Multiple Comparisons (Exclude Computing All Treatments With a Control) new!

download Multiple Comparisons (Exclude Computing All Treatments With a Control) new!

of 36

Transcript of Multiple Comparisons (Exclude Computing All Treatments With a Control) new!

  • 8/3/2019 Multiple Comparisons (Exclude Computing All Treatments With a Control) new!

    1/36

    MULTIPLE COMPARISON TEST

  • 8/3/2019 Multiple Comparisons (Exclude Computing All Treatments With a Control) new!

    2/36

    Learning Objective

    Understand the definitions used

    in multiple comparison test.

    Determine which means differ,using the multiple comparison

    test if the null hypothesis is

    rejected in the Kruskal-Wallis

    non-parametric ANOVA.

  • 8/3/2019 Multiple Comparisons (Exclude Computing All Treatments With a Control) new!

    3/36

    Learning Outcome

    Make inference to related sample using

    multiple comparison test.

  • 8/3/2019 Multiple Comparisons (Exclude Computing All Treatments With a Control) new!

    4/36

    When we use multiplecomparison test?

    When a hypothesis-testing procedure

    such as Kruskal-Wallis test leads us to

    reject the null hypothesis. Thus to conclude that not all sampled

    populations are identical.

    This test use to answer the question whichpopulations are different from which

    others.

  • 8/3/2019 Multiple Comparisons (Exclude Computing All Treatments With a Control) new!

    5/36

    Example 6.2

    Cawson et al.* reported the data shown in Table

    6.10 on cortisol levels in three groups of patients

    who were delivered between 38 and 42 weeks

    gestation. Group 1 was studied before the onsetof labor at elective Caesarean section, group 2

    was studied at emergency Caesarean section

    during induced labor, and group 3 consisted of

    patients in whom spontaneous labor occurred

    and who were delivered either vaginally or by

    Caesarean section.

  • 8/3/2019 Multiple Comparisons (Exclude Computing All Treatments With a Control) new!

    6/36

    Example 6.2

    We wish to know whether these data provide

    sufficient evidence to indicate a difference in

    median cortisol levels among the three

    populations represented.

    *Cawson, M. J., Anne B. M. Anderson, A. C. Turnbull, andL. Lampe, Cortisol, Cortisone, and 11-Deoxycortisol

    Levels in Human Umbilical and Maternal Plasma in

    Relation to the Onset of Labour, J. Obstet. Gynaecol. Br.

    Commonw., 81(1974), 737-745.

  • 8/3/2019 Multiple Comparisons (Exclude Computing All Treatments With a Control) new!

    7/36

    Table 6.10

    Antecubital vein cortisol levels in three groups

    of patients studied at time of delivery

    Group 1 262 307 211 323 454 339 304 154 287 356

    Group 2 465 501 455 355 468 362

    Group 3 343 772 207 1048 838 687

  • 8/3/2019 Multiple Comparisons (Exclude Computing All Treatments With a Control) new!

    8/36

    Kruskal-Wallis One Way Analysis of Variance

    by Ranks

    Hypotheses

    Ho: The three populations represented by

    the data are identical.

    H1: The three populations do not have the

    same median.

  • 8/3/2019 Multiple Comparisons (Exclude Computing All Treatments With a Control) new!

    9/36

    Ranks corresponding to data of Table 6.10

    Group Ranks Rank sums

    1

    2

    3

    4 7 3 8 14 9 61 5 12

    16 18 15 11 17 13

    10 20 2 22 21 19

    R1= 69

    R2= 90

    R3= 94

  • 8/3/2019 Multiple Comparisons (Exclude Computing All Treatments With a Control) new!

    10/36

    Test Statistic

    H=

  • 8/3/2019 Multiple Comparisons (Exclude Computing All Treatments With a Control) new!

    11/36

    N =22

    R1= 69

    R2= 90

    R3= 94

    H 222.2)222(22

    99

    2

    99

    22

    22

    )222(22

    22 222 =+

    ++

    +=

  • 8/3/2019 Multiple Comparisons (Exclude Computing All Treatments With a Control) new!

    12/36

    Since the sample sizes all exceed 5, we use the

    chi-square table to determine whether the

    sample medians are significant different.

    critical value:

    k=3

    degree of freedom = k-1, 3-1=2

    by looking at table A.11, the critical value

    is 9.210 for =0.01.

  • 8/3/2019 Multiple Comparisons (Exclude Computing All Treatments With a Control) new!

    13/36

    H= 9.232

    critical value= 9.210

    H> critical value

    9.232 > 9.210

    Decision :

    We reject Ho at the level of significance.

    Conclusion :

    There is enough evident to support the claim that is the three

    populations represented by the data are not identical. We conclude

    that the medians of the populations represented are not all equal

    that is the median cortisol levels are not the same for all three typesofpatients.The P value is between 0.01 and 0.005.

  • 8/3/2019 Multiple Comparisons (Exclude Computing All Treatments With a Control) new!

    14/36

    As we reject Ho, we must continue with multiple

    comparison test to see which populations are different

    from which others.

    When we apply this multiple-comparison procedure, we

    use what is known as an experimentwise error rate.

    The experimentwise error rate, which represents a

    conservative approach in making multiple comparisons,

    holds the probability of making only correct decisions at 1-

    when the null hypothesis of no difference among

    populations is true.

  • 8/3/2019 Multiple Comparisons (Exclude Computing All Treatments With a Control) new!

    15/36

    STEPS:

    1) Obtain the mean of the ranks for each sample, and let Ri be the

    mean of the ranks of the ith sample and Rj be the mean of therank of thejth sample.

    2) Select an experimentwise error rate of, which we think of as an

    overall level of significance. Our choice of is determined in part by2 k, the number of

    samples involved, and is larger for largerk.

  • 8/3/2019 Multiple Comparisons (Exclude Computing All Treatments With a Control) new!

    16/36

    If we have ksamples, there will be a total of

    pairs of sample that can be compared apair of time.

    We usually select a value of larger than2those customarily encountered in single-

    comparison inference procedures-for

    example, 0.15, 0.20,or 0.25 depending on the

    size ofk.

    a

  • 8/3/2019 Multiple Comparisons (Exclude Computing All Treatments With a Control) new!

    17/36

    3) Find the value ofzin Table A.2 that has area to its

    right.

    4) Form the inequalityIfksamples are all of the different size

    (6.7)

    where Nis the number of observations in all samples

    combined. The probability that Inequality 6.7 holds for all

    pairs of means, when H0 is true, is at least 1- .2

    )2( kka

    ++

    jikka

    ji

    nn

    NNZRR

    22

    99

    )2(||

    )2(2

  • 8/3/2019 Multiple Comparisons (Exclude Computing All Treatments With a Control) new!

    18/36

    If the k samples are all of the same size

    (6.8)

    Any difference that is larger than the

    right- hand side of Inequality 6.7 (or Inequality

    6.8 if applicable) is declared significant at the

    level.

    2)2

    (||)2(

    2+

    NNZRR

    kk

    aji

  • 8/3/2019 Multiple Comparisons (Exclude Computing All Treatments With a Control) new!

    19/36

    Continue from Example 6.2

    We will apply multiple comparison test to see

    which populations are different from which

    others.

    1)Hypothesis

    H0 : 1 = 2 H0 : 1 = 3 H0 : 2 = 3

    H1 : 1 2 H1: 1 3 H1: 2 3

  • 8/3/2019 Multiple Comparisons (Exclude Computing All Treatments With a Control) new!

    20/36

    To make all possible comparisons in order to

    locate where the differences occur, let choose

    an error rate of =0.15.

    Samples involved are k=3,so that there will

    be

    So there are 3 comparisons to make.

  • 8/3/2019 Multiple Comparisons (Exclude Computing All Treatments With a Control) new!

    21/36

    From table A.2, we find the z with 0.025 of

    the area to its right to be 1.96.

    The mean of the rank for the threesamples:

    222.2)2(2

    22.2

    )2(==

    kka

    2.222

    222

    ==R

    99.992

    99

    2==R

    222

    222

    ==R

  • 8/3/2019 Multiple Comparisons (Exclude Computing All Treatments With a Control) new!

    22/36

    TABLE A.2

  • 8/3/2019 Multiple Comparisons (Exclude Computing All Treatments With a Control) new!

    23/36

    2)Test to compare group

    Compare groups 1 and 2

    8.1 > 6.57

    |222.2||| = ji RR

    2.2=

    ++=

    2

    2

    99

    2

    99

    )222(2222.2

    22.2=

    +

    +

    jikk

    a nn

    NNZ

    22

    99

    )2(

    )2(2

    C 1 d 3

  • 8/3/2019 Multiple Comparisons (Exclude Computing All Treatments With a Control) new!

    24/36

    Compare group 1 and 3

    8.77 > 6.57

    +

    +=

    2

    2

    99

    2

    99

    )222(2222.2

    |22.222.2||| = ji RR

    22.2=

    22.2=

    ++

    ji

    kk

    a nn

    NNZ

    22

    99

    )2(

    )2(

    2

    C 2 d 3

  • 8/3/2019 Multiple Comparisons (Exclude Computing All Treatments With a Control) new!

    25/36

    Compare group 2 and 3

    0.67 < 7.35

    |22.2222||| = ji RR

    22.2=

    22.2=

    ++

    jikka

    nn

    NNZ

    22

    99

    )2(

    )2(2

    +

    +=

    2

    2

    2

    2

    99

    )222(2222.2

  • 8/3/2019 Multiple Comparisons (Exclude Computing All Treatments With a Control) new!

    26/36

    3) Decision

    a) 8.10 > 6.57, reject the H0. This

    comparison is significant.

    b) 8.77 > 6.57, reject the H0. Thiscomparison is significant.

    c) 0.67 < 7.35 , do not reject the H0. Thiscomparison is not significant.

  • 8/3/2019 Multiple Comparisons (Exclude Computing All Treatments With a Control) new!

    27/36

    4) Conclusion :

    a)There is enough evident to support the claimthat there are different between median group 1

    and 2, we conclude that subjects for elective

    Caesarean section before the onset of labortend to have lower cortisol levels that women

    experiencing induced labor at emergency

    Caesarean section.

    b)There is enough evident to support the claim that there

  • 8/3/2019 Multiple Comparisons (Exclude Computing All Treatments With a Control) new!

    28/36

    b)There is enough evident to support the claim that there

    are different between median group 1 and 3. We

    conclude that the women in population 1 also tend to

    have lower cortisol values than the women inpopulation 3.

    c)There is not enough evident to support the claim that

    there are different between median group 2 and 3. We

    cannot conclude that the populations represented by

    samples 2 and 3 differ with respect to their median

    cortisol levels.

  • 8/3/2019 Multiple Comparisons (Exclude Computing All Treatments With a Control) new!

    29/36

    Exercise 6.27, page 2541)Four tour guides are employed at a state-operated

    tourist attraction of historical interest. Each guideconducts tours of exactly 20 persons each during

    peak season. Each tour is timed. A researcher with

    the state department of tourism conducted a study

    compared the tour guides with respect to efficiency.

    For each guide the researcher selected a simple

    random sample of 6 tours conducted during the

    month of June, and the required for each wasrecorded. The results are shown in Table 6.32.

  • 8/3/2019 Multiple Comparisons (Exclude Computing All Treatments With a Control) new!

    30/36

    Time required to conduct tours by fourtour guides

    Tour guide_________________________________A B C

    D______________________

    37 36 38 30

    31 33 38 34

    33 38 39 31

    30 36 36 30

    39 37 38 35

    31 39 37 34

  • 8/3/2019 Multiple Comparisons (Exclude Computing All Treatments With a Control) new!

    31/36

    Can the researcher conclude on the basis of these

    data that the four tour guides differ with respect to

    the average time it takes them to conduct a tour? Let

    =0.05, use the Kruskal-Wallis test to determine the

    P value and make all possible comparisons.

  • 8/3/2019 Multiple Comparisons (Exclude Computing All Treatments With a Control) new!

    32/36

    Exercise 6.12, page 249

    2) Ali and Sweeney* determined the protoporphyrin levels in15 normal,

    healthy laboratory workers and in 26 patients admitted with acute

    alcoholism. Their results appear in Table 6.12. Can we conclude from

    these data that the normal workers and the two groups of alcoholics differ

    with respect to average protoporphyrin level? Let =0.15,

    determine the P value and make all possible comparisons.

    *Ali, M. A. M., and G. Sweney, Erythrocyte Corproporphyrin andPhotoporphyrin in Ethanol-Induced Sideroblastic Erythropoiesis, Blood,

    43(1974), 291-295.

    Protoporphyrin level milligrams/100 ml RBC in

  • 8/3/2019 Multiple Comparisons (Exclude Computing All Treatments With a Control) new!

    33/36

    Protoporphyrin level, milligrams/100 ml RBC, inthree groups of subjects

    Normal 22 27 47 30 38 78 28 58 72 56 30 39 53 50 36

    Alcoholicswith ringsideroblastsin bone

    marrow

    78 172 286 82 453 513 174 915 84 153 780

    Alcoholicswithoutringsideroblasts

    in bonemarrow

    37 28 38 45 47 29 34 20 68 12 37 8 76 148 11

  • 8/3/2019 Multiple Comparisons (Exclude Computing All Treatments With a Control) new!

    34/36

    Exercise 6.13, page 249

    3) Torre et al.* recorded the changes in rat cerebral and extracerebral (platelet)

    serotomin (5-HT) after intraperitoneal administration of LSD-25 and 1-

    methyl-d-lysergic acid butanolamide (UML). They also took measurements

    on 11 controls. The results are shown in Table 6.13. Do these data provide

    sufficient evidence to indicate a difference among the three groups? Let

    =0.20, determine the P value and make all possible comparisons.

    *Torre, Michele, Filippo Bogetto, and Eugenio Torre , Effect of LSD-25 and 1-Methyl-d-Lysergic

    Acid Butanolamide on Rat Brain and Platelet Serotonin Levels, Psychopharmacologia,36(1974), 117-122.

    Brain serotonin (5-HT) nanograms per gram in

  • 8/3/2019 Multiple Comparisons (Exclude Computing All Treatments With a Control) new!

    35/36

    Brain serotonin (5-HT), nanograms per gram, inthree groups of mice

    Controls 340 340 356 386 386 402 402417 433 495 557

    LSD, 0.5 mg/kg 294 325 325 340 356 371 385402

    UML, 0.5 mg/kg 263 309 340 356 371 371 402417

  • 8/3/2019 Multiple Comparisons (Exclude Computing All Treatments With a Control) new!

    36/36

    Link for youtube

    http://youtu.be/ZGy8glJ9t8w

    http://youtu.be/ZGy8glJ9t8whttp://youtu.be/ZGy8glJ9t8w