Wk 5 - CFA Flashcards
What are the 3 main purposes of CFA?
ID latent psych constructs that account for correlations among sub-sets variables
Assess how strongly each variable is associated with factors
Test hypotheses about factor structure underlying set of variables
What are the 2 main diffs between EFA and CFA? (x2, x2)
EFA looks for structure of data set
Is data-driven
CFA tests specific hypotheses about structure
Is theory driven
Explain how CFA works (x4)
Based on theory, predict which co-variances among vars should be low
Constrain them to zero in analysis
Predict zero correlation between variables and OTHER factor/s
Test how accurately model fits the data
What are model parameters in CFA? (x1)
Factor loadings
How does CFA use model parameters? (x2)
Pattern chosen predicts parameters
Iterative process adjusts free parameter loadings to max similarity between predictions and data
How are model predictions evaluated in CFA? (x2)
Chi-square test used to test diff between observed and predicted data
Significance = (unwanted) diff between model and data
What 2 important things do CFA results reveal? (x4, x3)
‘Quality’ of theory:
o Hypothesized structure ignore anything important (poor fit)?
o Closely capture data (good fit)?
o Parsimony? (minimum psych constructs, not just fit)
Separates in/essential:
o What factors/loadings are needed to describe data structure?
o Which aspects of data can we ignore?
When should we choose EFA? (x3)
To explore a set of variables
o Newly developed variables/measures
o No clear ideas about constructs underlying the variables
When should we choose CFA? (x3)
To test a priori hypotheses derived from:
o Existing theory
o Previous research using the same variables
What are the 4 steps in conducting a CFA?
Preliminaries
Evaluate model fit
Evaluate parameter estimates (pattern of factor loadings and correlations)
Evaluate alternative models
What are 4 preliminary considerations (step 1) for running a CFA?
Check assumptions
Specify the model
Model parameters
Sample size?
What assumptions should be checked before running a CFA?
Interval scales
Good variance in scores
LInear correlations between variables
Generally normally distributed
What are the considerations for specifying the model in CFA? (x1, x3)
Theory driven assumptions of number/pattern of factors
Capturing most important aspects of data:
What is essential for explaining the data?
What can be ignored?
What are the considerations when deciding on model parameters in CFA? (x2)
Which should be fixed (at correlations of zero)?
Which should be free (estimated from the data)?
What do free parameters do for CFA? (x2)
Add greater flexibility:
Model can account for more patterns in data
What is the benefit of the added flexibility brought by free parameters in CFA? (x2)
Better account of wider range of data
*Explain specific features in data, that aren’t zero correlations
What is the downside of the added flexibility brought by free parameters in CFA? (x2)
Model that accounts for everything doesn’t give insight:
* More freedom = more noise explained * Model predicts 'anything could happen'
Why is it important to consider sample size for CFA? (x2)
And we should aim for? (x1)
Stability of model: small samples = unreliable estimates
Statistical power - detect smaller effects
5 - 10 per parameter
In what ways is model fit evaluated (step 2) in CFA? (x3)
Against a null hypothesis
In terms of absolute fit
And relative fit
What test is used to assess similarity of estimated and observed covariance matrices in CFA? (x1, explain x3)
Chi-square test:
Null hypothesis of no diff between model and data
Significance means poor fit
ns means acceptable fit
What are the limitations of chi-square test? (x2)
Very sensitive to large samples (as used in CFA)
Assumptions of multivariate normality often not met
What is the accepted convention for dealing with significant chi-square results in CFA? (x2)
Evaluate the importance:
*If chi-square over df > 2, misfit can be downplayed/dismissed
What are alternative indices (to chi-square) used for in CFA? (x4)
Assess how much observed co/variance matrix accounted for by model
Absolute fit, and relative, against baseline model
Range 0 - 1 (no NHST)
*higher is better (goodness of fit)
What 6 alternative indices are used in CFA?
And you should… (x1)
Normed fit index (NFI) Non-normed fit index (NNFI) Incremental fit index (IFI) Comparative fit index (CFI) Goodness-of-fit index (GFI) Adjusted goodness-of-fit index (AGFI)
Report 2 chosen a priori
