6 Structural Equation Modelling Flashcards
What is the aim of path analysis?
To estimate the magnitude and significance of hypothesised causal connections between variables.
What is the path coefficient for a residual term? And why?
Always 1. It is assumed that error is not systematically associated with anything else.
In path analysis, what is disturbance?
Error/residual
What are endogenous variables? How do they work with arrows?
Presumed effects. Have arrows going into them.
What are exogenous variables? How do they work with arrows?
Presumed causes. Variables WITHOUT arrows going into them. Arrows go from them to other causes.
What is a path coefficient?
Value of the arrow/estimate of cause.
How many regression equations per endogenous variable do you need in a path model?
1
How do you calculate indirect effects?
By multiplying the path coefficients.
How do you calculate total effects?
Adding direct and indirect effects.
What’s the difference between recursive and non-recursive models?
Recursive – all paths go in one direction with no feedback loops. Can be solved with standard multiple regression equations.
Non-recursive –have reciprocal causality/feedback loops. Cannot be solved with regression equations.
What’s the difference between:
1) just-identified
2) under-identified
3) over-identified models?
1) Under-identified models have less than one possible solutions (i.e. they cannot be solved).
2) Just-identified – there is only one solution to what paths are.
3) Over-identified –have more than one solution. SEM programs determine which is most likely (e.g. CFA models which are rotated for best fit).
A just-identified (saturated) model has __ ___ elements in the covariance matrix as there are ________ __ __ _______.
A just-identified (saturated) model has as many elements in the covariance matrix as there are parameters to be estimated.
In an underidentified model there are more or less elements in the correlation matrix than there are parameters to be estimated?
Underidentified - less elements in the correlation matrix than there are parameters to be estimated.
How do you calculate the number of elements in the correlation matrix if the number of variables is greater than 3?
Use the formula for triangular numbers.
K= (N(N+1)) / 2
In an underestimated model in SEM, df is _______.
In an underestimated model in SEM, df is negative.
In an overidentified model, KNOWNS > UNKNOWNS. True or false?
True
What are regarded as the ‘knowns’ and ‘unknowns’ in model identification?
Knowns –elements in correlation matrix
Unknowns –parameters to be estimated (e.g. coefficients)
For a model to be identified it must have how many degrees of freedom?
0 or more
How do you calculate df in a model?
Number of parameters to be estimated
In an AMOS Regression Weight output, what is the Critical Ratio?
The estimate (unstandardized correlation coefficient) divided by standard error.
What are squared multiple correlations?
The same as R-squared. They represent the proportion of variance in endogenous variables that the model explains.
What is the logic of calculating model fit?
Estimate covariance matrix if model was 100% accurate in describing data. Compare this to actual covariance matrix. Calculate discrepancy with chi-square.
How can chi-square estimates be biased in SEM?
By having large sample –> significant chi-square.
What does a significant chi-square mean?
The model does NOT describe the data well. There is a large discrepancy between model and data.
What values would you want to find in the:
CFA (confirmatory fit index)
GFI (goodness of fit index)
RMSEA (root mean square error of approximation)
CFI and GFI –values over .90
RMSEA –values less than .6
If chi-square is less than df, what is this, what does it mean?
It’s called overfit, and it means that your model is too complex. Also results in maxing out GFI, NFI, CFI and RMSEA.