Factor AnalysisAssumptionsSlide 3Slide 4Slide 5Slide 6Slide 7Slide 8Extraction MethodsSlide 10Slide 11Slide 12Slide 13Slide 14Slide 15Slide 16Slide 17Rotation MethodsGeometric RotationSlide 20Slide 21Slide 22Slide 23Slide 24Slide 25Slide 26Slide 27Factor AnalysisPsy 524AinsworthAssumptionsAssumes reliable correlationsHighly affected by missing data, outlying cases and truncated dataData screening methods (e.g. transformations, etc.) can greatly improve poor factor analytic resultsAssumptionsSample Size and Missing DataTrue missing data (MCAR) are handled in the usual ways (ch. 4) but regression methods may overfitFactor analysis needs large samples and it is one of the only draw backs•The more reliable the correlations are the smaller the number of subjects needed•Need enough subjects for stable estimatesAssumptionsComrey and Lee•50 very poor, 100 poor, 200 fair, 300 good, 500 very good and 1000+ excellent •Shoot for minimum of 300 usually•More highly correlated markers less subjectsAssumptionsNormalityUnivariate - normally distributed variables make the solution stronger but not necessaryMultivariate – is assumed when assessing number of factors; usually tested univariatelyAssumptionsNo outliers – obvious influence on correlations would bias resultsMulticollinearity/SingularityIn PCA it is not problem; no inversionsIn FA, if det(R) or any eigenvalue approaches 0 -> multicollinearity is likelyAlso investigate inter-item SMCs approaching 1AssumptionsFactorable R matrixNeed inter-item correlations > .30 or FA is unlikelyLarge inter-item correlations does not guarantee solution either•Duos•MultidimensionalityMatrix of partials adjusted for other variablesOther testsAssumptionsVariables as outliersSome variables don’t workExplain very little varianceRelates poorly with factorLow SMCs with other itemsLow loadingsExtraction MethodsThere are many (dozens at least)All extract orthoganal sets of factors (components) that reproduce the R matrixDifferent techniques – some maximize variance, others minimize the residual matrix (R – reproduced R)With large stable sample they all should be relatively the sameExtraction MethodsUsually un-interpretable without rotation (next)Differ in output depending on combinations ofExtraction methodCommunality estimatesNumber of factors extractedRotational MethodExtraction MethodsPCA vs. FA (family)PCA begins with 1s in the diagonal of the correlation matrix; all variance extracted; each variable giving equal weight; outputs inflated communality estimateFA begins with a communality estimates (e.g. SMC) in the diagonal; analyzes only common variance; outputs a more realistic communality estimateExtraction MethodsPCA analyzes varianceFA analyzes covariance (communality)PCA reproduces the R matrix (near) perfectlyFA is a close approximation to the R matrixExtraction MethodsPCA – the goal is to extract as much variance with the least amount of factorsFA – the goal is to explain as much of the correlations with a minimum number of factorsPCA gives a unique solutionFA can give multiple solutions depending on the method and the estimates of communalityExtraction MethodsPCAExtracts maximum variance with each componentFirst component is a linear combination of variables that maximizes component score variance for the casesThe second (etc.) extracts the max. variance from the residual matrix left over after extracting the first component (therefore orthogonal to the first)If all components retained, all variance explainedExtraction MethodsPrincipal (Axis) FactorsEstimates of communalities (SMC) are in the diagonal; used as starting values for the communality estimation (iterative)Removes unique and error varianceSolution depends on quality of the initial communality estimatesExtraction MethodsMaximum LikelihoodComputationally intensive method for estimating loadings that maximize the likelihood (probability) of the correlation matrix.Unweighted least squares – ignores diagonal and tries to minimize off diagonal residualsCommunalites are derived from the solutionOriginally called Minimum Residual method (Comrey)Extraction MethodsGeneralized (weighted) least squares Also minimizes the off diagonal residualsVariables with larger communalities are given more weight in the analysisMany Other methodsRotation MethodsAfter extraction (regardless of method) good luck interpreting resultRotation is used to improve interpretability and utilityA orthogonally rotated solution is mathematically equivalent to un-rotated and other orthogonal solutionsStable and large N -> same resultGeometricRotationGeometric RotationFactor extraction equivalent to coordinate planesFactors are the axesLength of the line from the origin to the variable coordinates is equal to the communality for that variableOrthogonal Factors are at right anglesGeometric RotationFactor loadings are found by dropping a line from the variable coordinates to the factor at a right angleRepositioning the axes changes the loadings on the factor but keeps the relative positioning of the points the sameRotation MethodsOrthogonal vs. ObliqueOrthogonal rotation keeps factors un-correlated while increasing the meaning of the factorsOblique rotation allows the factors to correlate leading to a conceptually clearer picture but a nightmare for explanationRotation MethodsOrthogonal Rotation MethodsVarimax – most popular•Simple structure by maximizing variance of loadings within factors across variables•Makes large loading larger and small loadings smaller•Spreads the variance from first (largest) factor to other smaller factorsRotation MethodsOrthogonal Rotation MethodsQuartimax•Opposite of Varimax•Simplifies variables by maximizing variance with variables across factors•Varimax works on the columns of the loading matrix; Quartimax works on the rows•Not used as often; simplifying variables is not usually a goalRotation MethodsOrthogonal Rotation MethodsEquamax is a hybrid of the earlier two that tries to simultaneously simplify factors and variables•Not that popular eitherRotation MethodsOblique Rotation TechniquesDirect Oblimin•Begins with an unrotated solution •Has a parameter (gamma in SPSS) that allows the user to define
View Full Document