Workshop Notes
Workshop Notes
2. EFA (SPSS)
3. CFA (SEM)
5. Correlation
8. Mode Graph
BY:
Gulfam Murtaza (PHD Scholar) Basharat Javed (PHD Scholar)
Why?
If you are missing much of your data, this can cause several problems;
e.g., Means will be condensed and will not give accurate estimates
When?
Missing less than 10% from a variable or respondent is typically not problematic (I prefer less
than 5%).
How?
Option 1: Use only valid data. No imputation, just use valid cases or variables
Option 2: Use known replacement values. Match missing value with similar case’s values.
Option 3: Use calculated replacement values. For this mean or median would be the choice.
Treatment
1. Transform -> Replace missing values -> box opens
4. In “Method” box, select median for categorical variables and mean for continuous
variables
Note :“values and measures” of all the variables or items containing missing data should be
re-adjusted in variable view.
Best Method – Prevention!
FACTOR ANALYSIS
1. Exploratory Factor Analysis
Method:
2. Shift all items of the all variables in “variables” box. (demographics are excluded)
Note: if we want to take forced choice of fixed no. of factors then the option “fixed no. of
factors” would be selected.
7. Continue > ok
Adequacy
Two statistics on the SPSS output allow you to look at some of the basic
assumptions.
Kaiser-Meyer-Olkin (KMO)
It also tells whether the variables are able to be grouped into a smaller set of
underlying factors or not
AMOS
1.Open the amos from spss
(as amos is opened from spss, data file combine 3 is automatically linked with amos, otherwise if
we open amos from its short cut button, we have to link the file with it) the path would be as
follows:
ii.click on “select data file” button from tool box > select your data file
Safety
Abusive Psychological
Outcomes (SO)
supervision (AS) Distress (PD)
1.Abusive supervision
1. Select Draw tool from the tool box and draw an oval for latent variable and its
observed variables by clicking (5times in our case) the same tool in oval.
2. Take select tool from the tool box and double click in oval to write name of the
variable, box opens, type name (Abusive supervision)
3. For naming the error terms, click Plugins > name unobserved variables
4. For loading item in observable variables, take List variable tool from the tool box, it
shows the list of all items. Select AS1 to AS5 and drop them one by one in rectangles
5. For moving a complete variable, select two tools from tool box, Move Objects tool
and Preserve symmetries toll , now you can move this variable on the canvas
6. For rotating the variable, use rotate tool , this is used to set many variables on the
canvas
8. Draw all the variables, place them in sequence, connect them through Covariance tool
o Modification Indices
Model Fit Indices and their cut off values are given below
Goodness of fit index(GFI)
Adjusted Goodness of Fit Index (AGFI)
Normed Fit Index (NFI)
Comparative Fit Index (CFI)
Incremental Fit Index (IFI)
Tucker Lewis Index (TLI)
Root mean Square Error of Approximation (RMSEA)
Root Mean Square Residual (RMR)
Initial solution is close to the cutoff values (FIG1). Model fitness can be improved by joining the
error terms with greater values from modification indices (FIG 2). We get the final solution as
under:
MODIFICATION INDICES
Par
M.I. Change
e23 <--> SaftOut 4.368 -0.047
e23 <--> PsyDistt 9.618 0.062
e23 <--> e24 38.354 0.139
e22 <--> e24 11.486 -0.058
e21 <--> e23 23.196 -0.084
e21 <--> e22 11.689 0.045
e20 <--> AbuSup 9.874 0.1
e19 <--> PsySafty 7.683 -0.093
e19 <--> e23 4.038 0.05
e18 <--> SaftOut 6.255 -0.055
Measurement models CMIN /DF CFI GFI AGFI RMR NFI IFI TLI RMSEA
Threshold values <3 >0.95 >.95 >0.8 <0.09 >0.9 >0.9 >0.9 .05-0.1
Initial 4 factor solution 3.081 0.90 0.82 0.78 0.11 0.91 0.95 0.94 0.08
Final 4 factor solution 2.05 0.95 0.88 0.85 0.10 0.92 0.95 0.95 0.06
Most of the estimates lie within threshold values so model fit is good but
we have to compare it with other models for example 3 factor models, 2
factor models, 1 factor model etc.
There are many ways to pair the variables in different sets. Here our
variables are as follows:
3 FACTOR MODELS
2 FACTOR MODELS
3 Factor Model
There are many ways to pair the variables in different sets. Here our variables are as follows:
NOTE: always give regression weight to newly merged variables otherwise SEM will not
calculate estimates
3Factor models
2 factors models
1. AS-PD PS-SO
2. AS-PS PD-SO
3. AS-SO PD-PS
1 factor model
The results of these different models are summarized in a table with few estimates (here below,
in the table we presented almost all). If our proposed mode (here 4 factor model is our proposed
model) shows more fit estimates than other models, this makes our argument strong for the
fitness of model.
Measurement models CMIN /DF CFI GFI AGFI NFI IFI TLI RMSEA
Threshold values <3 >0.95 >.95 >0.8 >0.9 >0.9 >0.9 .05-0.1
Initial 4 factor solution 3.08 0.90 0.82 0.78 0.91 0.95 0.94 0.08
Final 4 factor solution 2.05 0.95 0.88 0.85 0.92 0.95 0.95 0.06
3Factor AS-PD 4.47 .83 .71 .65 .79 .83 .81 .10
3Factor AS-PS 4.80 .81 .71 .65 .78 .82 .79 .11
3Factor AS-SO 7.63 .67 .57 .48 .65 .68 .64 .15
3Factor PD-PS 8.48 .63 .42 .38 .60 .63 .59 .16
3Factor PS-SO 5.0 .80 .67 .60 .76 .80 .78 .11
2Factor AS-PD, PS-SO 8.91 .60 .53 .44 .58 .61 .57 .16
1Factor model 13.39 .39 .38 .26 .37 .39 .33 .20
From the above table, model fit indices clearly point out fitness of 4 factor model is greater than
other models through statistical estimates.
Now after cfa, clear picture of variables is evoked. It’s the right time to get reliability test.
Method:
Note: In “reliability statics table” of output, cronbach’s alpha’s value is mentioned. If the
alpha’s value of any variable is smaller than 0.7, the table “item-total statistics” last column
shows what item should be deleted to improve the reliability and to what level it will increase the
alpha’s value. It is the option for researcher to delete a certain item for improving reliability
Reliability is good now our data is ready for regression analysis. But before regression, we have
to calculate means. In spss, transform > compute variable command will be used for calculating
composite variables. The main point to remember is that only those items will be included which
are surfaced during EFA and CFA.
First calculate the means of those variables needed to calculate interaction term. In our case,
Psychological Safety Climate (PS) is the moderator and Abusive Supervision (AS) is
independent variable. So mean of these variables will be calculated for centering. For mean
calculations: method
3. ok
Statistics
AS PS
For centering
1. Transform >compute variables
2. shift the variable (Composite) e.g., AS in right box,
3. Name the variable (e.g., ASC)
4. Place the minus sign, type the mean calculated and click ok. (same for PS)
For Interaction Term
1. Transform > compute variables
2. Name the variable (e.g., ASCxPSC)
3. Shift ASC in the box, place * sign, shift PSC
4. ok
This test is performed to confirm the demographic variables with dependent variables. If the
results are significant, we have to carry that particular demographic variable to our next analysis
of correlation and regression.
Method
2. Shift dependent variable and categorical/demographic variables one by one after every
reading
3. Ok
Note : For categorical variable Gender, the result of spss is below in the table. Perform
tests for all demographic variables. The results of all variables have been summarized in
the second table.
ANOVA
SO
The second table shows that only AGE is significant in One Way ANOVA. So we shall carry
only age in our correlation and regression analysis.
This analysis explains the co-variance between two variables. This test shows that whether two
variables positively or negatively associated with one another. Positive correlation indicates that
both variables move in the same direction i.e., if one variables increases, the other will also
increase. While negative correlation refer to the fact that if one variable increases, the other
variable decreases and is indicated by negative sign with the co-efficient value. It is important to
note that correlation test just explains the association between variables but not the causal
impact.
For magnitude / Value, it must lie within 0.3 to 0.7. Lesser will show low correlation between
variables and greater will indicate multicollinearity issue in IVs.
Method
2. Shift composite variables in the box along with demographic variables (here in our case it
is only AGE)
AGE 1
PD -0.29*** 0.21*** 1
REGRESSION (SEM)
2. Link the amos file with spss data file (procedure explained above)
3. Change the orientation of the page form view > interface properties > landscape > apply
4. Click on list variable button from toolbox, it will show the list of variables.
Drop composite variables on the sheet through drag and drop method
6. Position the variables on the canvas of amos as shown is figure below. (use move tool)
8. For regression, add error terms to all DV’s and Mediators (select Add a Unique Variable
10. Join IV’s with Mediator and DV with Draw Path Tool
11. Joint demographic variables (AGE) with DV (SO) with Draw path tool
12. Name the error terms (Plugins > Name unobserved variables)
b. Output > standard estimates > 1. modification indices, 2. Indirect / direct total
effects
15. Take Calculate estimates tool from tool box > proceed with analysis it will
calculate the estimates.
16. Click on view text from tool box, it will show the output file.
17. For bootstrap effect from output, click Estimates > matrices > standard indirect effect
AND from Estimates bootstrap > bootstrap confidence > two tailed method
It will just show sig or not sig. here in our case the value is (0.004) which is significant
i.e., mediation exists. Without a mediator, there is no effect of AS on SO but the only
effect exists through mediator.
Grouping
Creating groups for moderation, mediation e.g., we want to get a separate results for male
and females or in different age groups etc
In a box, type the name of the group…e.g., male.. than click new button and type
“female” and close the box
Click of “select data file” button from the tools and select data file for both male
and female through “FILE NAME” button
For attaching grouping variables to both the groups, click on each group and than
through “grouping variable” button attach grouping variable e.g., gender to both
the male and female
For assigning value to the group, click on ‘male’ and double click on ‘1’ from the
box and ‘2’ for female from the box…. Hit ok