15/1/2015

Structural Equation Modeling

(SEM) using AMOS
Prepared by
Anuar Mohd Mokhtar, Ng Wen Jin, Fatin Athirah, Nur Ifzan, Narozieta & Dr. Amir Hamzah Sharaai

Faculty of Environmental Studies, Universiti Putra Malaysia

Introduction
1.

SEM specifically designed to analysis quantitative data.

2.

Parametric test for normal distribution data.

3.

Use model testing method to examine the cause-effect relationships

between a group of variables in a research.

4.

Hypothesis model is tested to determine its compatibility with the

research data collected from the respondents.

5.

SEM analysis is a combination of path analysis and factor analysis.

6.

SEM can be analyze using AMOS, LISREL, and EQS.

7.

AMOS is the newest software by IBM and provide attractive graphic.

15/1/2015

Two main functions of SEM

a)

Alternative to multiple regression analysis, path analysis, factor analysis,

time series analysis, ANCOVA, MANOVA to determine relationships
between variables.

b)

Identifying tools

i.

Identify whether the relationships between variables proposed in the model

is correct among the research respondents.

ii.

Identify whether the pattern of variance and covariance in research data are
matched with hypothesis model using Chi-Square Goodness-of-fit, baseline
comparison, RMSEA and others.

c)

Model Development tools

i.

Combined identifying and exploring functions.

ii.

SEM will suggest new relationships if the model is not compatible with
research data.

Procedures in performing SEM

Designing hypothesis
model based on theory
Designing research
tools

Data collection

Hypothesis model
testing

Reporting analysis
result

15/1/2015

Characteristics of variables in SEM

Variable
Indicator Variable

Characteristics
Variable measured by research tools.
A.K.A. observed variable. In SEM, it is represented by a
square with 2 arrows pointed to it.

Unobserved Variable

Variable is not measured by the research tools. It represented

by an oval/circle in SEM.

Exogenous Variable

Independent Variable in the regression model of SEM. Oneway arrow pointed out of it.

Endogenous Variable

Dependent Variable in the regression model of SEM. Pointed

by a one-way arrow.

Latent Variable

It is not measured directly from the research. It is represented

by its indicator variables.

Variable

EXAMPLE

Indicator

EMOSI 1, EMOSI 2, EMOSI

Variable

3, MOTIVASI 1, MOTIVASI
2, MOTIVASI 3.

Unobserved

EMOSI, MOTIVASI, e1, e2,

Variable

e3, e4, e5, e6, z1

Exogenous

EMOSI, e1, e2, e3, e4, e5, e6,

Variable

z1

Endogenous

MOTIVASI,

Variable

EMOSI

EMOSI

1,

2, EMOSI

3,

MOTIVASI 1, MOTIVASI 2,
MOTIVASI 3.
Latent Variable EMOSI, MOTIVASI

15/1/2015

Conditions that need to be fulfill for

SEM
1. Normality multivariate:
Every indicator variables should be normally distributed.
2. Type of data:
Interval and ratio (known as scale in SPSS).
3. Sample size:
Depends on the numbers of parameters.
1 parameter = 10 respondents
4. Numbers of variables in regression model:
Most suitable = 3 - 5 latent variables.
1 latent variable = 3 5 indicator variables.

5. Linearity:
Relationships between endogenous and exogenous variables should

be linear relationships (to avoid bias).

6. Random sampling:
Samples must be choose randomly from the population.
7. Free indicators variables:
Items in the questionnaire should not represents more than one

indicator variable.

15/1/2015

Tutorial
Some researchers are trying to determine the factors that contributed to

the household carbon emission at a residential area in Penang.

52 out 60 residents were chosen using Krejcie and Morgan formula and

they were chosen by using simple random methods.

To determine the relationships, they used SEM to analyze the data.

Research Hypothesis
Null Hypothesis (HO):
There is no relationships between family size, energy consumption, and
transportation with their carbon emission.

Research Hypothesis (Ha):

There is a relationships between family size, energy consumption, and
transportation with their carbon emission.

15/1/2015

Regression Model
Hypothesis model for our study:
Household Carbon emission = b1(household) + b2
(electricity) + b3 (transportation) + other factors.

Data preparation
Step 1: Type in your data in SPSS and save them as
Tutorial 1.

15/1/2015

Step 2: Open up AMOS Graphics from your computer

Step 3: Click File, then choose Data Files

15/1/2015

Step 4: Then, click File Name to select the data set,

and choose the SPSS file Tutorial 1 that you have saved
before, then click Open.

15/1/2015

As you can see, the numbers of sample is displayed (N = 30/30)

Then, click OK

Step 5: Next, click on View and choose Variables in Dataset.

15/1/2015

A pop up will come up and it lists all the variables available in

your Tutorial 1 (SPSS file).

Step 6: Click on one of the variables (continue pressing your

left cursor)

10

15/1/2015

.and drag it into the AMOS window (release your cursor).

Now, the variable Household is inserted into the model.

Continue Step 6 until all the variables are placed into the
model.

11

15/1/2015

Step 7: To rearrange the placement of the variables, click Move

objects button at the left panel.

Arrange them according to the regression model that you have

suggested.

12

15/1/2015

Step 8: Now, start to draw the path that represents the

relationship between the variables by clicking Draw paths
button at the left panel.

Click on Household as the first point and drag the arrow until
it reaches CO2.

13

15/1/2015

Now, you already have the first path. Continue to draw the paths
for other exogenous variables ( Electricity and Transport).

Once you have finish, the model will looks like this.

14

15/1/2015

Step 9: Then, you can start to draw the relationships between the
exogenous variables. Click Draw covariances at the left panel.

Start by drawing the covariance from Transport to

Electricity.

15

15/1/2015

Continue to draw the covariances between all the exogenous

variables.

Step 10: Now, click on Add a unique variable at the left

panel to add an unobserved variable

16

15/1/2015

and click on CO2 until you fit it in the right position.

Step 11: Double click at the circle (unobserved variable) to

name the variable.

17

15/1/2015

The Object Properties window will comes up. Name the

variable as Residue and type in its label as e1.

Step 12: Now, all the variables are named and the model is completed.
Then, you need to save the model before you can analyze it. Click
Save button at the left panel.

18

15/1/2015

Save it as Tutorial 1.

Step 13: Then, click on View and select Analysis Properties

to choose the output of the analysis.

19

15/1/2015

Step 14: In the Estimation tab, make sure that Maximum likelihood
and Fit the saturated and independence models are selected.

Then, switch to Output tab (at the right of Estimation tab

20

15/1/2015

Step 15: Tick the Maximization history, Standardized estimates,

Squared multiple correlations, and Modification indices. Change the
Threshold for modification indices into 10. Then, close the Analysis
Properties window.

Step 16: Now, click on Analyze and choose Calculate Estimates to

analyze the model.

21

15/1/2015

Step 17: Once the analysis is complete, you see the output in the model
itself, by clicking View the output path diagram at the left panel.

Step 18: You can also look at the full result by clicking View
and choose Text Output.

22

15/1/2015

Step 19: The Amos Output window will come up and choose
the Estimates option on the left list.

Finally, You can see the result of the analysis from this window.

23

15/1/2015

Results
Regression Weights: (Group number 1 - Default model)
Estimate

S.E.

CO2 <---Household

1.083

.966

.262

CO2 <--- Electricity

.987

.018

***

CO2 <--- Transport

1.000

.001

***

Label

Transport - lowest S.E. value, 0.001 which means it has the strongest ability to predict the
household carbon emission;
number of household - highest value of S.E., 0.966 which has the weakest ability to predict the
carbon emission.
Electricity: C.R. = 55.385, p < 0.05; Transportation: C.R. = 690.840, p < 0.05
C.R. value is out of 1.96 range - significant variable to household carbon emission.

Standardized Regression Weights: (Group number 1 - Default model)

CO2 <--- Household

CO2 <--- Electricity
CO2 <--- Transport

Estimate
.002
.098
1.023

Positive correlations
Transportation has the highest correlation.
Most significant contributor.

24

15/1/2015

Covariances: (Group number 1 - Default model)

Estimate

S.E.

C.R.

Household <--> Electricity

83.137

31.341

Household <--> Transport

242.999

284.014

Label

.008
.856

.392

Electricity <--> Transport -23818.259 16259.741 -1.465 .143

Significant correlations between

these variables. These variables are
affecting each others

Correlations: (Group number 1 - Default model)

Estimate
Household <--> Electricity
.566
Household <--> Transport
.161
Electricity <--> Transport
-.283

Correlation between
Numbers of Household
and electricity
consumption is the
highest

25

15/1/2015

Variances: (Group number 1 - Default model)

Estimate
S.E.
Household
2.632
.691
Electricity 8195.731
2152.304
Transport 866266.473 227492.721
Residue
40.439
10.620

P
***
***
***
***

Label

Since C.R. Value is more than 1.96, so it

shows exogenous variables are significantly
able to forecast any changes in endogenous
variable (CO2 emission)

Squared Multiple Correlations: (Group number 1 - Default model)

Estimate
CO2

1.000

It shows 100% variance in CO2 emission can be

predicted by all the variables.

26

15/1/2015

Reporting the results.

a)

The result of SEM Analysis has shown that the regression model designed by the
researcher is suitable, as three of the variables which are numbers of household,
electricity consumption, and transportation fuels are significant predictor variables for
carbon emission variable (Household: = .002, C.R. = 1.121, p < 0.05. Electricity =
.098, C.R. = 55.385, p < 0.05, and Transportation: = 1.023, C.R. = 690.840, p < 0.05).

b)

Overall, SEM analysis result has shown that the variance value in endogenous variable
(CO2 emission) that been predicted by the three exogenous variables is 1.00. It shows
that 100.0% variance in CO2 emission is predicted by the numbers of household,
electricity consumption, and transportation fuel.

27