Lect 11_EN

download Lect 11_EN

of 15

Transcript of Lect 11_EN

  • 7/30/2019 Lect 11_EN

    1/15

    Research Methodology: Tools

    Applied Data Analysis (with SPSS)

    Lecture 11: Repeated Measures ANOVA / ANCOVA

    May 2011

    Prof. Dr. Jrg Schwarz [email protected]

    MSc Business Administration

    Slide 2

    Contents

    Aims ___________________________________________________________________________________________________ 5Introduction _____________________________________________________________________________________________ 6Outline ________________________________________________________________________________________________ 13Concepts of repeated measures ANOVA / ANCOVA ___________________________________________________________ 14ANOVA with SPSS: Two detailed examples __________________________________________________________________ 19

  • 7/30/2019 Lect 11_EN

    2/15

    Slide 3

    Table of contents

    Aims ___________________________________________________________________________________________________ 5Aims of the lecture .................................................................................................................................................................................................5

    Introduction _____________________________________________________________________________________________ 6Example 1 ..............................................................................................................................................................................................................6Example 2 ..............................................................................................................................................................................................................9

    Outline ________________________________________________________________________________________________ 13Concepts of repeated measures ANOVA / ANCOVA ___________________________________________________________ 14Repeated measures ANOVA................................................................................................................................................................................14ANCOVA..............................................................................................................................................................................................................15Homogeneity of variances....................................................................................................................................................................................16

    Case of between-group ANOVA................................................................................................................................................................................................16Case of repeated measures ANOVA.........................................................................................................................................................................................17

    Covariate..............................................................................................................................................................................................................18

    Slide 4

    ANOVA with SPSS: Two detailed examples __________________________________________________________________ 19Repeated measures ANOVA (Dataset Diet1.sav) ................................................................................................................................................19

    SPSS Elements: .......................................................................................................................19How to define repeated measures.............................................................................................................................................................................................20SPSS Output ANOVA (Diet1) Mauchly's Test of Sphericity ...................................................................................................................................................21SPSS Output ANOVA (Diet1) Tests of Within-Subjects Effects.............................................................................................................................................22SPSS Output ANOVA (Diet1) Tests of Between-Subjects Effects .........................................................................................................................................23SPSS Output ANOVA (Diet1) Contrasts.................................................................................................................................................................................24SPSS Output ANOVA (Diet1) Multivariate Tests ....................................................................................................................................................................25

    ANCOVA (Dataset Diet2.sav)...............................................................................................................................................................................26SPSS Elements: ........................................................................................................................................26SPSS Output ANOVA (Diet2) Tests of Between-Subjects Effects .........................................................................................................................................27Homogeneity of regression........................................................................................................................................................................................................28Specify custom model................................................................................................................................................................................................................29Homogeneity of regression Output .........................................................................................................................................................................................30

  • 7/30/2019 Lect 11_EN

    3/15

    Slide 5

    Aims

    Aims of the lecture

    You will understand the key steps in conducting a repeated measures ANOVA / ANCOVA.

    You will understand the concept of repeated measures ANOVA / ANCOVA.

    You will understand the concept of homogeneity of variances.

    You will understand the concept of homogeneity of regression.

    You can conduct a repeated measures ANOVA / ANCOVA with SPSS.

    In particular, you will know how to

    interpret the output

    Mauchly's Test of sphericitycontrasts

    check of homogeneity of regression

    describe the output

    Slide 6

    Introduction

    Example 1 Repeated Measures ANOVA

    Medical research: Effect on body weight of a diet over 5 weeks. (Dataset: Diet1.sav)

    Typical questions

    Is the diet effective?

    Does it affect men and women differently?

    15 women

    15 men

    all

    Bodyweight[kg]

  • 7/30/2019 Lect 11_EN

    4/15

  • 7/30/2019 Lect 11_EN

    5/15

    Slide 9

    Example 2 ANCOVA

    Medical research: Effect of 3 different types of diet on body weight.

    Each type of diet was tested on a subsample of 5 randomly allocated overweight men.

    6

    7

    8

    9

    10

    11

    12

    13

    14

    15

    1 2 3

    Weightreduction[kg]

    Type of diet

    Dataset (Diet2.sav)

    Sample of n = 15 overweight men

    Variables for

    type of diet (1,2,3) (diet)

    weight reduction (kg) (reduction)

    weight before diet (kg) (weight)

    Typical questions

    Is there an impact of the type of diet?

    What is the influence of the initial weight?

    Slide 10

    The Result Impact of type of diet

    Significant ANOVA model ("Corrected Model" with p =.000).

    Significant variable diet

    There is a main effect of diet (types 1, 2, 3) on weight reduction (F(2, 12) = 20.812 p = .000).

    The value of Partial Eta Squared = .776 shows that 78% of the variation can be explained by the

    variable diet (all other variables fixed).

  • 7/30/2019 Lect 11_EN

    6/15

    Slide 11

    What is the influence of weight before diet?

    6

    7

    8

    9

    10

    11

    12

    13

    14

    15

    80 85 90 95 100 105 110 115

    It seems that "weight before diet" and

    "weight reduction" are related.

    The greater weight before diet, the more

    weight reduction is possible.

    This relation should also be modeled.

    "Weight before diet" is a metric variable.

    Enter the variable weightas a so called

    covariate into the model.

    Weightreduction[kg]

    Weight before diet [kg]

    Slide 12

    The Result Impact of type of diet plus weight before diet as covariate

    There is a main effect of diet (types 1, 2, 3) on weight reduction (F(2, 11) = 25.017 p = .000).

    The value of Partial Eta Squared = .820 shows that 82% of the variation can be explained by the

    variable diet (all other variables fixed).

    There is also a significant effect of the covariate weight (F(1, 11) = 6.659 p = .026).

    The constant ("Intercept") is no more significant (p = .542).This is due to the covariate which "binds" the variation of the intercept.

    The impact of diet is higher (2 = 82% vs. 2 = 78%) after controlling for the effects of weight

    before diet. The model is more "selective" respecting the main effect.

  • 7/30/2019 Lect 11_EN

    7/15

    Slide 13

    Outline

    Basic situation for repeated measures ANOVA

    Given: One dependent metric scaled variable which is measured multiple and many

    independent variables with categorical scales.

    Task: Find a relationship between the characteristics.

    Analysis of repeated measures ANOVA Statistical method is the same as in analysis of variance.

    The repeated measures model can test the main effects of within-subjects factors likesubsequent measurement times or different treatments.

    Basic situation for ANCOVA

    Given: One dependent variable with metric scale and many independent variables with categori-

    cal scales. Included is a metric scaled independent variable.

    Task: Find a relationship between the characteristics.

    Analysis of Covariance (ANCOVA)

    Controls for the effects of the metric scaled independent variable that covaries with thedependent variable.

    The metric scaled independent variable is called "covariate" or "control variable".

    Slide 14

    Concepts of repeated measures ANOVA / ANCOVA

    Repeated measures ANOVA

    Repeated measures

    General term for a study in which multiple measurements are applied to the same subject

    Can be several measurements over time or multiple treatments (e.g. Drug A, B, C)

    Advantages

    Source of variability between subjects is excluded from the experimental error

    Less participants are neededThis will be important if only a few subjects are available for the experiment

    Disadvantages

    Carry-over effects (outcome is influenced by previous treatment in an unwanted way) Example: Subject changes his opinion on certain issues.

    Order effects (outcome depends on the order of treatment)

    Example: The order "treatment A > B" is different to "treatment B > A"

  • 7/30/2019 Lect 11_EN

    8/15

    Slide 15

    ANCOVA

    Design of experiment

    At least one categorical variable (called factor) and one metric variable (called covariate)

    Blending of regression and ANOVA

    ANCOVA is used for different purposes

    Experiment: to control for factors that cannot be randomized but that can bemeasured on an metric scale

    Survey: to remove the effects of metric variables modifying the relationship of thecategorical independents to the dependent variable

    In general, the covariate is considered to be a control variable

    Advantages

    Irrelevant variability can be reduced ("controlled")

    Disadvantages If the covariate is correlated with the factors, then its inclusion will lead to an underestimation

    of the effect of the treatment factors

    Slide 16

    Homogeneity of variances

    Case of between-group ANOVA

    Back to the requirements of ANOVA in Lecture 10

    3. Homogeneity of variances

    Residuals (= Error) have constant variance

    Correction => weight variances

    Nurse example: Use homogeneity test

    p < .050 => no constant variance

    Weight with an appropriate variable

    Nurse example of lecture 10

    Homogeneity test "Options"

  • 7/30/2019 Lect 11_EN

    9/15

    Slide 17

    Case of repeated measures ANOVA

    Requirements of repeated measures ANOVA ("Compound Symmetry")

    1. Residuals (= Error) have constant variance

    => the same as in lecture 10 "3. Homogeneity of variances"

    2. Correlations between treatments are equal

    Assumption of sphericity ("Sphrizitt")

    Sphericity is a less restrictive form of compound symmetry

    Technical implementation: A test whether variances of the differences between each pair of

    treatment outcome are equal (=> at least three levels of treatment are needed)

    Example (Diet1)

    Calculate differences of scores between weeks 2 and 3,

    and weeks 3 and 4. Variances of the differences of the

    first pair must be the same as those of the second pair, etc.

    Sphericity is measured using Mauchly's Test.

    p > 0.05 => Sphericity is met

    p < 0.05 => Sphericity is violated

    If Mauchly's Test shows violation of sphericity, this may be compensated by an adjustment.

    66

    68

    70

    72

    74

    76

    78

    80

    82

    0 1 2 3 4 5 6

    Slide 18

    Covariate

    To control for metric variable x, introduce an additional (regression) term into the linear model.

    (In example 2 the metric variable x is weight before diet)

    gk g g gk gky y x= + + +

    g

    gk

    y grand mean

    effect of group g

    random term

    =

    =

    =

    g

    gk

    Additional term

    slope of covariate of group g

    x covariate

    =

    =

    Homogeneity of regression

    ANCOVA demands that homogeneity of regression exists.

    => For each level of the covariate xg., the slope g must be equal.

    The null hypothesis (H0) to verify is that all regression coefficients g have the same value

    The alternative hypothesis (HA) is that this is not the case

    H0: 1 = 2 = = G (= )

    HA: at least two coefficients are not equal

  • 7/30/2019 Lect 11_EN

    10/15

    Slide 19

    ANOVA with SPSS: Two detailed examples

    Repeated measures ANOVA (Dataset Diet1.sav)

    SPSS Elements:

    Variables week1, week2, week5 are repeated measures of body weight

    Define a name for a virtual Within-Subject-Factor

    For example "week"

    Enter the number of repetitions (called "levels")

    This example: 5 for week1 to week5

    Slide 20

    How to define repeated measures

    Allocate each week variable to its level

  • 7/30/2019 Lect 11_EN

    11/15

    Slide 21

    SPSS Output ANOVA (Diet1) Mauchly's Test of Sphericity

    Mauchly's Test is significant with p = .010 => sphericity is not given

    How to proceed if Mauchly's Test is significant: If Epsilon in Greenhouse-Geisser < .75 => correction as per Greenhouse-Geisser

    If Epsilon in Greenhouse-Geisser > .75 => correction as per Huynh-Feldt

    Slide 22

    SPSS Output ANOVA (Diet1) Tests of Within-Subjects Effects

    This is the first of two main tables

    The time elapsed (week 1, week 2, ) has a significant influence on body weight

    (F(2.792, 78.175) = 178.349 p = .000, correction as per Greenhouse-Geisser).

    The duration of the diet explains 86% of the variance.

    There is also an interaction of time elapsed and sex on body weight

    (F(2.792, 78.175) = 75.165 p = .000, correction as per Greenhouse-Geisser).

    The interaction term explains 73% of the variance.

  • 7/30/2019 Lect 11_EN

    12/15

  • 7/30/2019 Lect 11_EN

    13/15

    Slide 25

    SPSS Output ANOVA (Diet1) Multivariate Tests

    This is an alternative to the test of "within-subjects effects" if sphericity is not given.

    It is part of the MANOVA procedure, which makes no assumptions as regards sphericity.

    The Multivariate Tests table provides F tests of the within-subjects factor week

    and the interaction with the between-subjects grouping factor sex.

    Wilks' lambda is the most commonly reported

    Usually the same substantive conclusion emerges from any variant.

    Slide 26

    ANCOVA (Dataset Diet2.sav)

    SPSS Elements:

  • 7/30/2019 Lect 11_EN

    14/15

    Slide 27

    SPSS Output ANOVA (Diet2) Tests of Between-Subjects Effects

    There is a main effect of diet (types 1, 2, 3) on weight reduction (F(2, 11) = 25.017 p = .000).

    The value of Partial Eta Squared = .820 shows that 82% of the variation can be explained by the

    variable diet (all other variables fixed).

    There is also a significant effect of the covariate weight (F(1, 11) = 6.659 p = .026).

    The constant ("Intercept") is no more significant (p = .542).

    The covariate "binds" the variation due to the intercept.

    The impact of diet is higher (2 = 82% vs. 2 = 78% without covariate) after controlling for the ef-

    fects of weight before diet. The model is more "selective" respecting the main effect.

    Slide 28

    Homogeneity of regression

    Are the slopes g between weight before diet and weight reduction equal for all levels of diet?

    6

    7

    8

    9

    10

    11

    12

    13

    14

    15

    80 85 90 95 100 105 110 115

    6

    7

    8

    9

    10

    11

    12

    13

    14

    15

    80 85 90 95 100 105 110 115

    6

    7

    8

    9

    10

    11

    12

    13

    14

    15

    80 85 90 95 100 105 110 115

    Use a modification of the ANCOVA to quick-check whether the slopes g are equal.

    6

    7

    8

    9

    10

    11

    12

    13

    14

    15

    80 85 90 95 100 105 110 115

    all diets

    diet 1 diet 2 diet 3

  • 7/30/2019 Lect 11_EN

    15/15

    Slide 29

    Specify custom model

    Allocate main effects

    Allocate interaction term

    Slide 30

    Homogeneity of regression Output

    If the interaction term of a main factor and a covariate is significant,

    the assumption of homogeneity of regression is violated.

    Here the interaction term is not significant=> homogeneity of regression is given

    => the covariate weight(weight before diet) may be introduced into the model.