SlideShare una empresa de Scribd logo
1 de 32
Descargar para leer sin conexión
2009 German Stata Users Group Meeting

Performing within and between analysis
(WABA) in Stata
Sven-Oliver Spieß
Outline



Introduction
Where are constructs in hierarchical data really associated?



Within and between analysis
 The basic idea
 Simple ANOVA
 Partitioning of correlations
 Graphical demonstration



wabacorr.ado in Stata

Sven-Oliver Spieß

24.06.2009 ║ Page 2
24.06.2009
Introduction



Many names for one common problem:
Fallacy of composition, ecological fallacy, atomistic fallacy, individualistic
fallacy, Simpson’s paradox, …
 Fallacies of the wrong level



Therefore global correlations potentially misleading
 E.g. Corr(Job Satisfaction, Commitment) = .72
 But at which level is the association?
Individuals? Work groups? Departments?



Particularly problematic in applied settings
 No simple random samples
 Interventions

Sven-Oliver Spieß

24.06.2009 ║ Page 3
24.06.2009
Introduction


rJob Satisfaction,Commitment = .72

ρ?

Branch

ρ?

Department

ρ?

Employee
Observations

Job Satisfaction


Commitment

At which level is the association?

Sven-Oliver Spieß

24.06.2009 ║ Page 4
24.06.2009
Outline



Introduction
Where are constructs in hierarchical data really associated?



Within and between analysis
 The basic idea
 Simple ANOVA
 Partitioning of correlations
 Graphical illustration



wabacorr.ado in Stata

Sven-Oliver Spieß

24.06.2009 ║ Page 5
24.06.2009
The basic idea


The idea:
Let’s split up the total correlation into a component within the groups
and another component between the groups

 similar to idea behind analysis of variance (ANOVA)


Simply needs to be adjusted to correlations



Data prerequisites:
variables in question must be metric and levels must be nested

Sven-Oliver Spieß

24.06.2009 ║ Page 6
24.06.2009
Simple ANOVA

8



Xij = (Xij – μ•j) + μ•j



Random Values

6

SSTotal = SSError + SSGroup
(Xij – μ••)² = (Xij – μ•j)² + (μ•j – μ••)²

4

Grandmean

2



η² = SSGroup / SSTotal

==> η² between measure

Sven-Oliver Spieß

1

2

3
Group

24.06.2009 ║ Page 7
24.06.2009

4
Partitioning of correlations


Adjusted to correlations:
rxy = ηBx * ηBy * rBxy + ηWx * ηWy * rWxy



rxy = CB + CW
 ηB = corr(μ•j,Xij)
 ηW = corr[(Xij – μ•j),Xij]



Central question: is between or within component (i.e. higher or lower
level, or both) of total correlation more important?

Sven-Oliver Spieß

24.06.2009 ║ Page 8
24.06.2009
Partitioning of correlations


Typical procedure in 3 steps:
1.

Univariate comparison of the
within and between variances

rxy = ηBx * ηBy * rBxy + ηWx * ηWy * rWxy

2.

Bivariate comparison of the
within and between
correlations

rxy = ηBx * ηBy * rBxy + ηWx * ηWy * rWxy

3.

Summary judgment on the
importance of the within and
between components for the
total correlation

rxy = CB + CW

Sven-Oliver Spieß

24.06.2009 ║ Page 9
24.06.2009
Partitioning of correlations

4 possible outcomes/inductions:
1. Parts
 lower level/within
2. Wholes
 higher level/between
3. Equivocal
 meaningful association at both levels
4. Inexplicable
 noise

Sven-Oliver Spieß

24.06.2009 ║ Page 10
24.06.2009
Graphical illustration: Step 1
40

WABA1: Scatterplot Negotiation
Y=

30

Raw Scores

20
Y
10

0

-10
-10

Sven-Oliver Spieß

0

10
20
Raw Scores

30

40

24.06.2009 ║ Page 11
24.06.2009
Graphical illustration: Step 1 (ηB)
40

WABA1: Scatterplot Negotiation
Y=

30

Raw Scores

20

xj

Y
10

Dyad 2
0

-10
-10

Sven-Oliver Spieß

0

10
20
Raw Scores

30

40

24.06.2009 ║ Page 12
24.06.2009
Graphical illustration: Step 1 (ηB)
40

WABA1: Scatterplot Negotiation
Y=

30

Raw Scores
Group Means

20

xj

Y
10

Dyad 2
0

-10
-10

Sven-Oliver Spieß

0

10
20
Raw Scores

30

40

24.06.2009 ║ Page 13
24.06.2009
Graphical illustration: Step 1 (ηB)
40

WABA1: Scatterplot Negotiation
Y=

30

Raw Scores
Group Means

20
Y
10

20-16.5
0

-10
-10

Sven-Oliver Spieß

0

10
20
Raw Scores

30

40

24.06.2009 ║ Page 14
24.06.2009
Graphical illustration: Step 1 (ηW)
WABA1: Scatterplot Negotiation

40

Y=
30

Raw Scores
Group Means

20
Deviations from Means

Y
10

0

-10
-10

Sven-Oliver Spieß

0

10
20
Raw Scores

30

40

24.06.2009 ║ Page 15
24.06.2009
Graphical illustration: Step 1
WABA1: Scatterplot Negotiation

40

Y=
30

Raw Scores
Group Means

20
Deviations from Means

Y
10

Linear Fit:

Parts
Means

0

Deviations
-10
-10

Sven-Oliver Spieß

0

10
20
Raw Scores

30

40

24.06.2009 ║ Page 16
24.06.2009
Graphical illustration: Step 1
WABA1: Scatterplot Performance

40

Y=
30

Raw Scores
Group Means

20
Deviations from Means

Y
10

Linear Fit:

Means

0

Deviations
-10
-10

Sven-Oliver Spieß

0

10
20
Raw Scores

30

40

24.06.2009 ║ Page 17
24.06.2009
Graphical illustration: Step 2

20
10
0

Parts
-10

Negotiation

30

40

Scatterplot Negotiation over Satisfaction

0

10

20
Satisfaction

Raw Scores
Group Means
Deviations from Means

Sven-Oliver Spieß

30

40

Fitted Scores
Fitted Means
Fitted Deviations

24.06.2009 ║ Page 18
24.06.2009
Graphical illustration: Step 2

20
10
-10

0

Performance

30

40

Scatterplot Performance over Taskclarity

0

10

20
Taskclarity

Raw Scores
Group Means
Deviations from Means

Sven-Oliver Spieß

30

40

Fitted Scores
Fitted Means
Fitted Deviations

24.06.2009 ║ Page 19
24.06.2009
Outline



Introduction
Where are constructs in hierarchical data really associated?



Within and between analysis
 The basic idea
 Simple ANOVA
 Partitioning of correlations
 Graphical illustration



wabacorr.ado in Stata

Sven-Oliver Spieß

24.06.2009 ║ Page 20
24.06.2009
wabacorr.ado in Stata


General syntax:
wabacorr varlist [if] [in] [fweight], by(grpvar) [detail]



Examples based on Detect Data set A
 40 persons in 20 dyads in 10 groups in 4 collectivities
 4 metric variables: negotiation, satisfaction, performance, taskclarity

Sven-Oliver Spieß

24.06.2009 ║ Page 21
24.06.2009
wabacorr.ado in Stata
. wabacorr

negotiation satisfaction performance taskclarity, by(dyad)

Within and between analysis
Group variable: dyad

Number of obs
Number of groups

=
=

40
20

Obs per group: min =
avg =
max =

2
2.0
2

Within- and between-groups Etas and Eta-squared values:
------------------------------------------------------------------------------Variable | Eta-betw Eta-with
Eta-b^2
Eta-w^2
F
p>F
------------------+-----------------------------------------------------------negotiation |
0.2846
0.9586
0.0810
0.9190
10.7769
0.0000
satisfaction |
0.2783
0.9605
0.0774
0.9226
11.3170
0.0000
performance |
0.9988
0.0493
0.9976
0.0024 431.6194
0.0000
taskclarity |
0.9944
0.1054
0.9889
0.0111
93.7529
0.0000
-------------------------------------------------------------------------------

 Parts

Sven-Oliver Spieß

24.06.2009 ║ Page 22
24.06.2009
wabacorr.ado in Stata
Within- and between-groups correlations:

--------------------------------------------------------------------Variables |
r-betw
r-with
z'
p>z'
----------------------------+---------------------------------------negotiation-satisfaction |
-0.1973
0.8441
-3.0614
0.0011
negotiation-performance |
0.1413
-0.0477
0.2794
0.3900
negotiation-taskclarity |
-0.0589
0.0502
0.0257
0.4897
satisfaction-performance |
-0.0346
0.0695
-0.1037
0.4587
satisfaction-taskclarity |
0.0526
0.1568
-0.3119
0.3776
performance-taskclarity |
-0.9679
-0.1157
5.7429
0.0000
---------------------------------------------------------------------

Sven-Oliver Spieß

 Parts

 Noise
 Wholes

24.06.2009 ║ Page 23
24.06.2009
wabacorr.ado in Stata
Total correlation and components:

------------------------------------------------------------------------------Variables |
r-total betw-comp with-comp
z'
p>|z'|
----------------------------+-------------------------------------------------negotiation-satisfaction |
0.7616
-0.0156
0.7772
-3.0239
0.0025
negotiation-performance |
0.0379
0.0402
-0.0023
0.1122
0.9107
 Parts
negotiation-taskclarity |
-0.0116
-0.0167
0.0051
0.0343
0.9726
satisfaction-performance |
-0.0063
-0.0096
0.0033
0.0187
0.9851
satisfaction-taskclarity |
0.0304
0.0145
0.0159
-0.0039
0.9969
performance-taskclarity |
-0.9620
-0.9614
-0.0006
5.8045
0.0000
-------------------------------------------------------------------------------



Induction for the correlation between negotiation and satisfaction is parts



Thus variables should not be aggregated, but higher level information could
be disregarded without a big loss

Sven-Oliver Spieß

24.06.2009 ║ Page 24
24.06.2009
wabacorr.ado in Stata


What if induction is wholes (as with performance and taskclarity) or
equivocal?



If possible repeat WABA at the next higher level until induction is parts
 New number of cases N equals the number of groups M during the
previous analysis
 Input/initial values are correspondingly the means μ•j of the previous
analysis
 This is called multiple WABA
 In unbalanced data the means must be weighted to avoid distortions
(wabacorr supports frequency weights)
 Aggregate data no higher than level of first parts induction, but do not

disregard levels where inductions were equivocal


Stata again:

Sven-Oliver Spieß

24.06.2009 ║ Page 25
24.06.2009
wabacorr.ado in Stata
. collapse (mean) performance taskclarity group collectivity
(count) obs=performance, by(dyad)
. wabacorr performance taskclarity [fweight=obs], by(group)
Within and between analysis
Group variable: group

40

=
=

20
10

Obs per group: min =
avg =
max =
Number of weighted obs =

Number of obs
Number of groups

2
2.0
2

Weighted obs per group: min =
avg =
max =

4
4.0
4

:::
Output omitted
:::
. collapse (mean) performance taskclarity collectivity
(rawsum) obs [fweight=obs], by(group)
. wabacorr performance taskclarity [fweight=obs], by(collectivity)
Sven-Oliver Spieß

24.06.2009 ║ Page 26
24.06.2009
wabacorr.ado in Stata
. wabacorr performance taskclarity [fweight=obs], by(collectivity)

Within and between analysis
Group variable: collectivity

40

=
=

10
4

Obs per group: min =
avg =
max =
Number of weighted obs =

Number of obs
Number of groups

2
2.5
3

Weighted obs per group: min =
avg =
max =

8
10.0
12

:::
Output omitted
:::



Induction remains wholes even at the highest level



Data could thus be aggregated by collectivities

Sven-Oliver Spieß

24.06.2009 ║ Page 27
24.06.2009
Example for an Analysis:
Dansereau et al. (2006)

 Initial value
 Step 2
 Step 1
 Step 3
 Step 2
 Step 1

 Step 3
 Step 2
 Step 1
 Step 1

Sven-Oliver Spieß

24.06.2009 ║ Page 28
24.06.2009
Conclusions I


Within and between analysis
 provides a detailed picture of patterns of associations between variables
at different levels in nested hierarchical data instead of an all-or-nothing
decision as with ANOVA or intra-class correlations (ICC)
 has its greatest added value in equivocal cases
 can reveal important results even if total correlation is nil
 can be employed at two levels (single WABA) or successively at more

levels (multiple WABA)
 can also be employed in multivariate contexts like regression analysis (cf.

Dansereau et al. (2006))
 can inform further analyses, like the choice of levels in multi level

modeling (MLM), and selection of starting points for interventions
Sven-Oliver Spieß

24.06.2009 ║ Page 29
24.06.2009
Conclusions II


wabacorr.ado
 performs WABA of correlations in Stata 9.2 or higher
 also provides tests of practical significances with ‘detail’ option
 supports frequency weights to allow multiple WABA with unbalanced data
 stores results for further use by the user

Sven-Oliver Spieß

24.06.2009 ║ Page 30
24.06.2009
Further sources


Method:
 Dansereau, F., Cho, J. and Francis J. Yammarino. (2006).
Avoiding the "Fallacy of the Wrong Level": A Within and Between Analysis
(WABA) Approach. Group & Organization Management, 31, 536 - 577.
 O'Connor, B. P. (2004). SPSS and SAS programs for addressing
interdependence and basic levels-of-analysis issues in psychological
data. Behavior Research Methods, Instrumentation, and Computers, 36
(1), 17-28.
 Detect software: http://www.levelsofanalysis.com



wabacorr.ado:
 http://www.wip-mannheim.de/
 http://www.svenoliverspiess.net/stata
 Soon: Statistical Software Components

Sven-Oliver Spieß

24.06.2009 ║ Page 31
24.06.2009
Thank you!

Más contenido relacionado

Similar a Within and Between Analysis (WABA).

Why Study Statistics Arunesh Chand Mankotia 2004
Why Study Statistics   Arunesh Chand Mankotia 2004Why Study Statistics   Arunesh Chand Mankotia 2004
Why Study Statistics Arunesh Chand Mankotia 2004
Consultonmic
 
C.V.P Analysis of Nishat Mills
C.V.P Analysis of Nishat MillsC.V.P Analysis of Nishat Mills
C.V.P Analysis of Nishat Mills
Maryam Khan
 
Iso 9001 guidelines
Iso 9001 guidelinesIso 9001 guidelines
Iso 9001 guidelines
jomsatgec
 

Similar a Within and Between Analysis (WABA). (20)

Why Study Statistics Arunesh Chand Mankotia 2004
Why Study Statistics   Arunesh Chand Mankotia 2004Why Study Statistics   Arunesh Chand Mankotia 2004
Why Study Statistics Arunesh Chand Mankotia 2004
 
Javier Garcia - Verdugo Sanchez - Six Sigma Training - W2 Correlation and Reg...
Javier Garcia - Verdugo Sanchez - Six Sigma Training - W2 Correlation and Reg...Javier Garcia - Verdugo Sanchez - Six Sigma Training - W2 Correlation and Reg...
Javier Garcia - Verdugo Sanchez - Six Sigma Training - W2 Correlation and Reg...
 
Qm1 notes
Qm1 notesQm1 notes
Qm1 notes
 
Qm1notes
Qm1notesQm1notes
Qm1notes
 
Demand Forcasting
Demand ForcastingDemand Forcasting
Demand Forcasting
 
Consistently Modeling Joint Dynamics of Volatility and Underlying To Enable E...
Consistently Modeling Joint Dynamics of Volatility and Underlying To Enable E...Consistently Modeling Joint Dynamics of Volatility and Underlying To Enable E...
Consistently Modeling Joint Dynamics of Volatility and Underlying To Enable E...
 
Regression Presentation.pptx
Regression Presentation.pptxRegression Presentation.pptx
Regression Presentation.pptx
 
LESSON 04 - Descriptive Satatistics.pdf
LESSON 04 - Descriptive Satatistics.pdfLESSON 04 - Descriptive Satatistics.pdf
LESSON 04 - Descriptive Satatistics.pdf
 
C.V.P Analysis of Nishat Mills
C.V.P Analysis of Nishat MillsC.V.P Analysis of Nishat Mills
C.V.P Analysis of Nishat Mills
 
Eco
EcoEco
Eco
 
Iso 9001 guidelines
Iso 9001 guidelinesIso 9001 guidelines
Iso 9001 guidelines
 
correlation.pptx
correlation.pptxcorrelation.pptx
correlation.pptx
 
Topic 4.1
Topic 4.1Topic 4.1
Topic 4.1
 
Statistics
StatisticsStatistics
Statistics
 
Using Microsoft Excel for Weibull Analysis by William Dorner
Using Microsoft Excel for Weibull Analysis by William DornerUsing Microsoft Excel for Weibull Analysis by William Dorner
Using Microsoft Excel for Weibull Analysis by William Dorner
 
Javier Garcia - Verdugo Sanchez - Six Sigma Training - W2 Simple Variance Ana...
Javier Garcia - Verdugo Sanchez - Six Sigma Training - W2 Simple Variance Ana...Javier Garcia - Verdugo Sanchez - Six Sigma Training - W2 Simple Variance Ana...
Javier Garcia - Verdugo Sanchez - Six Sigma Training - W2 Simple Variance Ana...
 
Friedman-SPSS.docx
Friedman-SPSS.docxFriedman-SPSS.docx
Friedman-SPSS.docx
 
lanen_5e_ch05_student.ppt
lanen_5e_ch05_student.pptlanen_5e_ch05_student.ppt
lanen_5e_ch05_student.ppt
 
Akuntansi Manajemen Edisi 8 oleh Hansen & Mowen Bab 11
Akuntansi Manajemen Edisi 8 oleh Hansen & Mowen Bab 11Akuntansi Manajemen Edisi 8 oleh Hansen & Mowen Bab 11
Akuntansi Manajemen Edisi 8 oleh Hansen & Mowen Bab 11
 
One-way ANOVA
One-way ANOVAOne-way ANOVA
One-way ANOVA
 

Más de COSTARCH Analytical Consulting (P) Ltd.

Más de COSTARCH Analytical Consulting (P) Ltd. (14)

Hospitality Analytics: Learn More About Your Customers
Hospitality Analytics: Learn More About Your CustomersHospitality Analytics: Learn More About Your Customers
Hospitality Analytics: Learn More About Your Customers
 
Dedh Ishqia: Social Sentiments
Dedh Ishqia: Social SentimentsDedh Ishqia: Social Sentiments
Dedh Ishqia: Social Sentiments
 
Karle Pyaar Karle: Social Sentiments
Karle Pyaar Karle: Social SentimentsKarle Pyaar Karle: Social Sentiments
Karle Pyaar Karle: Social Sentiments
 
Logistic Regression Analysis
Logistic Regression AnalysisLogistic Regression Analysis
Logistic Regression Analysis
 
Student's T-Test
Student's T-TestStudent's T-Test
Student's T-Test
 
Dyadic Data Analysis
Dyadic Data AnalysisDyadic Data Analysis
Dyadic Data Analysis
 
Sexiest of the Sexiest Job Profile: Sports Analyst
Sexiest of the Sexiest Job Profile: Sports AnalystSexiest of the Sexiest Job Profile: Sports Analyst
Sexiest of the Sexiest Job Profile: Sports Analyst
 
Structural Equation Modelling (SEM) Part 3
Structural Equation Modelling (SEM) Part 3Structural Equation Modelling (SEM) Part 3
Structural Equation Modelling (SEM) Part 3
 
Functional Data Analysis
Functional Data AnalysisFunctional Data Analysis
Functional Data Analysis
 
Digital Marketing
Digital MarketingDigital Marketing
Digital Marketing
 
Structural Equation Modelling (SEM) Part 2
Structural Equation Modelling (SEM) Part 2Structural Equation Modelling (SEM) Part 2
Structural Equation Modelling (SEM) Part 2
 
Structural Equation Modelling (SEM) Part 1
Structural Equation Modelling (SEM) Part 1Structural Equation Modelling (SEM) Part 1
Structural Equation Modelling (SEM) Part 1
 
Data mining and its applications!
Data mining and its applications!Data mining and its applications!
Data mining and its applications!
 
Approaches to the_analysis_of_survey_data
Approaches to the_analysis_of_survey_dataApproaches to the_analysis_of_survey_data
Approaches to the_analysis_of_survey_data
 

Último

Salient Features of India constitution especially power and functions
Salient Features of India constitution especially power and functionsSalient Features of India constitution especially power and functions
Salient Features of India constitution especially power and functions
KarakKing
 

Último (20)

Spatium Project Simulation student brief
Spatium Project Simulation student briefSpatium Project Simulation student brief
Spatium Project Simulation student brief
 
REMIFENTANIL: An Ultra short acting opioid.pptx
REMIFENTANIL: An Ultra short acting opioid.pptxREMIFENTANIL: An Ultra short acting opioid.pptx
REMIFENTANIL: An Ultra short acting opioid.pptx
 
Jamworks pilot and AI at Jisc (20/03/2024)
Jamworks pilot and AI at Jisc (20/03/2024)Jamworks pilot and AI at Jisc (20/03/2024)
Jamworks pilot and AI at Jisc (20/03/2024)
 
Making communications land - Are they received and understood as intended? we...
Making communications land - Are they received and understood as intended? we...Making communications land - Are they received and understood as intended? we...
Making communications land - Are they received and understood as intended? we...
 
ICT Role in 21st Century Education & its Challenges.pptx
ICT Role in 21st Century Education & its Challenges.pptxICT Role in 21st Century Education & its Challenges.pptx
ICT Role in 21st Century Education & its Challenges.pptx
 
On National Teacher Day, meet the 2024-25 Kenan Fellows
On National Teacher Day, meet the 2024-25 Kenan FellowsOn National Teacher Day, meet the 2024-25 Kenan Fellows
On National Teacher Day, meet the 2024-25 Kenan Fellows
 
Graduate Outcomes Presentation Slides - English
Graduate Outcomes Presentation Slides - EnglishGraduate Outcomes Presentation Slides - English
Graduate Outcomes Presentation Slides - English
 
Salient Features of India constitution especially power and functions
Salient Features of India constitution especially power and functionsSalient Features of India constitution especially power and functions
Salient Features of India constitution especially power and functions
 
Towards a code of practice for AI in AT.pptx
Towards a code of practice for AI in AT.pptxTowards a code of practice for AI in AT.pptx
Towards a code of practice for AI in AT.pptx
 
Micro-Scholarship, What it is, How can it help me.pdf
Micro-Scholarship, What it is, How can it help me.pdfMicro-Scholarship, What it is, How can it help me.pdf
Micro-Scholarship, What it is, How can it help me.pdf
 
How to Give a Domain for a Field in Odoo 17
How to Give a Domain for a Field in Odoo 17How to Give a Domain for a Field in Odoo 17
How to Give a Domain for a Field in Odoo 17
 
Basic Civil Engineering first year Notes- Chapter 4 Building.pptx
Basic Civil Engineering first year Notes- Chapter 4 Building.pptxBasic Civil Engineering first year Notes- Chapter 4 Building.pptx
Basic Civil Engineering first year Notes- Chapter 4 Building.pptx
 
HMCS Max Bernays Pre-Deployment Brief (May 2024).pptx
HMCS Max Bernays Pre-Deployment Brief (May 2024).pptxHMCS Max Bernays Pre-Deployment Brief (May 2024).pptx
HMCS Max Bernays Pre-Deployment Brief (May 2024).pptx
 
80 ĐỀ THI THỬ TUYỂN SINH TIẾNG ANH VÀO 10 SỞ GD – ĐT THÀNH PHỐ HỒ CHÍ MINH NĂ...
80 ĐỀ THI THỬ TUYỂN SINH TIẾNG ANH VÀO 10 SỞ GD – ĐT THÀNH PHỐ HỒ CHÍ MINH NĂ...80 ĐỀ THI THỬ TUYỂN SINH TIẾNG ANH VÀO 10 SỞ GD – ĐT THÀNH PHỐ HỒ CHÍ MINH NĂ...
80 ĐỀ THI THỬ TUYỂN SINH TIẾNG ANH VÀO 10 SỞ GD – ĐT THÀNH PHỐ HỒ CHÍ MINH NĂ...
 
UGC NET Paper 1 Mathematical Reasoning & Aptitude.pdf
UGC NET Paper 1 Mathematical Reasoning & Aptitude.pdfUGC NET Paper 1 Mathematical Reasoning & Aptitude.pdf
UGC NET Paper 1 Mathematical Reasoning & Aptitude.pdf
 
Python Notes for mca i year students osmania university.docx
Python Notes for mca i year students osmania university.docxPython Notes for mca i year students osmania university.docx
Python Notes for mca i year students osmania university.docx
 
SKILL OF INTRODUCING THE LESSON MICRO SKILLS.pptx
SKILL OF INTRODUCING THE LESSON MICRO SKILLS.pptxSKILL OF INTRODUCING THE LESSON MICRO SKILLS.pptx
SKILL OF INTRODUCING THE LESSON MICRO SKILLS.pptx
 
Key note speaker Neum_Admir Softic_ENG.pdf
Key note speaker Neum_Admir Softic_ENG.pdfKey note speaker Neum_Admir Softic_ENG.pdf
Key note speaker Neum_Admir Softic_ENG.pdf
 
Beyond_Borders_Understanding_Anime_and_Manga_Fandom_A_Comprehensive_Audience_...
Beyond_Borders_Understanding_Anime_and_Manga_Fandom_A_Comprehensive_Audience_...Beyond_Borders_Understanding_Anime_and_Manga_Fandom_A_Comprehensive_Audience_...
Beyond_Borders_Understanding_Anime_and_Manga_Fandom_A_Comprehensive_Audience_...
 
SOC 101 Demonstration of Learning Presentation
SOC 101 Demonstration of Learning PresentationSOC 101 Demonstration of Learning Presentation
SOC 101 Demonstration of Learning Presentation
 

Within and Between Analysis (WABA).

  • 1. 2009 German Stata Users Group Meeting Performing within and between analysis (WABA) in Stata Sven-Oliver Spieß
  • 2. Outline  Introduction Where are constructs in hierarchical data really associated?  Within and between analysis  The basic idea  Simple ANOVA  Partitioning of correlations  Graphical demonstration  wabacorr.ado in Stata Sven-Oliver Spieß 24.06.2009 ║ Page 2 24.06.2009
  • 3. Introduction  Many names for one common problem: Fallacy of composition, ecological fallacy, atomistic fallacy, individualistic fallacy, Simpson’s paradox, …  Fallacies of the wrong level  Therefore global correlations potentially misleading  E.g. Corr(Job Satisfaction, Commitment) = .72  But at which level is the association? Individuals? Work groups? Departments?  Particularly problematic in applied settings  No simple random samples  Interventions Sven-Oliver Spieß 24.06.2009 ║ Page 3 24.06.2009
  • 4. Introduction  rJob Satisfaction,Commitment = .72 ρ? Branch ρ? Department ρ? Employee Observations Job Satisfaction  Commitment At which level is the association? Sven-Oliver Spieß 24.06.2009 ║ Page 4 24.06.2009
  • 5. Outline  Introduction Where are constructs in hierarchical data really associated?  Within and between analysis  The basic idea  Simple ANOVA  Partitioning of correlations  Graphical illustration  wabacorr.ado in Stata Sven-Oliver Spieß 24.06.2009 ║ Page 5 24.06.2009
  • 6. The basic idea  The idea: Let’s split up the total correlation into a component within the groups and another component between the groups  similar to idea behind analysis of variance (ANOVA)  Simply needs to be adjusted to correlations  Data prerequisites: variables in question must be metric and levels must be nested Sven-Oliver Spieß 24.06.2009 ║ Page 6 24.06.2009
  • 7. Simple ANOVA 8  Xij = (Xij – μ•j) + μ•j  Random Values 6 SSTotal = SSError + SSGroup (Xij – μ••)² = (Xij – μ•j)² + (μ•j – μ••)² 4 Grandmean 2  η² = SSGroup / SSTotal ==> η² between measure Sven-Oliver Spieß 1 2 3 Group 24.06.2009 ║ Page 7 24.06.2009 4
  • 8. Partitioning of correlations  Adjusted to correlations: rxy = ηBx * ηBy * rBxy + ηWx * ηWy * rWxy  rxy = CB + CW  ηB = corr(μ•j,Xij)  ηW = corr[(Xij – μ•j),Xij]  Central question: is between or within component (i.e. higher or lower level, or both) of total correlation more important? Sven-Oliver Spieß 24.06.2009 ║ Page 8 24.06.2009
  • 9. Partitioning of correlations  Typical procedure in 3 steps: 1. Univariate comparison of the within and between variances rxy = ηBx * ηBy * rBxy + ηWx * ηWy * rWxy 2. Bivariate comparison of the within and between correlations rxy = ηBx * ηBy * rBxy + ηWx * ηWy * rWxy 3. Summary judgment on the importance of the within and between components for the total correlation rxy = CB + CW Sven-Oliver Spieß 24.06.2009 ║ Page 9 24.06.2009
  • 10. Partitioning of correlations 4 possible outcomes/inductions: 1. Parts  lower level/within 2. Wholes  higher level/between 3. Equivocal  meaningful association at both levels 4. Inexplicable  noise Sven-Oliver Spieß 24.06.2009 ║ Page 10 24.06.2009
  • 11. Graphical illustration: Step 1 40 WABA1: Scatterplot Negotiation Y= 30 Raw Scores 20 Y 10 0 -10 -10 Sven-Oliver Spieß 0 10 20 Raw Scores 30 40 24.06.2009 ║ Page 11 24.06.2009
  • 12. Graphical illustration: Step 1 (ηB) 40 WABA1: Scatterplot Negotiation Y= 30 Raw Scores 20 xj Y 10 Dyad 2 0 -10 -10 Sven-Oliver Spieß 0 10 20 Raw Scores 30 40 24.06.2009 ║ Page 12 24.06.2009
  • 13. Graphical illustration: Step 1 (ηB) 40 WABA1: Scatterplot Negotiation Y= 30 Raw Scores Group Means 20 xj Y 10 Dyad 2 0 -10 -10 Sven-Oliver Spieß 0 10 20 Raw Scores 30 40 24.06.2009 ║ Page 13 24.06.2009
  • 14. Graphical illustration: Step 1 (ηB) 40 WABA1: Scatterplot Negotiation Y= 30 Raw Scores Group Means 20 Y 10 20-16.5 0 -10 -10 Sven-Oliver Spieß 0 10 20 Raw Scores 30 40 24.06.2009 ║ Page 14 24.06.2009
  • 15. Graphical illustration: Step 1 (ηW) WABA1: Scatterplot Negotiation 40 Y= 30 Raw Scores Group Means 20 Deviations from Means Y 10 0 -10 -10 Sven-Oliver Spieß 0 10 20 Raw Scores 30 40 24.06.2009 ║ Page 15 24.06.2009
  • 16. Graphical illustration: Step 1 WABA1: Scatterplot Negotiation 40 Y= 30 Raw Scores Group Means 20 Deviations from Means Y 10 Linear Fit: Parts Means 0 Deviations -10 -10 Sven-Oliver Spieß 0 10 20 Raw Scores 30 40 24.06.2009 ║ Page 16 24.06.2009
  • 17. Graphical illustration: Step 1 WABA1: Scatterplot Performance 40 Y= 30 Raw Scores Group Means 20 Deviations from Means Y 10 Linear Fit: Means 0 Deviations -10 -10 Sven-Oliver Spieß 0 10 20 Raw Scores 30 40 24.06.2009 ║ Page 17 24.06.2009
  • 18. Graphical illustration: Step 2 20 10 0 Parts -10 Negotiation 30 40 Scatterplot Negotiation over Satisfaction 0 10 20 Satisfaction Raw Scores Group Means Deviations from Means Sven-Oliver Spieß 30 40 Fitted Scores Fitted Means Fitted Deviations 24.06.2009 ║ Page 18 24.06.2009
  • 19. Graphical illustration: Step 2 20 10 -10 0 Performance 30 40 Scatterplot Performance over Taskclarity 0 10 20 Taskclarity Raw Scores Group Means Deviations from Means Sven-Oliver Spieß 30 40 Fitted Scores Fitted Means Fitted Deviations 24.06.2009 ║ Page 19 24.06.2009
  • 20. Outline  Introduction Where are constructs in hierarchical data really associated?  Within and between analysis  The basic idea  Simple ANOVA  Partitioning of correlations  Graphical illustration  wabacorr.ado in Stata Sven-Oliver Spieß 24.06.2009 ║ Page 20 24.06.2009
  • 21. wabacorr.ado in Stata  General syntax: wabacorr varlist [if] [in] [fweight], by(grpvar) [detail]  Examples based on Detect Data set A  40 persons in 20 dyads in 10 groups in 4 collectivities  4 metric variables: negotiation, satisfaction, performance, taskclarity Sven-Oliver Spieß 24.06.2009 ║ Page 21 24.06.2009
  • 22. wabacorr.ado in Stata . wabacorr negotiation satisfaction performance taskclarity, by(dyad) Within and between analysis Group variable: dyad Number of obs Number of groups = = 40 20 Obs per group: min = avg = max = 2 2.0 2 Within- and between-groups Etas and Eta-squared values: ------------------------------------------------------------------------------Variable | Eta-betw Eta-with Eta-b^2 Eta-w^2 F p>F ------------------+-----------------------------------------------------------negotiation | 0.2846 0.9586 0.0810 0.9190 10.7769 0.0000 satisfaction | 0.2783 0.9605 0.0774 0.9226 11.3170 0.0000 performance | 0.9988 0.0493 0.9976 0.0024 431.6194 0.0000 taskclarity | 0.9944 0.1054 0.9889 0.0111 93.7529 0.0000 -------------------------------------------------------------------------------  Parts Sven-Oliver Spieß 24.06.2009 ║ Page 22 24.06.2009
  • 23. wabacorr.ado in Stata Within- and between-groups correlations: --------------------------------------------------------------------Variables | r-betw r-with z' p>z' ----------------------------+---------------------------------------negotiation-satisfaction | -0.1973 0.8441 -3.0614 0.0011 negotiation-performance | 0.1413 -0.0477 0.2794 0.3900 negotiation-taskclarity | -0.0589 0.0502 0.0257 0.4897 satisfaction-performance | -0.0346 0.0695 -0.1037 0.4587 satisfaction-taskclarity | 0.0526 0.1568 -0.3119 0.3776 performance-taskclarity | -0.9679 -0.1157 5.7429 0.0000 --------------------------------------------------------------------- Sven-Oliver Spieß  Parts  Noise  Wholes 24.06.2009 ║ Page 23 24.06.2009
  • 24. wabacorr.ado in Stata Total correlation and components: ------------------------------------------------------------------------------Variables | r-total betw-comp with-comp z' p>|z'| ----------------------------+-------------------------------------------------negotiation-satisfaction | 0.7616 -0.0156 0.7772 -3.0239 0.0025 negotiation-performance | 0.0379 0.0402 -0.0023 0.1122 0.9107  Parts negotiation-taskclarity | -0.0116 -0.0167 0.0051 0.0343 0.9726 satisfaction-performance | -0.0063 -0.0096 0.0033 0.0187 0.9851 satisfaction-taskclarity | 0.0304 0.0145 0.0159 -0.0039 0.9969 performance-taskclarity | -0.9620 -0.9614 -0.0006 5.8045 0.0000 -------------------------------------------------------------------------------  Induction for the correlation between negotiation and satisfaction is parts  Thus variables should not be aggregated, but higher level information could be disregarded without a big loss Sven-Oliver Spieß 24.06.2009 ║ Page 24 24.06.2009
  • 25. wabacorr.ado in Stata  What if induction is wholes (as with performance and taskclarity) or equivocal?  If possible repeat WABA at the next higher level until induction is parts  New number of cases N equals the number of groups M during the previous analysis  Input/initial values are correspondingly the means μ•j of the previous analysis  This is called multiple WABA  In unbalanced data the means must be weighted to avoid distortions (wabacorr supports frequency weights)  Aggregate data no higher than level of first parts induction, but do not disregard levels where inductions were equivocal  Stata again: Sven-Oliver Spieß 24.06.2009 ║ Page 25 24.06.2009
  • 26. wabacorr.ado in Stata . collapse (mean) performance taskclarity group collectivity (count) obs=performance, by(dyad) . wabacorr performance taskclarity [fweight=obs], by(group) Within and between analysis Group variable: group 40 = = 20 10 Obs per group: min = avg = max = Number of weighted obs = Number of obs Number of groups 2 2.0 2 Weighted obs per group: min = avg = max = 4 4.0 4 ::: Output omitted ::: . collapse (mean) performance taskclarity collectivity (rawsum) obs [fweight=obs], by(group) . wabacorr performance taskclarity [fweight=obs], by(collectivity) Sven-Oliver Spieß 24.06.2009 ║ Page 26 24.06.2009
  • 27. wabacorr.ado in Stata . wabacorr performance taskclarity [fweight=obs], by(collectivity) Within and between analysis Group variable: collectivity 40 = = 10 4 Obs per group: min = avg = max = Number of weighted obs = Number of obs Number of groups 2 2.5 3 Weighted obs per group: min = avg = max = 8 10.0 12 ::: Output omitted :::  Induction remains wholes even at the highest level  Data could thus be aggregated by collectivities Sven-Oliver Spieß 24.06.2009 ║ Page 27 24.06.2009
  • 28. Example for an Analysis: Dansereau et al. (2006)  Initial value  Step 2  Step 1  Step 3  Step 2  Step 1  Step 3  Step 2  Step 1  Step 1 Sven-Oliver Spieß 24.06.2009 ║ Page 28 24.06.2009
  • 29. Conclusions I  Within and between analysis  provides a detailed picture of patterns of associations between variables at different levels in nested hierarchical data instead of an all-or-nothing decision as with ANOVA or intra-class correlations (ICC)  has its greatest added value in equivocal cases  can reveal important results even if total correlation is nil  can be employed at two levels (single WABA) or successively at more levels (multiple WABA)  can also be employed in multivariate contexts like regression analysis (cf. Dansereau et al. (2006))  can inform further analyses, like the choice of levels in multi level modeling (MLM), and selection of starting points for interventions Sven-Oliver Spieß 24.06.2009 ║ Page 29 24.06.2009
  • 30. Conclusions II  wabacorr.ado  performs WABA of correlations in Stata 9.2 or higher  also provides tests of practical significances with ‘detail’ option  supports frequency weights to allow multiple WABA with unbalanced data  stores results for further use by the user Sven-Oliver Spieß 24.06.2009 ║ Page 30 24.06.2009
  • 31. Further sources  Method:  Dansereau, F., Cho, J. and Francis J. Yammarino. (2006). Avoiding the "Fallacy of the Wrong Level": A Within and Between Analysis (WABA) Approach. Group & Organization Management, 31, 536 - 577.  O'Connor, B. P. (2004). SPSS and SAS programs for addressing interdependence and basic levels-of-analysis issues in psychological data. Behavior Research Methods, Instrumentation, and Computers, 36 (1), 17-28.  Detect software: http://www.levelsofanalysis.com  wabacorr.ado:  http://www.wip-mannheim.de/  http://www.svenoliverspiess.net/stata  Soon: Statistical Software Components Sven-Oliver Spieß 24.06.2009 ║ Page 31 24.06.2009