Structural Equation Modelling

Question 1

Arrows

Answer 1

Curved - Covariance

Double headed arrow - covariance (as above)

Straight Line - Regression - directional pathway

Question 2

Bivariate means…

Answer 2

2 variable

Question 3

CMIN or CHi Squared

Answer 3

Looking for NON significance -

when it is a non significant result the model is a good fit

Question 4

Error

The small circle with the e1 in it

Answer 4

What is not explained by what is coming in from any of the independent variables

error = 1 - r²

Question 5

Errors

Answer 5

e.g. measuring happiness some variability due to happiness and some due to error, ie heat in room etc, question systematic error - the way i’s worded gives a higher score, random error

Question 6

‘Global goodness of fit’ (N Biased)

Steps to tell if the MODEL FITS

Answer 6

The goodness of fit tells us how well the structural coeffcients “match” the bivaiate correlations

1. Look at result or chi-squared is it significant? We are looking for a non-significant result for a goodness of fit

NULL hypothesis..they match or 0 difference which would mean a significant result for chi squared - the model is “significantly NOT” like the data

Question 7

Goodness of fit indicators - other

Answer 7

1.Global Goodness of Fit (N biased)

–Chi Squared (must be non significant)

2.Absolute Fit Indices

–AIC, CAIC, BIC, Hoelter’s N (CN), ECVI, BCC

3.Incremental Fit Indices (baseline comparison)

–CFI, NFI, IFI, RFI

•GFI (.95) >NFI (.80-.90)

4.Parsimony Adjusted Fit Indices (less complex)

–PRATIO, RMSEA (< .05 or .08; the 95% CI should be less than .10), P/G/N/CFI

Question 8

Indentification

and

Overidentification

Answer 8

Identification - is it a test?

If you want to interrelate 3 variables p=3 (2 IV’s & 1DV)

p(p+1) = amount of variance covariances which are the data

The total known amount of information will be 6 - so if I want to take arrows away that is fine (ie less) but if more it will get an erroyou ned more information than what you are asking for

Overidentification - is ok as you have asked for less that what is in the data

Underidentfication - is not ok where you ask fo more than ther data can offer

Question 9

Latent Variables

Answer 9

Concepts that are hypothetical - they are latent in people at some level, they cant be directly observed, like intelligence etc - we can measure using observable indicators (e.g. questionnaire)

Question 10

Mediation

Answer 10

Flow on effects - flowing through different towns etc

Question 11

Mediation

Answer 11

Flow on effect can e direct or indirect relationship

Question 12

Moderation

Answer 12

Interaction effects

Question 13

Number on the arrows

Answer 13

the correlation between factor and indicator

e.g. box (indicator) arrow (number .77) round shape (DV - factor Happiness) Correlation between job satisfaction and happiness is 77% so as job satisfaction goes us so does happiness

Question 14

r² - r squared

the number sitting on the top right had side of the box

Answer 14

is the amount of variance of the DV explained by the IV (IV’s)

e.g 77% of the variation of current salary is explained by beginning salary

.77 or r2 comes from the Standardised BETA coefficient e.g. .88x.88 =77.4 =77%

Question 15

SEM - mediation

Answer 15

interested in mediated effects - complex - or indirect effects - flow on through other variables

Question 16

SEM - techniques involved in .. also known as

Answer 16

covariance structure analysis, latent variable analysis, confirmatory actor analysis (CFA) LISERAL; path analysis

Question 17

SEM definition

Answer 17

SEM has 2 characteristics 1. Multiple IV’s and multiple DV’s 2. Represent factors (latent) as well as variables (manifest) (e.g. latent construct is “depression” underneath a whole lot of symptoms) 3. confirmatory focus on a theory you wish to test (path analysis using latent variables)

Question 18

SEM suited to

Answer 18

Useful for more complex concepts and constructs and systems of relationships - it is modelling a causal system

Question 19

SEM unpacks bivariate correlations into….

Answer 19

Direct effect (partital correlations) and ‘other influence pathways’ indirect effects

Question 20

SMC’s (r² ) - what is that?

Answer 20

that is the number to the top right hand side of the box whicch is the total amount of variance in the DV explained by the IV

Squared Multiple Correlations - like Pearson’s r

Question 21

Standardised Regression Weights P Value

Answer 21

Looks at Regression Weights: p values (see image)

If it is not significant then the arrow does not exist

Question 22

The Indirect Path Coefficient is?

The total correlation is?

Answer 22

The general rule is:

The product (multiply) of the indirect paths = the indirect pathway coeff between any two variables

The total correlation is then = Sum of Direct and Indirect coeffs presented in the particular path diagram.

(indirect (multiplied by direct_a) + (Direct_b)

e.g. .63 x .71 + .17= total correlation

.63 x .71 = indirect pathway coefficient

Question 23

What are the structural coefficients?

Answer 23

The coefficient numbers on the line between each box

Question 24

What are the total effects also called?

Answer 24

Bivariate Correlations

Question 25

What is over identified?

What is unsaturated?

What is under Idenetified?

Question 26

Wrights Rules for Pathways

Answer 26

No LOOPs, no forward then backward -no same variable twice

No going forward then backwad

1 curved error per route