CE 452: Transportation Engineering Lab II
Lecture: 11
Structural equation modeling
Course Teacher: Saurav Barua (SB)
Assistant Professor, Dept. of CE, DIU
Email: [Link]@[Link]
Phone: 01715334075
Outline of the lecture
Introduction to SEM
Terminology and Basics of SEM
Jamovi v.2.3.28 Statistical software
User interface and Jamovi modules
SEM data processing in Excel
Transfer data to Jamovi and analysis
Output and interpretation of SEM
What is SEM?
Structural equation modeling (SEM) is a multivariate statistical
analysis technique that is used to analyze structural
relationships. This technique is the combination of factor
analysis and multiple regression analysis, and it is used to
analyze the structural relationship between measured variables
and latent constructs.
Terminology and Basics of SEM
Terminology and Basics of SEM
Terminology and Basics of SEM
Terminology and Basics of SEM
Jamovi Statistical Software
Download and installation
Download and website link: [Link]
User interface
Modules for SEM
Analysis tab ->
Modules + ->
select Jamovi
Library from
dropdown
Install “semlj-
SEM”
Raw data Question survey data collected for train
riding service quality (SQ) using
questionnaire form
Demographic variables are Age and
Gender
Service Quality variables are station
seating condition, station information
service, train physical appearance, train
fair and train seat quality.
Data processing in Excel
Assign Dummy
variable for
string data
Import or copy data to Jamovi
Assigning variable title (Column)
->
Variables
tab
Click
checkbox
any and
entry name
in ‘Data
variable’
window
Run SEM
-> Analyses
tab
Click SEM
Choose
SEM
(interactive)
Run SEM
Assign Endogenous and Exogenous variables
Using arrow tab include
variables from left-most list
Click Add new latent to add
new variables
Endogenous 1 are Sta_seat,
Sta_info
Endogenous 2 are Train_app,
Train_fare, Train_seat
Exogenous 1 is Age and
Exogenous 2 is gender.
Path diagram option
Below path diagram option
Click checkbox show residuals,
show intercepts
From Path radio button click
Betas
Model options
Keep default setting
Estimation method is
automatic.
If normally distributed data
then it will run maximum
likelihood (ML).
If sample size is small, model
is complex then, generalized
least square (GLS) is
preferable.
Parameter options
Keep default setting
Scale/standardize variables ->
Check Latent vars: Fix first
indicator (to 1) radio button
Output options
Under Additional outputs
-> Check Additional fit
measures, Mardia’s coefficient,
Reliability indices
Under Covariance and
correlations
-> Check Residual
Modification indices
-> Check Modification indices
Results
χ2 test
χ2 test measures similarity or
discrepancy between user model
and baseline model.
Whether or not model good fit
data is measured by χ2 test
depends on sample size
If p<0.001 there is similarity
between model and dataset.
Model value is good fit with
dataset.
Other statistical output
SRMR should be around 0.06 or
less
RMSEA should be <= 0.05
And RMSEA lie within lower and
upper 95% CI.
Additional output
CFI >= 0.9 is acceptable for
model fit
TLI >= 0.9 is acceptable for model
fit
Reliability statistics
Latent variable α should be
around 0.8 and
ω1, ω2 and ω3 value the
higher are the more preferable.
Latent variable should not have
negative α, ω values. If those
have negative values, the
model does not fit with dataset
and need to drop the variable or
reversing order is required.
Correlation of residual
The values of residual
correlation should be <0.5.
Look at the values of upper
triangle.
Here residual correlation is
explained by data in the
model.
Any correlation above 0.5,
need to investigate
descriptive statistics of
variables and drop
corresponding one variable
from model.
User interface
If residual correlation
>=0.5, look at the
standardized estimate β.
If any of the
corresponding variable
has small β, it should be
drop from the model.
Reverse order in any
variable having β
negative value.
Path diagram
The main cause of variance
in Trn_S, Trn_f and Tr_p is
latent variable Endg2.
Exgn1 variable represented
by St_nf and St_st.
Endg1 and Endg2 are
caused by (or predicted by)
Exgn1 and Exgn2 latent
variables.
Exogenous variables Exgn1
and Exgn2 cause variance
in the Endogenous
variables Endg1 and Endg2
.
Information regarding Path diagram
Latent variables are measured by observed variables i.e. those are test
items or questionnaire items,
Observed variable is represented by rectangle
Latent variable is represented by circle
Errors e is represented by triangle. Every factors got errors associated
with it.
Endogenous variables have correlation with each other which represent
by two headed arrow.
2- headed arrow indicates correlation.
1- headed arrow represent causation. It runs from endogenous variable
to observed variables represents cause effect relationship.
Example problem
SEM model for a dataset have been analyzed using Jamovi v.2.3.28.
The output and path diagram are given. Interpret following: (1) No. and
name of endogenous and exogenous variables, Comments on -(2) χ2
test, (3) SRMR, RMSEA, (4) CFI and TLI, (5) residual correlation, (6) β
value and (7) path diagram.
Recap of the lecture
SEM model terminology
Interpretation of SEM analysis
Comments on outputs
Path diagram
Example problem
Source data link: [Link]
7c03EWk1SxnzOlQgExTKu_aKoKF?usp=drive_link