M22 - Structural Equation Modeling Flashcards
SEM in a nutshell
= tries to reproduce an observed …/… …. of measured variables as … as possible by …. … … (and intercepts) in a … set of (usually …) … … derived from theory.
A good model:
SEM in a nutshell
= combines path models and confirmatory fachor analysis
=tries to reproduce an observed VARIANCE/ COVARIANCE MATRIX of measured variables as ACCURATELY as possible by ESTIMATING REGRESSION WEIGHTS (and intercepts) in a LIMITED set of (usually LINEAR) REGRESSION EQUATIONS derived from theory.
A good model:
Parsimonious
Theoretically justifiable
Reproduces the underlying corr matrix based on the constraints imposed
What SEM can do
Its a combi of statistical data and qualitative causal assumptions
- graphical way to present an assumed theory
- estimates relship in the assumed theory
- estimates the causal effects of 1var on another (positive, negative, zero and signicabt)
- proves if the experimentally changed var actually causes an outcome
Ocean model
The big five
Openness, conscientiousness, extraversion, agreeableness, neuroticism
Model that characterizises personality into these dimensions
By microtargeting, means measuring people’s personality from their digital footprints based on ocean model
–> manipulate election
Importance in conjoint analyes
- the impirtance of an attribute is prop to the diff between. The part-worths of the best & the worst level of this attribute
- the impirtance values are normalized in such a way that their sum yields 100%
- the choice of levels of an attribute will in general affect the attribute’s importance
What does it mean:
The importance of an attribute is proportional to the difference between the part-worths of the best and the worst level of this attribute
If i have 3 attributes
Each of them has differnet deltas between best and worst part-worths
The one with the biggest delta, has the highest importance
Panel regression
If individual,
If individual, time-const effects are corr with the explanatory variable, then random effects regression is biased.
Fixed regression is still unbiased
What SEM cannot di?
- prove if 1v actually causes anpther var
- prove the direction of causal order betw variables
- distinguish between models that result in identical corr patterns
Regression model
Path model
Regression: several iv predict 1 dv
Path: several regr models are solved simultaneously
Exploratory factor analysis
Confirmatory factor analysis
Exploratory: when a set of items is correlated, the items can be combined to yield a score
Confirmatory: test whether a set of items define a construct
Regression models description
- only manifest var
- several iv predict 1 dv
- directed relationship –> factor loadings –> requires an error term
Path models description
- combine several regr models
- only manifest var
- directed relship –> factor loadings–> require an errotr term
- indirect effect for x1 on y2 is mediated through: x1 on y1 and y1 on y2
- total effect: the sum of direct effect x1 on y2 and indirect effect
Exploratory faczor analysis description
Factor loading represents whcih indicators correlate with the factor
No of factors such as to minimize factor loadings from indicators to one factor while minimizing cross-factor loadings
All factors may affect all items
Crinbachs aloha determines factor reliability
Confirmatory factor analysis description
–> tests assumed structure
- priori defined no of factors and existence of corr
- only possible for reflexive indicators
- checks for unidimensionality
- typically an indicator loads only on one factor
Advantage of path model plus confirmatory factor analysis?
Path: test relship between variables, but can only deal with observable variables
Cfa: tests cobstructs consisting of factors and. Indicators, and can use latent variables
Reflective and formative operationalization of constructs
- reflective: the indicators reflect the construct (drunkenness)
- formative: the indicators form the construct (drunkenness)
