Structural Equation Modelling
(SEM)
An Introduction (Part 1)
What is Structural Equation Modelling?
• SEM is a general statistical modelling technique used to establish relationship among
variables.
• SEM is a confirmatory data analysis technique, i.e.

 it tests models that are conceptually derived, beforehand
 it tests if the theory fits the data

• SEM can be thought of as a combination of factor analysis and multiple regression
 it can simultaneously test measurement and structural relationships

• SEM is a family of related procedures. It is alternately defined by the following terms

 Path Analysis, Path Modelling, Causal Modelling, Analysis of Covariance Structures, Latent
Variable Analysis, Linear Structural Relations
Covariance: At the Heart of SEM
• Covariance is a measure of how much two random variables change
together. Alternately, it can be defined as the strength of association
between the two variables and their variabilities.
𝑛
𝑖=1(𝑋 𝑖 −

𝑋)(𝑌 𝑖 − 𝑌)

𝑁−1

OR

𝑐𝑜𝑣 𝑥𝑦 = 𝑟 𝑥𝑦 𝑆𝐷 𝑥 𝑆𝐷 𝑦

• The basic statistic of SEM
 Understanding patterns of correlations among a set of variables
 Explain as much of their variance as possible with the model specified
Logic of SEM
• Every theory (model) implies a set of correlations
 And why variables are correlated
• Necessary (but insufficient) condition for the validity of the theory is
that it should be able to reproduce the correlations that are actually
observed
 i.e., the implied covariance matrix should = the actual covariance
matrix
Why SEM over Regression?
• Regression allows for only a single dependent variable,
whereas SEM allows for multiple dependent variables.
• SEM allows for variables to correlate, whereas regression
adjusts for other variables in the model.
• Regression assumes perfect measurement, whereas SEM
accounts for measurement error.
USES OF SEM
• Theory testing
 Strength of prediction/association in models with multiple DVs
 Model fit
• Mediation/tests of indirect effects
• Group differences
 Multiple-sample analysis
• Longitudinal models
• Multilevel nested models
Looking for Online SEM
Training?
Contact us: info@costarch.com

Visit: http://tinyurl.com/costarch-sem
www.costarch.com

Structural Equation Modelling (SEM) Part 1

  • 1.
  • 2.
    What is StructuralEquation Modelling? • SEM is a general statistical modelling technique used to establish relationship among variables. • SEM is a confirmatory data analysis technique, i.e.  it tests models that are conceptually derived, beforehand  it tests if the theory fits the data • SEM can be thought of as a combination of factor analysis and multiple regression  it can simultaneously test measurement and structural relationships • SEM is a family of related procedures. It is alternately defined by the following terms  Path Analysis, Path Modelling, Causal Modelling, Analysis of Covariance Structures, Latent Variable Analysis, Linear Structural Relations
  • 3.
    Covariance: At theHeart of SEM • Covariance is a measure of how much two random variables change together. Alternately, it can be defined as the strength of association between the two variables and their variabilities. 𝑛 𝑖=1(𝑋 𝑖 − 𝑋)(𝑌 𝑖 − 𝑌) 𝑁−1 OR 𝑐𝑜𝑣 𝑥𝑦 = 𝑟 𝑥𝑦 𝑆𝐷 𝑥 𝑆𝐷 𝑦 • The basic statistic of SEM  Understanding patterns of correlations among a set of variables  Explain as much of their variance as possible with the model specified
  • 4.
    Logic of SEM •Every theory (model) implies a set of correlations  And why variables are correlated • Necessary (but insufficient) condition for the validity of the theory is that it should be able to reproduce the correlations that are actually observed  i.e., the implied covariance matrix should = the actual covariance matrix
  • 5.
    Why SEM overRegression? • Regression allows for only a single dependent variable, whereas SEM allows for multiple dependent variables. • SEM allows for variables to correlate, whereas regression adjusts for other variables in the model. • Regression assumes perfect measurement, whereas SEM accounts for measurement error.
  • 6.
    USES OF SEM •Theory testing  Strength of prediction/association in models with multiple DVs  Model fit • Mediation/tests of indirect effects • Group differences  Multiple-sample analysis • Longitudinal models • Multilevel nested models
  • 7.
    Looking for OnlineSEM Training? Contact us: info@costarch.com Visit: http://tinyurl.com/costarch-sem www.costarch.com