6.10: Nursing Home Data Nursing homes can be classified by the owners: private (271), non-profit (138), government (107) Costs: nursing labor, dietary labor, plant operation and maintenance labor, housekeeping and laundry labor To investigate the effects of ownership on costs
4242 One-Way MANOVA
4343 Assumptions about the Data
4444 Univariate ANOVA
4545 Univariate ANOVA
4646 Univariate ANOVA
4747 Univariate ANOVA
4848 Concept of Degrees of Freedom
4949 Concept of Degrees of Freedom
5050 Examples 6.7 & 6.8
5151 MANOVA
5252 MANOVA
5353 MANOVA
5454 Distribution of Wilk’s Lambda
5555 Test of Hypothesis for Large Size
5656 Popular MANOVA Statistics Used in Statistical Packages
5757 Example 6.9
5858 Example 6.8
5959 Example 6.9
6060 Example 6.9
6161 Example 6.10: Nursing Home Data Nursing homes can be classified by the owners: private (271), non-profit (138), government (107) Costs: nursing labor, dietary labor, plant operation and maintenance labor, housekeeping and laundry labor To investigate the effects of ownership on costs
6262 Example 6.10
6363 Example 6.10
6464 Example 6.10
6565 Bonferroni Intervals for Treatment Effects
6666 Result 6.5: Bonferroni Intervals for Treatment Effects
6767 Example 6.11: Example 6.10 Data
68 Test for Equality of Covariance Matrices With g populations, null hypothesis H 0 : 1 = 2 =... = g = Assume multivariate normal populations Likelihood ratio statistic for testing H 0
69 Box’s M-Test
70 Example 6.12 Example nursing home data
7171 Example 6.13: Plastic Film Data
7272 Two-Way ANOVA
7373 Effect of Interactions
7474 Two-Way ANOVA
7575 Two-Way ANOVA
7676 Two-Way MANOVA
7777 Two-Way MANOVA
7878 Two-Way MANOVA
7979 Two-Way MANOVA
8080 Bonferroni Confidence Intervals
8181 Example 6.13: MANOVA Table
8282 Example 6.13: Interaction
8383 Example 6.13: Effects of Factors 1 & 2
8484 Profile Analysis A battery of p treatments (tests, questions, etc.) are administered to two or more group of subjects The question of equality of mean vectors is divided into several specific possibilities –Are the profiles parallel? –Are the profiles coincident? –Are the profiles level?
8585 Example 6.14: Love and Marriage Data
8686 Population Profile
8787 Profile Analysis
8888 Test for Parallel Profiles
8989 Test for Coincident Profiles
9090 Test for Level Profiles
9191 Example 6.14
9292 Example 6.14: Test for Parallel Profiles
9393 Example 6.14: Sample Profiles
9494 Example 6.14: Test for Coincident Profiles
9595 Example 6.15: Ulna Data, Control Group
9696 Example 6.15: Ulna Data, Treatment Group
9797 Comparison of Growth Curves
9898 Comparison of Growth Curves
9999 Example 6.15
Example 6.16: Comparing Multivariate and Univariate Tests
Example 6.14: Comparing Multivariate and Univariate Tests
Strategy for Multivariate Comparison of Treatments Try to identify outliers –Perform calculations with and without the outliers Perform a multivariate test of hypothesis Calculate the Bonferroni simultaneous confidence intervals –For all pairs of groups or treatments, and all characteristics
Importance of Experimental Design Differences could appear in only one of the many characteristics or a few treatment combinations Differences may become lost among all the inactive ones Best preventative is a good experimental design –Do not include too many other variables that are not expected to show differences