lecture 6 Flashcards
PCA
principle component analysis
- when we try to measure a single psychological phenomena - many things said to measure a particular process
- all questions asked have to tap into same biological construct
- can be the same for behaviour
what tests measures?
- self-report impulsivity
- go/no-go task
- time estimation
- stop signal task performance
PCA allows us..
to take many items and reduce dimensionality of a construct
Allison et al., 2014 - scripted responses, taking about unrelated topics - can be combined into single measure
- reduces likelihood of false positives because we test one construct now 3 separate measures
- any PCA will find very large number of components + uses eigenvalue to just whether factors are worth keeping
R-matrix
- looks at which items correlate with each other + cause the component factor
Eigenvalue
- tells use how important each measure is
Kaiser rule
Eigenvalues larger than 1 mean component is valid
Joliffe (1972,1986)
Eigenvalues
factor loadings
PCA will give us number of components - doesn’t tell us which of our individual measure make up each component
- factor loading tells us this - association between factor and measure
pearsons correlation between item and factor
component matrix
factor loadings can be negative
factor can load into multiple
Rotation
- do not interpret component matrix - apply rotation first
- Yaremenko et al., 1986 - shifts the factors in a 2 dimension space to make loadings clearer
forms of loading
Orthogonal rotation methods assume factors in analysis are uncorrelated
more commonly used is Varimax rotation
- choose rotation according to whether evidence suggests your components/factors will be correlated or not
assumptions
- variables should not be nominal
- sampling adequacy - Kaiser-Meyer Oskin measure of sampling adequacy should be above .5 to be acceptable
- get of sphericity
- tests the null hypothesis that the correlations represent an identity matrix - want to be significant
