Professional Documents
Culture Documents
PSY417 Week10
PSY417 Week10
PSY417 Week10
0373
• What is SEM?
• Confirmatory Factor Analysis
• Path analysis
• SEM basics
• Running SEM
• Journal club
2
Readings
3
What is Structural Equation Modelling?
4
Structural Equation Modelling
by loading variables
(referred to as ‘indicators’)
onto factors without any a
priori knowledge
11
Source: towardsdatascience.com/confirmatory-factor-analysis-theory
Characteristics of CFA
• The number of factors and their observed
indicators (i.e. items or variables) must be
specified prior to analysis
13
Source: towardsdatascience.com/confirmatory-factor-analysis-theory
Structural Equation Modelling
• SEM is a collection of tools to analyse the possible
connections and their strengths
• Might reflect:
1. Imprecise measurement of the
endogenous variable, and/or
20
SEM terminology
Observed Observed
variable variable
Latent
variable
Direct effects
Covariances
21
Type of research questions best suited for
SEM
• When key constructs are complex and multi-faceted
1. Measurement model
2. Structural model
24
Measurement model
• Confirmatory factor analysis model
(𝑝 + 1)
𝑝= > 𝑝𝑎𝑟𝑎𝑚𝑒𝑡𝑒𝑟𝑠 𝑡𝑜 𝑏𝑒 𝑒𝑠𝑡𝑖𝑚𝑎𝑡𝑒𝑑
2 28
Source: Shanmugam & Marsh (2015)
Running SEM
29
1. Determine the model specification based
on literature
31
Source: Streiner (2005)
3. Determine model identification
32
4. Select measures, collect, prepare and
screen data
33
5. Estimate the model parameters
Run the analyses.
Model estimation determines how the tested model fits the data based on how the
data generated (“observed covariance matrix”) is equivalent to the model-implied
matrix (“hypothetical model”):
Σ = Σ(𝜃)
Think of the ‘hand-in-glove’ metaphor: the glove is the model, Σ 𝜃 , and the hand is
the data, Σ.
To find a good fit for the data, the lack of fit (too big or too small) is given by the
vector 𝜃. Source: Shanmugam & Marsh (2015)
34
6. Evaluate the model fit
This is done in SEM using the chi-square (𝜒 ! )statistic.
39
40
Introduction
• Social and familial support, and social group membership,
can positively impact resiliency and mental health.
• SCS and SDS are predictors of mental health but have only been studied in
isolation.
• Research aimed to test the causal relationships between these variables, focussing on the
prediction of symptoms of mental illness from SCS, SDH, and classic predictors (5 different
SEMs).
• Hypotheses
1. Significant prediction values and good model fit of SEMs containing SCS, SDH, and a
global SEM where SCS, SHD and classic predictors are integrated.
2. Most accurate prediction of symptoms of mental illness will be by combining SCS, SDH,
and classic predictors.
3. There will be reciprocal relationships between social and classic factors in predicting
symptoms of mental health. 44
Methods
Participants
• 18-59 years (mean age = 42.6 years, +/- 16.5 years)
• N = 2947
45
Methods
Instruments
• Symptoms of mental illness: SRQ-29 (outcome variable)
SCS
1. Emotion recognition task
2. Empathy for pain task
SHD
1. Social adverse and social protective factors
46
Methods
Instruments
Classic factors
1. Psychiatric antecedents
2. Physical-somatic factors (e.g. diabetes, hypertension)
3. Cognitive functioning
4. Assessment of covariables (sex and age)
47
1: The SCS-SEM
48
2: The SDH-SEM
49
3: The global social-SEM
50
4: The classical psycho-social-physical SEM
51
5: The global-integrated SEM
52
Results: Descriptive analysis of measures
53
Result: Parameters of goodness-of-fit (Fit
indices)
The global-integrated model showed the highest fit indices in comparison to the
other models.
54
Discussion
“In this randomized probabilistic design, we evaluated to what
extent a combined set of social factors (SCS and SDH) is able to
predict symptoms of mental illness.
SEMs using SCS and SDH reached high fit values and showed that
those factors are able to predict symptoms of mental illness.
Moreover, our results show that the integrated social model (SCS
and SDH) reached even higher prediction scores of symptoms of
mental illness compared with models using classical predictors.”
55
Limitations
• Self-reported measures of mental illness can
underestimate or overestimate predictors due to recall
bias.
56
Conclusion
“Our results highlight the importance of social factors as
robust predictors of symptoms of mental illness when they
are compared to classical factors.