Karteikarten
RM - Lecture 4 - Factor Analysis
Karten in diesem Stapel (36)
- 1What did Charles Spearman do?He noticed students test scores were correlated, and thought this could be due to an underlying factor.
He called this g and found that subjects could predict g - 2Why do we use questionnaires?- We can ask lots of questions
- Cheap and quick
- We think the answers can often be explained by a small number of factors - 3What can't the computer work out when looking at underlying factors?They can't work out what the questions have in common eg. love of chocolate
- 4What is factor analysis good for?- Testing for clusters of variables
- Whether different measures have common aspects (underlying factors)
- See what the factors might be
- See how many factors there are - 5What is factor analysis not?- A direct way of modelling data (regression)
- Cant tell what the factors are - 6Look at personality traits example.
- 7What does SPSS give us when analysing correlation?An R-Matrix (correlation matrix)
- 8What does factor analysis try to do to the R-matrixReduce it to a small set of uncorrelated dimensions
- 9What does it mean if a factor is orthogonal?You can vary along it independently (eg. of another factor)
- 10What are factor loadings and where are they stored?- How much each question contribute to each of the factors.
- The weight of a variable on a factor
- Stored in the factor matrix - 11How would you get the total score from the factor matrix?Times factor 1 and 2's loading on question one together, then question 2 etc.
- 12What is the equation for gFactor loading * Test score
- 13Why do we often use a 5 point scale?People can usually only discriminate between 5 points
- 14Look at statistics example..
- 15What are the initial considerations before making a questionnaire for factor analysis?- Test variables should correlate (r > .30)
- No multicollinearity
- No singularity
- Examine correlation matrix to eliminate concern variables
- Is there enough measurements?
- Are there data clusters to find in the first place? - 16Multicollinearity- Several variable highly correlated (r > .80 is bad)
- Can happen if you enter the same question twice - 17Singularity- Some variable correlate perfectly (r = 1)
- You should drop one as they tell you the same thing - 18Determinant- Indicates multicollinearity
- Should be greater than 0.00001 - 19Keiser-Meyer-Olkin (KMO)- Measures sampling adequacy
- Should be greater than 0.5 - 20Bartlett's test of sphericity (cloud of data)- Checks for low correlation across all predictors
- Should be significant (p=0.5) - 21What is Principle Component Analysis (PCA) and example of?Ways of finding factors (factor analysis)
- 22What does PCA do?- FInds the principle axis in a cloud of data
- The axes are at right angles to each other - 23What are eigenvectors?- The axis that PCA finds
- A way of reducing complex, multidimensional datasets to a manageable set of components - 24What are eigenvalues?- Each eigenvector has an eigenvalue which SPSS lists
- Factors with small eigenvectors are useless - 25What is a scree plot?Graph of number of questions/components agains eigenvalues.
- 26How do Cattell and Kaiser say to decide which factors to include?Cattell - Everything above the point of inflection
Kaiser - Use factors with an eigenvalue over 1 - 27Why do we do factor rotation?To maximise the loading of a variable on to one factor whilst minimising its loading on other factors.
- Makes it interpretable
- More isolation of variables - 28What are the two types of rotation?Orthogonal
- Factors uncorrelated
- Rotated
Oblique
- Factors correlated
- Squashed
We only use orthogonal/varimax - 29What does SPSS give each factor?A % of the variance that it explains
- 30What do we assume about these algebraic factors?They represent psychological constructs
- 31What ways are there to test reliability of PCA?- Test-retest method
- Split-half method
- Chronbach's alpha - 32Test-retest methodRetest it to see if theres any practice, mood etc effects
- 33Split-half method- Split the questionnaire into two random groups.
- Calculate scores and correlate them - 34What does Chronbach's alpha do?Splits questionnaire into all possible halves then calculates scores and correlates them. Then it averages the correlation for all splits.
- Ranges from 0-1 with 1 being completely reliable
- Measures what happens to the reliability if we remove one question - 35How do we interpret Chronbach's alpha?- Kline said it's reliable if it's over .70
- More questions = bigger alpha
- Treat sub scales sperately
eg. if CA is higher than groups CA if item were deleted then it should be deleted. - 36What is flagged up in CA if we have forgotten to do it?Reverse scoring
Informationen
- Beschreibung
- Karteikarten am RM - Lecture 4 - Factor Analysis, erstellt von hannahbevans am 02/01/2015.
- Erstellt von
- hannahbevans · Jan 2015
- Aktualisiert
- 19 Feb 2016
- Aufrufe
- 2
- Bewertung