Factor analysis Flashcards
EFA steps
- Correlation matrix
- Plot the Eigenvalues and make Scree plot
- Select a number of factors
(Extract the factors and rotate to find meaning)
- Interpret
Factor rotation for uncorrelated factors
Orthogonal rotation.
Most common
Varimax rotation
Factor rotation for correlated or uncorrelated
Oblique rotation
It allows factors to show if they ‘want to be’ correlated or uncorrelated.
Rotation is done to make the factors look more like a simple structure
Factor loadings
With use of rotation, you can see loading on a factor depending what rotation you choose;
Oblique:
- Pattern coeffictients; reflects unique association, the degree to which an item is associated with a factor, controlling for the correlation between the factors.
- Structure coeff; Correlations between item response and level of underlying factors. Not controlling for correlations
Simple structure
Occurs when item is strongly linked to only one factor. The closer to 0 on one of the factors, the simpler the structure. It reveals clearly which items should be scored together (those that load on the same factor)
Factor correlation matrix
Shows correlation between the two extracted factors.
Confirmatory factor analysis (CFA)
When you know the dimensions and on which factors items are supposed to load. used to confirm/disconfirm hypothesis about test’s dimensions
Eigenvalues
Tell you something about the amount of variability in the data that can be explained by that item
- First is always the biggest, accumulates till 100%
- Skip Kaizer (above 1)
- Make Scree plot to asses inflection point (Factors is nr before the point)
Factor scores
After rotating, it’s an option to use the factor scores of the rotation instead of the sum score. (dimensions)
- It reduces the data (eg only 3 variables io 20)
- Has advantage over composite score (more natural, takes factor loadings into account)
- Avoid multicollinearity; measures that measrue the same ends up in high correlations. Due to less veriables, less chance
