Test 3 Flashcards
What does factor analysis do?
this analytic technique attempts to find groupings of items that constitute sub-factors within a single measure.
Be aware that these are groupings of ITEMS, not individual participants.
Deep dive: what is the diff between factor analysis and PCA anyway?
- Run factor analysis if you assume or wish to test a theoretical model of latent factors.
- Run principal component analysis If you want to simply reduce your correlated observed variables to a smaller set of important independent composite variables
What is an eigenvalue?
The eigenvalue is a mathematical representation of the degree of clustering of items created by the PCA program. The larger the eigenvalue, the better the clustering of items for that particular grouping. Some groupings will be very poor (eigenvalue near zero) but others will be larger.
Eigenvalue cut off
Eigenvalues greater than one is an arbitrary cut-off to try to catch the better clusterings of items. But it’s a crude and insensitive cut-off.
What is the deal with rotation anyway?
Orthogonal is the case where you force the factors to be maximally uncorrelated, whereas oblique allows more correlation among the factors. Varimax is orthogonal, and is often used. Oblimin is oblique.
Where is the point where the mountain ends and the loose gravel (scree) begins
One in from the point of inflection, i.e. one in from the elbow
Steps for deciding on number of subfactors
- Where is the kink or elbow in the scree plot?
- Do we have a relatively small number of subfactors with a reasonable number of items in each (e.g., 4 or more items)?
- Do these items yield an adequate Cronbach’s alpha (i.e., greater than .70) for the separate subfactors?
- Does a parallel analysis (PA) support the other converging evidence?
what’s a parallel analysis?
It is a Monte Carlo-generated set of eigenvalues that would occur by chance based on the number of items and participants that you have.
‘Monte Carlo’ refers to a data simulation that is run by the computer generating random values within a given range. i.e. under the line is under chance
Crossover on scree plot with parallel analysis
tells you how many sub factors we have, above the line or at cross over points
KMO and Bartlett’s Test
Kaiser-Meyer-Olkin Measure of Sampling Adequacy
Bartlett’s Test of Sphericity Approx. Chi-Square
Common errors with factor analysis
- Using the ‘eigenvalues greater than one’ rule
- Misinterpreting the kink in the scree plot. The proper number is usually the number to the left of the kink.
- Ignoring the parallel analysis approach.
- ‘Forcing’ a particular solution based on your theoretical views (or biases). Let the data tell you what is going on there.
What are the four criteria for determining factor structure of a PCA?
-Suppressed factor loading’s less than .3
Look to see if clusters make sense
KMO .6 minimum cutoff
Check Cronbach Alpha
What is confirmatory factor analysis
- It occurs AFTER someone has identified a factor structure in EFA
- It serves to determine whether the previously obtained factor structure is REPLICABLE in another sample
- It is NOT done in SPSS; instead it is done in structural equation modeling (SEM)
Comparing EFA to CFA
EFA: All items will load onto all factors. Does NOT generate model fit indices.
CFA: Not all items load onto all factors. It generates model fit indices.
Point: CFA tests how well the proposed model ‘fits the data’
model fit value for CFA
It needs to be lower than 7.0. If not it means that there is significant ‘misfit’ between the proposed model and the actual data
Classify by variable or case
Factor analysis classifies variables or items.
However, one can classify cases (individual subjects) within your dataset
What’s the advantage of this approach? It attempts to identify relatively homogeneous groupings of individuals who share one or several characteristics.
Then you can use those groupings to compare and contrast on other variables. CLASSIFICATION OF INDIVIDUALS BASED ON THEIR MOST LIKELY LATENT CLASS MEMBERSHIP