Sie sind auf Seite 1von 54

CEV – Corporate ethical value – 5

Type 1 EC – Ethical Climate – 10


TI – Turnover intention - 5
OC – Org commitment - 8
JP – Job performance - 10
JS- Job Satisfaction - 8

Chi square test is used for testing existence of association between two categorical variables
Won’t tell you the strength, direction of association

Correlation - tool used to test the relation between two metric variables
1. Magnitude of relation calculated
2. Direction of relation identified
3. Significance of relation assessed (Significance - possibility of committing type I error -
should not be more than 5%) (SIG <= 0.05 → This means our hypothesis is upheld.)
a. Hypothesis - 3 forms
i. Involving testing relations or associations
ii. Testing of Difference
iii. Predictive Hypothesis - Deals with predictions
iv. Involving testing of Impact - Closer to predictive hypothesis. Ex. Job
satisfaction has -ve impact on employee turnover intention.
v. Used in experimental scenario - used to test various tests and
experiments - causal Hypothesis - involves testing cause/effect
relationship
vi. Univariate Hypothesis - Only one variable to test. Ex. mean compensation
of employees in this firm is more than 5000 USD/month. Here only one
variable is present - income.
vii. Sometimes 2 variables are used - Bivariate Hypothesis.
viii. For multivariate hypothesis - many variables. Ex. JS, TI to have
significant impact on job performance. Here we have 2 independent + 1
dependent. Sometimes we might have 1 independent + 2 dependent.(JS
having impact on both JP & TI)
4. Types of variables
1. Extraneous variable
2. Moderator variable - (conditioning variable. Doesn’t have impact on variables)
3. Mediator variable - Impact of one variable on another variable is mediated through ??
e.g. Organisational Justice builds Organisational Trust, which results in Organisational
Citizenship Behaviour
4. Higher level - organisational justice, employee trust
a. also known as concept - because there's no universal agreement on how to
measure these variables
b. Subjective in nature
5. Lower level variable
a. Concrete in nature, not a concept
b. Known as Indicators, Manifest variable
c. Eg:-age,income,years of experience, gender, roll no.
d. Measures the stage at which something is measured

Construct vs concept
Construct - the way in which concept is operationalised and measured. Eg - Employee
commitment to org can be global commitment
Different level of construct- 0 level, 1st level and so on (medium level, high level
variable)
Concept -
Employee commitment - 3 distinct elements
Normative Commitment
Affective Commitment
Continuous Commitment

Same concept measured with different constructs

Q. Why is concrete (indicator/ manifest variable) a lower level variable?


Ans When you try to convert qualitative into quantitative (eg strongly agree, agree, disagree)

No inferior superior relationship b/w lower and higher level variable

Eg Concept - Employee Engagement (higher order construct)


It has 3 subdimensions (mid level constructs)
- Employee rigour - (strong rigour, OkOk , frax - lower level)
- Employee absorption
- Employee dedication
There’ll be indicator error in the lower level variable (epsilon)

Different researchers use different constructs to measure the same concepts


This creates great difficulty in terms of arriving at a consensus on a theory

This is both a limitation and an added advantage <- because it broadens the theoretical
perspective, which helps evolve new theories and refute outdated ones

3 types of research -
1. Exploratory - MEthod that is used when your research objective is to gain new insights.
This type of research begins with questions like why and how (primarily the two ques
asked)
Eg - Why all of a sudden there's a huge attrition rate in our org? , why child labour?, why
poverty? - these make us go and gather more information to gain insight and understand
the research better. You EXPLORE !
a. Features - no specific research …… No Hypothesis. No concrete research
objective. You rely on research methods like library, literature survey,
discussions with experts, people who have connections with the issue under
investigation, in-depth interviews, etc. You don't use statistical techniques/ data
analytics. At Max percentage analysis.
b. You are focused on getting qualitative information.Thus it is also called
qualitative research. It is not scientific. No random sample. No inferences.
Nothing. No generalization of research.
c. Methods of doing
i. In-depth one to one interview (also called clinical interview). -
Unstructured Interview with only guiding questions. No specific
questionnaire.
ii. Focused GD. Limitations - data analysis is difficult. Analysis becomes
highly subjective depending upon the moderator. Pre conceived notions
to have major role in influencing the outcome of Focused GD
iii. Projective Techniques -
1. Word association test. (ZNMD type game).
2. Sentence completion (You use detergent powder …..) - Let user
complete the test and you get the data then.
3. Story completion technique - you tell first part of story and let user
complete the story.
4. Cartoon test - Respond on the empty test (dont know what he
said here)
Cartoon is shown with two panels - for one character, something
is written; for another, nothing is written. Respondent is
supposed to fill the empty space for the second panel
5. Role Play - Ask customers to enact how would he behave in any
particular situation
6. Case study - obtain qualitative info. Pertaining to an event or to an
entity like an org or to an indiv etc
d. Exploratory research generates hypothesis and nothing else

2. Descriptive - (I will write this - Jajoo)


It is concerned about answering questions of What, when, where, how often etc.
You can describe relation, study impact, interdependence, make predictions etc.
Common forms of desc research
1. Self administration
2. Personal Interview method - You add questions, you record response
3. Online questionnaire - survey. To maintain anonymity of responses
4. Telephone survey (Computer Aided Telephone Interview - CATI) - Responses
are recorded and coded into centrally located computer catalogue

3. Experimental -
Causal research. 3 conditions are required
a. There has to be a co variation between treatment and outcome(experimental)
variable should be present
b. There should be continuity - continuity means before we observe change in the
outcome variable we have to make changes in the treatment variable itself. Ex -
Dark clouds appear before rain. Change in the treatment variable should occur
first. treatment variable = dark clouds;outcome variable = rain
c. Covariate control - Controlling for extraneous variables (noise/ contaminating
variables) will have impact on outcome variables.They are limitless. You choose
and measure and remove atleast 1 or 2 of such extraneous variables , those
variables become control variables.
Similar to accuracy
How do you control covariate - Design - NAture of questionnaire design and ...
Second is what you call analtyical control

Choose a particular technique to control the impact of these variables

Heirarchial regression analysis


Effect of covariate based on????
Abbe kch samajh nhi aata h kya bolta h

Main focus of experimental research is to test or establish Causal effect/ causality - Impact &
To establish causality - correlation is basic necessity

Methods to establish causal effect-


Methods for experimental
1. PED (pre experimental design) - What is it, how is it used, how is it figuratively
represented?
they follow basic experimental steps but fail to include a control group. In other
words, a single group is often studied but no comparison between an equivalent
non-treatment group is made.
2. TED (true experimental design)
3. QED (quasi-experimental design)
4. CED/SED (complex/statistical experimental design)

Lab experiment (in house,happens in controlled environment) or field experiment (real world)
E.g. of field experiment = test marketing
Tradeoff of lab v/s field = accuracy v/s generalisability, since some other marketing events might
be impacting our field result (we can’t control noise variables basically?)

Accuracy = internal validity - change in treatment variable causing change in outcome variable
External validity -you do extended research outside

------------------------------------------
Class 4

1. Divide the respondents on the basis of median job satisfaction and turnover intention
2. Is there an association between employee's JS and TI on the basis of these categorisations?

Chi-squared over correlation because data is categorical

Step 1: Summate JS & TI scores and call it as JS Total/TI Total


Step 2: Compute median for JS Total and TI Total
Step 3: Divide the respondents into two categories based on the median and assign code 1 and
2
respectively. Declare what is 1, what is 2 in Variable View, (1: Low, 2: High)
Step 4: Perform the analysis (Chi-Squared) on coded JS Total and TI Total

Analyse -> descriptive -> cross tab : give dependent variable (TI) as row & treatment variable as
column (JS)
Step 5: Draw inference

Cramers test - for multiple rows and columns - won’t give you direction of relationship (limitation)
Otherwise - use phi (will usually be same as Cramer’s)

Phi gives direction

Magnitude of association between these two variables is 0.352, which is negative


=> TI and JS are negatively associated

Direction, magnitude, significance


Magnitude of association b/w these two variables is 0..352 (-ve) i.e JS and TI are -vely
assoCiated

-------------

Q. Is there an association between an employee’s TI & Job Performance upon their median
classification? If so, what is the magnitude of such association?
Q. Is the strength of association between TI & JP significant for employees who are low job
satisfied and for those who are high job satisfied?
Ans - It’ll be significant if the asymptotic value of chi-squared analysis is <0.05

Q. Verify your result for the prev question with the following:
Consider JP_Total and TI_Total
No need for taking mediator
Is the correlation b/w employee TI and their JP different for employees in different
category of JS

Step 1: Select only the original JP_Total and TI_Total, and compute the correlation (Analyze ->
Correlate -> Bivariate)
Step 2: You have JS_recoded (median based), select this variable and split the file on the basis
of this variable
Step 3: Study the correlation b/w TI and JP
Step 4: Make a table & draw your inferences
Factor Analysis

1. Technique used to identify the existence of dimensions across a given set of variables.
2. We need metric variables (continuous variables)
3. Terms used in factor analysis research -
a. KMO statistics - Used for testing the sampling adequacy
b. KMO value has to be at least 0.5
c. BTS used for testing the hypothesis that variables are interrelated. It is based on
chi-squared distribution. Thus it should be significant at 0.05 level
d. Loadings - A measure that indicates strength of relation between an item and a
factor. If item has to be at least 0.5 loaded onto a specific factor only
e. The above is known as a clear factor loading
f. On the other hand if the loading is 0.5 or more on more than a single factor we
call it as a cross loading; whereas if loading is undue factor to which it is not
supposed to come up to the extent of 0.5 or more, we call it as misloading.
Similarly if an item is not at all loaded onto your component (to the extent of
atleast 0.5 or more) it is known as a non loading.
g. Sometimes items pertaining to two different dimensions or components are fallen
under a single component itself. This is known as mixed loading. Finally its quite
possible that an item will form a dimension of its own that is known as solo
loading.

When we encounter different sort of loading, we need to discard these erroneously loaded items
as follows. The order of deletion of items follows the following pattern.

Priority 1:-Delete an item whose communality is less than 0.3


Priority 2:- Delete the non loaded item if there is any
Priority 3:- Delete a cross loaded item
Priority 4:- Delete a misloading
Priority 5:- Delete a solo loaded item

In case of mixed loading force fit the factor solution

Communality means the contribution made by each item or variable or question 10.1 to the
extraction of factor solution. This has to be at least 0.3. This means the communality with
relation less than 0.3 will be removed first.
Cross loading is a situation when an item falls under more than one component to the extent of
0.5 or more
By misloading we mean an odd man item under a particular component for eg- suppose the
extracted component contains 5 item of which 4 items belong to turnover intention and the
remaining item belongs to job satisfaction, we need to remove that mis loaded item of JS.
By mixed loading we mean a situation where more than one item from different components
trying to form a separate component of its own.
As a rule of thumb we expect at least 3 items for each factor. In exceptional case the factor
goes with only 2 items or 2 questions.
Eigen value means the amount of variance explained by each factor across a set of items. Its
value has to be at least one. This is computed as sum of squared loadings for a factor or
component. Remember, communality is computed by summating the squared values of loading
for each item across components of factor and is represented as h2.
A factor solution is tenable if it explains a minimum 50% cumulative variance for all the factors
extracted.
It is customary that we look at the rotated component matrix based on Varimax rotated solution
(method). The uniqueness of this is that it makes factors orthogonal.

First Method: Principal Factor Analysis (Most superior)

Variance - Amount of information contributed/explained to factor solution


Q. Assess the dimensionality of the construct job performance.
Analyse -> Dimension Reduction -> Factor -> select set variables (Jp1-Jp10) ->
Descriptive button -> KMO and Bartlett button
Rotation button -> Select Varimax
Options button -> Select Sorted by size & suppress small coeff & absolute value below = 0.5

KMO should be 0-1 and closer to 1 for it to be absolutely fine


Significance in BTS - Variables are interrelated

This set of conditions is----


Set of conditions for Communality table:
Initially each variable has maximum
20 variables implies 20 pieces of information-
Who is the highest contributor - The one whose Extraction value is max in the table
If the commonality is less than .3- 94
Total Variance explained table :
1st component gives 81.84% of information if it has % of Variance = 81.843

Component Matrix Table:


If anything has <0.5 - Bad component
At its foot you need to write KMO = ?,BTS, corresponding degree of significance and Eigen
Value, % of variance

Factor analysis of job performance items (k=10) yielded a unidimensional factor structure.

Loadings are 0.5, amount of variance explained is 81.343%

Clear Factor solution is a situation where there is no question of non-loading, misloading etc

Once you have a CFS, go for computing factor score and interpret the results.

Q. Compute factor scores for jp for each employee and interpret the result for employee 1,5 and
10
In factor Analysis box only -> Scores button -> Select “Save as Variables” checkbox
Result - FAC1_1 column comes up in the data view
Compute mean (=0) and std dev of this (=1)
Poorest performer is 10 (Farthest from the mean)
Potential for good performance is 5 (Coz its closest to 0 (mean))
More the positive the better even if its far from mean.

Assess the dimensionality of organisational commitment item and interpret your results, present
your table and relevant values

Q. Compute factor score for OC. how do you interpret respondent 5 based on his jp factor score
and oc factor score.

Both are negative. What does it imply- one being more negative than the other

Q. 1Do Factor analysis on js item.

We found 2 clear factors with six items under Factor 1 and two items under Factor 2

First Component:
Second Component:

First component will always have higher value of variance explained


Q. Compute strength of relation between employee organisation commitment and job
performance ?

Strong positive correlation between OC and JP.

r jp,oc = 0.834*
*-> significant correlation

Q. Study correlation between employee JS and employee JP: what’s your inference?
Results indicate Global JS is positively moderately (GAJS_JP=0.578*) related to JP.
WA JS component is significantly positively but lowly (WASJS_JP = 0.137*) related to JP

Limitation of summated scores - Still contains errors


Adv of Factor score - more rigorous/clean coz all variables are orthogonal - considered to be
error free

Q. Compute correlation b/w JP and OC using summated score ?


Do you find any diff between values of correlation coeff for JP and OC under summated score
approach and factor approach

------------------------------------------------------------------------------------------------------------
How to assess reliability

● Relativity coefficient(Alpha) is given by le Cronbar and is known is Cronbar Alpha.


● Value ranges between 0 and 1.
● It is based on consistency princple.
● Computed for each component (or sub components)
● Alpha of at least 0.7 is expected if we have 3 or more items in our component
● Alpha of at least 0.5 is expected if we have 2 items in our component
● Items can be deleted at this stage also. (Factor analysis and reliability assessment) by
looking at CITC (Corrected Item Total Correlation) which is expected to be at least 0.35.
i.e . If CITC < 0.35, it will be disqualified from that particular component and reliability
constant is Recalculated

Navigation
Analyze - scale - reliability - select variables - statistics and check box ‘Scale item deleted’

Reliability - to check consistency of variables.

Q. Assess the consistency/relativity coefficient for JS, JP, OC components based on the
individual component wise factor solution applied till now.

Q. Assess the dimensionality of OC, JS and TI (Taking all together)


-----------------------------------------

Item Deletion History (All these we check in Rotated Component matrix)

Step Item Deletion Reason

1 OC5,OC2 CrossLoading(CL)

2 JS6, TI4 Misloading(only one element in a component of an item,JS6, and


other items has more than one, then removed JS6)

Forced fit solution (When substantial number of items of a single


component is found in another single component)

For forced fit -> In the factor analysis -> click on extraction button -> Select fixed number of
factors = 5
If Forced fit also doesn’t help then perform separate factor analysis on all 3

Item Retention Table

Component Items Alpha (Reliability)

C1 OC1,7,6,3,4 0.902

C2 JS8,1,2,7,5 0.871

C3 TI5,3,2 0.854

C4 JS4,3 0.807

Last component is meaningless, ignore it entirely

Surrogate variable: highest loaded item under a component


Q. Assess the dimensionality of constructs - elements, OCB, TI(IQ),JS, and LMX for
department heads in Mumbai plant whose years of experience <= 10 years

Factor Analysis : Based primarily on data from questionnaire

1. Assess dimensionality
2. Compute reliability for each component
3. Summate the score or identify surrogate variables or compute factor score for
each component extracted

Ans :-
STEP 1. Data->select cases->if condition is satisfied (separated by &)
STEP 2. Access dimensionality (do factor analysis)
Two condition for factor analysis(check in each iteration) :-
1. KMO(Kaiser-Meyer-Olkin Measure of Sampling Adequacy.) should be 0-1 and closer to 1 for
it to be absolutely fine
2. Significance in BTS - Variables are interrelated less than .05
Now step wise step deal with each priority :-
Priority 1:- Delete an item whose communality is less than 0.3(absolute value)
Priority 2:- Delete the non loaded(NL, blank) item if there is any
Priority 3:- Delete a cross loaded(CL, horizontally in more than 1 column) item
Priority 4:- Delete a misloading( two or more elements in 1 column with 1 having more than 1
element and other having just 1 element)
Priority 5:- Delete a solo loaded(SL) item
Priority 6:- Deal with Mixloading by Forced Factor (two or more elements in a column with more
than 1 elements of each)
For forced fit -> In the factor analysis -> click on extraction button -> Select fixed number of
factors = n+1
If Forced fit also doesn’t help then perform separate factor analysis on all 3
STOP IF ONLY SOLO LOADING IS LEFT AT THE END.

SNR Items Deleted Reason

1. OCB4, OCB9, OCB10 Non Loaded

2. JS4 CL

3 IQ3, OCB7 SL

4 OCB2 CL

FORCED FACTOR

5 JS1, JS5 CL
6 JS2 ML

7 OCB 16 NL

STEP 3. Do reliability test


Analyze->Scale->Reliability Analysis(Statistics- Check “Scale if item deleted)
Corrected Item-Total Correlation should be more than .35 for each
Record Cronbach's Alpha Value in the table

Item Retention Table


Column Name Those in the same column Alpha = Value of reliability
C1 LMX 6,3,4,7,1,5 .914

C2 OCB 8,14,13,12 .813

C3 OCB 5,6,11 .744

C4 OCB 3,1 .775

C5 IQ 2, IQ 1 .911

Don’t Add the single loaded elements in the above table

Also note the Eigen values (In total Variance explained table -> Rotation sums of squared
loadings):
Also report the KMO and Bartlett's Test result.

Surrogate Variables : The ones that are highest(first one in the order) in the list
Example from above - LMX 6, OCB 8 etc

Step 4:Summate Score: Transform-> Compute variables ( Summate column wise and give name
to the total, name multidimensional Factor’s total acc to the components)
Multidimensional Factor : The factor that comes in more than one component. (Eg OCB in C2,C3
and C4 in Item Retention Table)

Reporting the result (write the entire process) :-

Data pertaining to Employee JS,LMX,OCT obtained from 631 respondents of the manufacturing
org with multiple location was provided with a question of assessing the dimensionality of the
responses collected from dept heads in mumbai plant.

Towards this an exploratory factor analysis was done with verimax rotation.
What we have observed:
After eliminating items for non loading, cross loading, misloading and solo loading we observe 7
factor solution with items belonging to JS and IQ falling under a single component. Because this
is a mix loading pattern we resorted to a constraint model (a forced fit factor solution) with 8
factors. After eliminating the items for various improper loadings we obtained a clear 5 factor
solution (Although there were three other factors with solo loading, those solo loaded factors were
ignored).
Surprisingly, the entire construct of JS was lost, with OCB exhibiting a multi dimensional pattern,
while LMX and IQ items were retained interpretable. The multi dimensional structure of OCB was
named as OCB proactive (OCB 8,14,13,12), OCB defensive (OCB 5 ,6,11) and OCB altruistic
(OCB 3,1).

Q2. Your client wants the following information.


a. Correlation between OCB Components
b. Correlation between each of the OCB Components and TI
c. Number of department heads to have high OCB - proactive and defensive yet have
high TI.
d. Is there an association between OCB Proactive and experience Leader member
exchange (LMX) upon classifying the respondents in low and high on each attribute.
If there is a significant association between these two attributes, report the strength
of association.

Ans

a. Analyze -> Correlate -> Bivariate ( OCB proactive (OCB 8,14,13,12), OCB defensive
(OCB 5 ,6,11) and OCB altruistic (OCB 3,1).
Record Pearson Coeff pair wise in Star marked means significant or not
There is significant correlation b/w each. However Correlation b/w defensive and
Altruistic is low. For other two its moderate

b. Analyze -> Correlate -> Bivariate ( OCB proactive (OCB 8,14,13,12), OCB defensive
(OCB 5 ,6,11), OCB altruistic (OCB 3,1) and IQ_T(IQ 2, IQ 1).
With altruistic its significant. Interpretation: OCB altruistic and TI are inversely lowly
related, yet it is significant.

c. Analyze->Descriptive Statistics->Frequencies->Statistics (Check Median)


Transform-> Recode into diff variables ( lowest to Median ) name it as
OCB_”X”_Encoded & IQ_T_Encoded

Analyse -> descriptive -> cross tab : give dependent variable (OCB_”X”_Encoded ) as
row & IQ_T_Encoded
ALternatively,
Include all conditions in select cases and then calculate the frequency of filter_$ by Analyze-
>Descriptive Statistics->Frequencies

d. Cramers test - for multiple rows and columns - won’t give you direction of relationship
(limitation)
Otherwise - use phi (will usually be same as Cramer’s)

Phi gives direction


Magnitude of association between these two variables is 0.352, which is negative=> TI and JS
are negatively associated

https://docs.google.com/document/d/1e4XOObj0SnDBU5UfOr92aSxM4vonVKPCGmdY8VLQ9
gU/edit
Factor analysis works on principal component method, i.e. first component attracts most
variables, and so is most important, while last factor is least important

=> Most important factor

--------------------
1. Identify the number of respondents primarily driven by Overcoming Loneliness for the
usage of Facebook.
2. Assess the profile of the respondents identified in ques-1, on the other factor extracted
3. Is there any association between the relationship motivation and overcoming loneliness
motivation, upon classifying the respondents into low and high on each of the two
motivation?
4. Is there any relationship between Relationship Motivation and Communication Motivation
of using facebook? What explanation you give for your findings?
Note: Use factor score approach for answering these questions.

For model testing, based on factor analysis and reliability, you need to follow one of the 3
approaches:-

1. Factor-based approach
2. Summated score approach -
3. Surrogate variables approach

Summated Score: Raw - items falling under each criteria

Surrogate Variable: Consider the highest loaded item under each component.

Factor Score: Don’t ask for factor score unless you have a clear factor solution. Don’t activate the
option in the Factor Analysis box.

In Facebook case we have clear factor solution - i.e. all loadings are > 0.5, and each factor is
interpretable
Analyse -> dimension reduction -> Factor Analysis -> score button -> save as variables
Any factor score that is postive may be important, depending on the value of the factor score

Identify the number of respondents primarily driven by Overcoming Loneliness for the
usage of Facebook:-
Ans.

Overcoming_Loneliness > Comunication_Motive


AND Overcoming_Loneliness > Relationship_Motive
AND Overcoming_Loneliness > Entertainment_Motive

Identify the number of respondents driven by Overcoming Loneliness for the usage of
Facebook:-
Overcoming_Loneliness > 0
Discriminant Analysis ( This is going to come in exam ) (Read carefully)

Technique used to reveal how groups are different from each other in respect of a set of metric
variables.
● Dependent variable - categorical and predictors or discriminators metric variables.
● Discriminant function = D^ = alpha + b1x1 + b2x2 + …..bnXn. Here,
○ Alpha - constant
○ b1, b2 - coefficients
○ x1, x2 - actual values

Some of the techniques/terms we come across are as follows:-


1. Canonical Correlation: A measure used to test the strength of association between
discriminant score and group membership. A square of this Canonical correlation gives
information about what amount of variance is explained in differentiating between the
groups by a set of discriminators or predictors.
2. ANOVA F-Statistic: Used to assess the statistical significance of each discriminator or
predictor in differentiating between the groups. Variables which are not significant are
not considered as important in the interpretation.
3. Eigenvalue: Indicates the ratio of between group variation to within group variation. A
higher Eigenvalue is always looked at.
4. Wilk’s Lambda: Somewhat dissimilar to eigenvalue as it is the ratio of within group
variation to total group variation. We want this value to be closer to Zero and a
statistically significant based on Chi Square distribution (Therefore the overall
discriminant model significance is assessed by Chi Square value).
5. Structure Coefficient: A measure of association between discriminant score and each
of the predictors . This value is used in reporting discriminant analysis results. Higher the
value, greater the discriminating ability of the predictor.
6. Hit Ratio: Talks about the success of model developed by reclassifying the given cases
into correct and incorrect groups. A value of 0.7 is recommended for considering the
reliability and accuracy of the model in classifying and predicting cases(respondents)
into their respective groups(biased v/s non-biased)

Discriminant Analysis test results are quoted in the following format :-


Predictors G1 G2 Structure F Statistics
coefficient
Mean Std Dev Mean Std Dev

X1 *

X2

X3

X4

Constants: ________________________ Eigenvalues:


_____________________________
Canonical Correlation: _______________ Canonical Co Square:
______________________
Wilk’s Lambda: ______________ Chi Square: ____________ Degree of Freedom:
__________
Hit Ratio: ___________

Q. The client is interested in learning how families that visited the resort are different from those
who have not visited the resort.

Steps:-

1. Analyze -> Classify -> Discriminant

2. Statistics window
3. Classify Window

Output:

Canonical Discriminant Function Coefficients


Function table gives value of constant
Eigenvalues table gives value of eigen value and canonical Correlation

Wiki’s Lambda table give value of wiki lambda

Classification Result table gives hit ratio (last line below the table..from a.)

Filling the Discriminant Analysis table:-


Mean And Std Dev from Group Statistics Table
Structure Coeff from Structure Matrix- Star those whose values of significance is <0.05 in the
“test of equality of group Means” table. Those values are significantly different.
F Statistics from Tests of Equality of Group Means

From group statistics table


Predictors G1 G2 Structure F Statistics
coefficient
Mean Std Dev Mean Std Dev

Income 60.52 9.8307 to be taken from F column’si


structure matrix value from
table Tests of
Equality of
Group
Means

X2 Star those
whose values of
significance is
<0.05 in the
“test of equality
of group Means”
table

X3

X4

What percent of discrimination is explained by these 5 variables?


Ans. Square of canonical correlation coefficient = Variance of Canonical Correlation

Wilk’s Lambda talks about homogeneity of objects on the basis of the given variables.
How away each object is in this particular group -> extent of within group variation
We want to minimise this

Structure Coefficient: Function value from Structure Matrix: Higher the value, higher the
responsibility for difference between the groups

First variable is highest discriminator, second less so etc. and the last is the lowest discriminator

In ANOVA table we’re checking the value of each variable: should be > 0.05

Hit-Ratio: Percentage of cases which are correctly classified, i.e. data predicted matches actual
data

Pooled within-groups correlations between discriminating variables and standardized canonical


discriminant functions
Variables ordered by absolute size of correlation within function.

Report
Give small introduction about the context
Manager of the resort is interested in knowing how families are different who visit the
resort
Explain the data set
Explain the different variables given.

Once you introduce a context, by respecifying the research problem, you identify the process.

E.g.

Discriminant analysis

The findings are reported in Table 1. Table 1 is presented:-

Continue writing remaining interpretation.

We will begin with overall model. Mention all the below-table values.

All the 5 variables are found to be influencing group differences to the extent of 64% (Canonical
correlation square).

The discriminant model gives a hit ratio of 90% which is higher than minimum required value of
0.7. This indicates 90% families are correctly classified into who have visited and not visited the
resort

Now we need to interpret the findings.

In respect of income, attitude vacation, household size it is evident that families who have
visitied the resorts is more than who have not visited.(Also write the values)
Others are not significantly different.

Thus marketing strategy should be focused on the variables (both discriminant and non
discriminant)

Prediction is done based on the group-centroid ( From Functions at Group Centroid), i.e.
1.291 - 1.291. If average is positive, next family is predicted to go to positive side.

SNR Items Deleted Reason

1. CI_1 CL

2. CI_8 SL
FORCED FACTOR

5 BI_4 ML

6 BI_3 CL

IRT
Column Name Those in the same column Alpha = Value of reliability
C1 PI_1,2,3 .854

C2 BI_1,2 .875

C3 CI_2,3 .811

C4 CI_5,4 .782

C1 = Satisfaction
C2 = Reliability
C3 = People

Techniques:

1. ANOVA
2. ANCOVA
3. Multiple Regression

1. ANOVA
Technique used to study group differences in mean
It is based on F-Distribution
If ANOVA model is significant it is followed by a post hoc test as recommended by Tukey
to study which group is different from which other group
The o/p of ANOVA is reported in the following format

BI_Trust: Descriptive Statistics and ANOVA Test results on Brand Image by


Preferred Brand
Groups mean SD F-value

G1: Iphone 9.3525 .89170 * mark if


significant
G2: Nokia 9.0769 1.15587 64.039

G3 7.7818 1.49010
* Significant @ 0.5 level

Tukey’s table (Mean comparison table)

G1 G2 G3

G1 Mark * whichever
cell is significant

G2

G3
* Significant @ 0.5 level

Q. From the country image data, find whether there is any significant difference in the
mean brand image for customers who intend to buy iPhone, Nokia, Samsung brand of
phones?

Here there are two variables - Brand image and Brand name

ANOVA requires you to have a metric-dependent variable and a categorical or nominal-


independent variable

Brand image is based on metric scale, Brand Type is categorical

Step 1: This particular case involve a constraint/abstract/high level (brand image in this case).
You need to proceed with assessing dimensionality followed by consistency followed by
summating the score, followed by model testing followed by drawing inference and
reporting the results.

Step 2: In this case, country image data, Preferred brand variable - contains 7 different brand,
This analysis requires only 3. So select these three brands by selecting cases.

Step 3: Do factor analysis for getting dimensionality

Rotated Component Matrixa


Component
1 2
BI_2 .927
BI_1 .902
BI_4 .919
BI_3 .798

Extraction Method: Principal Component Analysis.


Rotation Method: Varimax with Kaiser Normalization.
a Rotation converged in 3 iterations.

Step 4: Get Reliability

BI_1, BI_2
Reliability Statistics
Cronbach's Alpha N of Items
.851 2

BI_4, BI_3
Reliability Statistics
Cronbach's Alpha N of Items
.715 2

Analyze -> Compare Means -> One way Annova


Both BI as Dependent list
Factor as Preferred_brand

Post Hoc-> Tukey

Option -> Descriptive, Means plot


Descriptive Statistics and ANOVA test result on brand image by preferred brand
The o/p of ANOVA (mean and std) are from Descriptives table and F value from Anova
table

Groups mean SD (from std F-value


column)

iPhone 9.35 .89

64.03*
Nokia 9.079 1.15

Samsung 7.78 1.49

Tukey’s comparison test result between pairs of preferred brands (Post hoc multiple
comparison test result on brand image by preferred brand)
(From Tukey HSD table) Value taken from Multiple Comparison Table (mean
difference I-J) Only * values are noted, rest Not Significant

This table is for BI_Trust


IPhone nokia Samsung

iPhone -

Nokia N.S (Not significant) -

Samsung 1.57* 1.29* -

Since nokia and iphone are not significant -> they are one group and samsung is
another group
Samsung Nokia IPhone

Brand image trust perception of customers is significantly different for customers who
have got either iphone or samsung or nokia.

Inferences:
1. Bi_trust perception of customers is significantly different for the ones who have
got iphone, samsung or Nokia. The mean brand image score is significantly
different for customers. Df: (between groups;within groups) -(2;340).
2. Tukey’s post hoc test mean comparison test reveals that brand image of nokia
and iphone are perceived alike whereas brand image perception for samsung is
different from nokia and iphone

Observation - the perceived brand image for samsung is the least in comparison to nokia
and iphone.
ANOVA TABLE
This table is for BI_vALUE
Groups mean SD (from std F-value
column)

iPhone 8.8 .89

64.03*
Nokia 7.8 1.15

Samsung 8.6 1.49

Table for BI_Brand value

IPhone nokia Samsung

iPhone -

Nokia 0.995* -

Samsung 1.57* 0.753* -

2. ANCOVA - analysis of covariance - An extension of anova wherein covariates


An extension of ANOVA where in covariates are controlled for(their effects are removed before
you study the main effects of independent variable).

Condition: Covariate has to be metric variable. That means it should not be categorical

Illustration:-
Apply the question in the previous illustration and controlled for the covariate Age:-
Check whether Age is categorical variable (if age would have been given in a band -> eg 25-30)
or continuous variable (exact age is given eg. 25)?
Age is continuous, therefore we can perform ANCOVA.

3 variables are there in ANCOVA

1. Brand Image
2. (indep variable) Preferred Brand - Independent variable
3. Covariate - has to be metric
Apply ANCOVA - Univariate

ANCOVA is found in GLM (General Linear Model)


Corrected mean is the mean after controlling for covariate

Free to present uncorrected mean -> Give footnotes saying which mean has been taken (mean,
corrected or covariate, make assumptions explicit)

Is this model significant?

You find Corrected Model - i.e. after eliminating customer age, is brand image perception still
significantly different for different brands of purchase? (check “significance value” for “corrected
Model” row in “Tests of Between-subject effects” table .. if its still <0.05 then the difference is
still significant.

In Multiple Comparisons table: If values of sig > 0.05 then there is no impact of one variable on
other
First inference:-

Even after controlling for customer's age, there is evidence of significant mean difference
among brand image perception of the three brands selected - Nokia, iPhone & Samsung

Second inference:-
Check “Tests of Between-subject effects” table
You find that there is one more value, partial eta square (partial = after removing effect of
covariate) (0.277 in our case)

This means that the impact of brand type on brand image perception is to the tune of 27.7%
after controlling for customer’s age, i.e. Age and Brand Name together account for 27.7% of
brand image perception

Unique contribution/impact of Brand Name alone: 27.6%


Eta square for Brand Name alone = 27.6 percent (0.276 in same table)

Q. What is the contribution of Brand Name on Brand Image for the three brands chosen without
controlling for any variables (i.e no covariates) ?

A. Compute Hay’s Omega square = between group sum of sq/total group sum of sq = 27.3
These values need to be taken from ANOVA table

Q. How much impact is explained on customer’s perception of Brand Image is due to Brand
Name without controlling any of the covariates (i.e absolute impact)?
A. Since after taking age and without taking it there is not much difference (27.3 and 27.6)
the role of age is not that much in analysing brand trust

ANCOVA TABLE
For Brand Value :-
Eta square for both Age and Brand Type together = 7.8%
Eta square for only Brand Type (after controlling age) = 6.8%
Omega sq for impact of Brand Type on Brand Image without controlling any of the covariates
(i.e absolute impact between group sum of sq/total group sum of sq)? 5.73%

Inference - Since with age the impact is 7.8% which is > 5.73% which is without age Thus age is
a significant variable.
In brand value - nokia alone in one segment. For brand trust nokia is considered superior but
poor in brand value

Practice Question

Q. Test the impact of Preferred brand(Nokia, samsung and Iphone)(Independent) on purchase


Intention(outcome variable) after controlling for country of origin (COO) and consumer
ethnocentrism (CE)

Regression analysis

● A technique used for testing the impact (overall impact - R2, absolute impact of each
predictor - B, relative impact of each predictor - Beta)
● The overall model is tested based on F statistic
● The significance of each predictor is tested based on T-statistic
● Severe correlation among predictors is tested based on Tolerance whose value has to
be not less than 0.2 and VIF (Variance Inflation Factor) whose value should not exceed
4
● Adjusted Rsquare stands for the minimum Rsquare one can expect by selecting another
sample from same population.
● Regression model is as follows
○ y^ = alpha + b1X1 + B2X2 + … + BnXn
○ Alpha is constant
○ B1,B2..Bn = absolutive value of each factor
○ X1, X2 = predictors
○ Y^ = Predicted outcome variable
Regression results are reported in following format

Predictor Mean SD Beta T-value Collinearity

X1 * for
significant
value

X2

X3
○ * significant @ 0.05 level
○ R = __________ R^2 = ___________ Adjusted R^2 = ________________

Illustrations:

Find out the impact of Country image, brand image and country of origin on PI.
Dependent PI and independent are CI and BI

Linear Regression
After factor analysis on these things do regression analysis
Analyze -> Regression -> Linear regression
Statistics -> select descriptives and collinearity
Check Also Descriptives
in ANNOVA Table
If overall model is significant then at least one is significant

In Coefficient Table
Tolerant should not be <0.2
VF should not be less than 0.4
And Report B and Beta value

In Model Summary Table


R stands for multiple correlation
R square stands for how all the COO,CI,BI are affecting the PI
r stands for
Adjusted r2 = 0.255 → if
1 - R^2 -> impact of extraneous (like noise) variables not considered in the model.

Predictor Mean SD Beta T-value Collinearity

X1 * for
significant
value

X2

X3
○ * significant @ 0.05 level
○ R = __________ R^2 = ___________ Adjusted R^2 = ________________

Take home assignment - 1

Your client wants you to assess the impact of employee's LMX, JS, & OCB on TI for those
located in Mumbai branch with at least 5 years of experience.
(LMX Data file)
Select cases based on Mumbai branch and at least 5 years of experience

Predictor Mean SD Beta T-value Collinearity

X1 * for
significant
value

X2

X3
○ * significant @ 0.05 level
○ R = __________ R^2 = ___________ Adjusted R^2 = ________________

PARTIAL CORRELATION:
Also tick the Mean and Std Deviation checkbox
NEW TECHNIQUE- PARTIAL CORRELATION

Use the country image data set


1. How country image(CI), age and brand image (BI) influence the brand purchase
intention (PI)
2. If the mean brand image is different for these different brand of mobile phones
3. If there is a relation between brand image and purchase intention after controlling for the
age of customers
Note: Use summated score approach. For assessing dimensionality, your client requires you to
take the following items: PI1, PI2, PI3, CL2, CL3, CL4, CL5, BI1, BI2. Your client further
requires you to assess the dimensionality of purchase intention separately; and a separate
dimensionality assessment for the rest of the given items. Present a report to your client.

Solution

Predictor Mean SD Beta T-value Collinearity

CI_People 12.7699 2.95389 0.075 1.815 1.082

BI_Reliability 8.3121 1.73221 .520 12.369* 1.117

Age 29.66 5.189 .243 6.014* 1.034

○ R = 0.558 R^2 = 0.311 Adjusted R^2 = 0.307

2) ANOVA

For preferred Brand


Groups mean SD (from std F-value
column)

iPhone 9.3525 0.891 28.364

Blackberry 7.769 1.912

Micromax 5.625 1.526

Nokia 9.0769 1.155

Sony 10.00 0
HTC 8.0 1.732

Samsung 7.7818181 1.49

Others 8.1428 2.81

IPhone Blackb Microm Nokia Sony HTC Samsu Others


erry ax ng

iPhone - - - - - - - -

Blackb 1.583* - - - - - - -
erry

Microm 3.72* 2.14* - - - - - -


ax

Nokia .27559 -1.307 - - - - - -

Sony -.6475 -2.230 - - - -

HTC 1.32 -2.230 - - -

Samsu 1.57* -.0125 - -


ng

Others 1.209* .373 -

For Existing Brand

Groups mean SD (from std F-value


column)

iPhone 8.9038 1.68336

Blackberry 8.7636 1.05345

Micromax 6.0000 .000000


4.577
Nokia 8.4286 1.59714

Sony 8.5500 1.14593


HTC 7.8889 2.60670

Samsung 8.0813 1.65754

Others 7.6000 2.37225

IPhone Blackb Microm Nokia Sony HTC Samsu Others


erry ax ng

iPhone -

Blackb N.S -
erry (Not
signific
ant)

Microm 2.9038 - -
ax 5* 2.7636
4*

Nokia -

Sony -

HTC - - 1.8888 - -
1.0149 0.8747 9 0.5396
6 5 8

Samsu - - 2.0813 - - 0.1924 -


ng 0.8225 0.6823 0 0.3472 0.4687 1
5 4 7 0

Others - - 1.6 - -0.95 - -0.4831 -


1.3038 1.1636 0.8285 0.2888
5 4 7 9

iiii) PARTIAL CORRELATION


Analyze-> Correlate-> Partial Correlation
Options-> Check 1)Mean & Std Dev
2) Zero Order Correlation

Predictor Mean SD

PI_BUY 11.6150 2.40715

BI_QUALITY 8.3121 1.73221

Age 29.6629 5.18914

Which Test to use in which situation

dependent independent covariate Analysis

metric metric - Regression


(Impact)

metric nominal/ - Anova (Mean


categorical Difference)

metric metric metric Hierarchical


regression
analysis
(Impact)

metric nominal metric Ancova

metric metric Non-metric Not discussed

metric Nominal Nominal Not discussed

categorical metric - Discriminant

Two variables which are Nominal/Categorical(like Chi-Squared


Gender) in nature, we may or may not know
dependent/independent relationship. When we want
to test their association
Two variables which are nominal in nature, and we Phi Coefficient
want to compute the magnitude of association or Cramer’s
Coefficient

Two variables which are metric in nature and you Correlation


want to study their relationship/association

Two variables which are metric, and you want to Partial


study their relationship, where there is a covariate Correlation
that is also metric

When all variables are metric and researcher’s task Factor Analysis
is to achieve parsimony or data reduction,
dimensionality or validity

When you want to assess the extent of measurement Reliability


error in the data you collected from the field survey on
metric variables

Two types of errors Systematic error


(validity) and
measurement/R
andom error
(Reliability)

We try to assess (IMP -- not establishing validity but assessing validity/ DONOT ASSESS
RELIABILITY ALONG WITH VALIDITY) systematic error through Factor Analysis.

● Of all the significant ones, look for the value of Beta to identify the relatively more
significant.
● Tolerance <0.2 indicates multi-collinearity (means severe overlap between two
variables) -> Thus those should be removed. Do model testing again after removing that
variable.
● If VIF > 0.4 remove that variable. Do model testing again after removing that variable.

Q. Which of the following types of research helps in generating the hypothesis?

a. Exploratory
b. Descriptive
c. Experimental (looks for testing the effect)
d. Longitudinal
Ans . Mostly D/ (A-GOOGLE) Refer session 1 notes

Q. Which of the following is definitely not a formulated research?


a. Exploratory
b. Descriptive
c. Experimental (looks for testing the effect)
d. Longitudinal
e. Both C and D

Ans . E

Q. Incorporating additional predictor in regression model always increases R^2-


Ans. mostly FALSE (once check)

Q. Using factor analysis it is not feasible to assess the extent of measurement error?
True

Q. Reliability guarantees validity (False); validity also guarantee reliability (False)


False (check)

Q. If a scale of questionnaire is reliable, it must always be valid (T/F)

Q. If a scale of questionnaire is reliable, it may or may not be valid (T/F)

Q. Testing of diff in mean b/w 3 or more groups is followed by a Tukey test? - T/F

Q. A factor solution yielded four factors with the Eigenvalues as follows:-


Eigenvalues: 2.7, 2.3, 1.9, 1.1.How many factors are extracted
1. 4
2. 3
3. 2
4. Cannot be decided

Q. A factor solution yielded four factors with the Eigenvalues as follows:-


Eigenvalues: 2.7, 2.3, 1.9, 1.01.
Q. Which of the following is the least important
Highest eigen value -> Most imp

Q. Factor analysis varimax rotation yields that these factors are:


a. Independent
b. Orthogonal
c. Interrelated
d. Insufficient information
e. Both A and B

Q. Correlation b/w 2 factor scores (varimax rotated within the same factor solution) will always
be equal to 0
Yes (mostly)

Q. Researcher wanted to find the correlation between PI and BI, after controlling for both age
and income (both being metric measures)

a. First-order correlation
b. Second-order correlation (When 2 covariate its 2nd order)
c. Third-order correlation
d. Zero-order correlation (Also called pearson correlation/product moment correlation, not
controlled for any of the covariates)

Q. Which of the following means a non-metric in nature?


a. Nominal
b. Interval
c. Ratio
d. Ordinal
e. Both A & B

Ans. Both A & B

Q. Whereas the correlation b/w 2 factor scores is assumed to be 0; the correlation between the
corresponding surrogate variables from these factors is not = 0 .. Why?
Ans.

Q. Which of the following is not a comparative scale?


a. Paired comparison scale
b. Rank order scale
c. Likert Scale (Strongly disagree to strongly agree)
d. Constant Sum Scale
Ans. D Constant Sum Scale (check)

Q. Which of the following scale contains only the adjectives at the extreme stages
a. Likert
b. Constant sum scale
c. Semantic scale (Spacious, somewhat spacious, somewhat cramped, cramped)
d. Staple scale
Q. Which of the following scale is vertical in nature?
Ans. Staple scale
Hierarchical Regression Analysis (HRA)

Considering = Controlling for / after eliminating

For such questions, we use HRA

Q. What do you get in HRA?

Ans. Your intention is to find out the incremental contribution of a variable over and above the
variable that you control for. You explain the variance in the outcome variable.
Denoted by ΔR2

Use the country image data set.

1. How country image (CI) and Brand Image (BI) influence the brand purchase intention
(PI) (dependent var), after considering the customers’ age [use HRA]
Covariate age (Should be metric measured)
If Covariate is nominal then model cannot be tested

Put covariate under Independent variables, here, age, then press next. Then put CI and BI
again in the independent variables.
Chi s

R^2= 0.289

Check if Delta R^2 is significant or not (Sig @ <0.05)

Hierarchical Test analysis Result of Country Image, brand image on Brand PI with Age as
covariate

Predictor Mean Std dev Beta T R Square

Model 1 From From From From From Model


Covariate: Descriptive Descriptive Coefficients Coefficients Summary
Age Statistics Statistics table table table

Model 2 - -
Age - -
CI 0.311
BI (Single Value
in this column
per group)

From Model Summary Table


Delta R^2: 0.289
Delta F: 91.217
Sig @ <0.05 level

Test the impact of Brand Image on Purchase Intention after controlling for customer age and
Country Image

Covariate Impact- (1st R^2 data in the Model Summary Table)


Incremental Contribution of BI -(2nd R^2 data in the Model Summary Table)

Chi Square Test

Q. Do you find any evidence of association of CI and Brand PI upon classifying them into low
and high categories. If there is an association compute the magnitude of such association.

Higher-order variable can be converted to lower-order variable, but not vice-versa


Ans :-
Hint: Both variables are metric.. But association word is there which is used for non metric
variables
Step 1:- Convert Higher order to lower order by using separation through median
Divide the variables above median is high..below median is low. Thus converting metric variable
into nominal variable
Step 2:- Analysis-> Descriptive-> Crosstab
Dependent in Row, hence PI in Row:
Check Chi-square and Pearson and Cramer V in statistics

Strength of association (From table Symmetric Measures) is sig@<0.05 level


Cramers V Ranges between 0 to 1 ( 0 being low)
There is low association however the association is significant

How to Present:
Draw the cross-table
Chi squared value= ___ ; Df=_____ ; Pearson (Phi value) = _____

Phi value : Less than 0.4 low association


0.4- 0.6 Moderate
0.6- 0.8 High
Above 0.8 Very strong Forced Fit solution possible results and remedies :
Possibilities

1. In a constrained model, if dependent construct is still inseparable, the following remedy is


available:
Perform a separate factor analysis for dependent and another separate factor analysis on all the
items of independent and control constructs (together)

2. In a constrained model, dependent variable is clear and 2 independent constructs form a


separate component, the following remedy is available:
Perform a separate factor analysis on dependent construct and another separate analysis on
each of the independent and control constructs
3. In a constrained model, either dependent or one of the independent or control constructs is
lost, then the following remedy is available:
Perform a separate factor analysis on each of the constructs.

4. While performing the dimensionality assessment, if you’re unable to obtain a Rotated


Component Matrix, this would imply that the variables formed unidimensional solution.
Unidimensionality is acceptable if all variables belong to a single construct. On the other hand,
if rotation is not converged due to lack of iterations, we manually reset the number of iterations
from the default setting of 25 anywhere up to 9,999 to obtain a Rotated Component Solution. If
even exhausting this does not yield a RCM, we resort to performing a factor analysis separately
on each construct.

Practice on LMX data:-

Test the impact of executives’ job satisfaction and LMX on their Intention to Quit (IQ) the
organisation for executives located in Mumbai plant with at least 5 years of experience, after
controlling for their age and OCB.

How are executives in Low IQ different from High IQ categories on the basis of their JS, LMX,
OCB and age.

Note: Use summated score approach; There is no need to reverse-score anyh of the variables;
Present your report with relevant information and table; Write clearly.
Do not show the workings.

Using directly this ?


Yes - Arushi & Shreya, but why?
Yes but in remedY 1 u take independent and covariate 2gthr Okay so it remains same ? yes
And in remedy 3 ?

Remedy 1 result mein which variable is deletd?


No no sahi hai

IQ 1,2 IQ_Positive

LMX 3,4,5,6,7 LMX_Support


LMX 1,2 LMX_Power
JS 1,2,4,5 JS_Pleasant

OCB 8,14,10,12,4 OCB_Loyalty


OCB 5,6,16 OCB_Reputation
OCB 3,11,1 OCB_HelpOther_T
OCB 13,7 OCB_ExtraEffort_T

Do we have to remove OCB 15 yes? Or something else 15 only..


Coz i dont remeber last time krke dekh ek baar...1 baar me aa jayega

Variables names different likho no


Itz fi8ne

ITS HRA na ?
Descriptive Statistics
Mean Std. Deviation N
IQ_T 4.5983 2.46829 351
AGE 34.69 6.734 351
JS_T 13.4587 2.27103 351
LMX_Trust 6.2536 2.09519 351
LMX_Favour 19.2821 4.50621 351
OCB_Affection 22.7179 2.46407 351
OCB_Image 12.6809 1.94368 351
OCB_Help 12.7151 1.74643 351
OCB_Assist 8.3704 1.40799 351

Tests of Between-Subjects Effects


Dependent Variable: IQ

2
Source Type III Sum of Squares df Mean Square F Sig. Partial Eta Squared
Corrected Model 214.703a 4 53.676 39.083 .000 .200
Intercept 767.798 1 767.798 559.060 .000 .472
JS1 160.919 1 160.919 117.171 .000 .158
Exp_New 31.188 3 10.396 7.570 .000 .035
Error 859.731 626 1.373
Total 4607.000 631
Corrected Total 1074.434 630
a R Squared = .200 (Adjusted R Squared = .195)
OYE
SURROGATE VALUE IS IQ2??
Yes
Anwer
Omega square - .01682
R square - .315
Adjusted R square - .312

Before removing covariate, eta square - .315


After removing - .304

Das könnte Ihnen auch gefallen