Factor Analysis Flashcards
What is a factor analysis?
the values of observed data are expressed as functions of a number of possible causes in order to find which are the most important
What is factor rotation?
a process by which we simplify the structure that we have
What do we need eigenvectors and eigenvalues for?
Allows us to make calls about the structure - a structural element that represents the factors that allow us to make judgements
What is factor loading?
Shows us how the individual elements relate to that structure and to those factors
What elements do we need for a factor analysis?
- Principle component analysis (including factor rotation)
- Eigenvector and values
- Factor loadings
What is a principle component analysis?
- How we are going to extract the data
- Takes a cloud of data points and finds the ‘principle axes’ of that cloud
- Lets you find the factors in factor analysis
What does a principle component analysis produce?
Eigenvectors and associated eigenvalues
What are eigenvectors?
Essentially our factors - mathematical foundation of the factors/components
How do we interpret eigenvectors?
- Each has an associated eigenvalue that tells you how important it is
- Factors with high eigenvalues are important
What is Kaiser’s Extraction?
- Used to tell which factors are important
- Retain factors with Eigenvalues > 1
How do we use scree plots to tell which factors are important?
Use the ‘point of inflexion’
- Factors above this point are important and we keep and the ones below are unimportant
What types of factor rotation are there?
- Orthogonal
- Oblique
What is orthogonal factor rotation?
The factors are uncorrelated
What is oblique factor rotation?
Factors inter-correlate
How do we interpret factor loadings?
- Between 0 - 1
- High loadings are above 0.5
- Above 0.3 is interesting
- Can be + or - depending on the nature of the relationship to the element
Can individual elements load on more than one factor?
Yes, but can be tricky to interpret
What assumptions need to be met for a factor analysis?
- Variables need to be correlated together (r>.3)
- But not too correlated (multicollinearity is an issue)
- Avoid singularity
- Sphericity
What is the use of Bartlett’s Test of Sphericity?
Testing to see if there are correlations between the variables
What are the sample size requirements for the data?
- 100 is low
- 300 is good
- 1000 is great
- Minimum 2 PPs per variables
What test can determine how good your sample size is?
Kaiser-Meyer-Olkin
- Should be above 0.5 to be adequate
What methods can we use to test reliability?
- Test-retest method
- Split-Half Method
- Cronbach’s Alpha
Explain the test-retest method
Test once and then test later after a certain time
What is the split-half method?
Splits the questionnaire into two random halves, calculates scores and correlates them
What is a disadvantage of the test-retest method?
Doesn’t always work as once the PP has been tested they may not be able to be tested again as they are not as naïve and will know the aim of the test (practice effect)
What is cronbach’s alpha?
Splits the questionnaire into all the possible halves, calculates the scores, correlates them and averages the correlation for all splits
- Ranges from 0 (no reliability) to 1 (complete reliability)
What are factor scores?
- You can get scores for each factor for each participant
- calculated by a combination of the factor loadings and the actual scores