principal component and exploratory factor analysisprincipal component and exploratory factor...

24
Principal component and exploratory factor analysis Hans Baumgartner Penn State University

Upload: others

Post on 30-Jan-2021

7 views

Category:

Documents


0 download

TRANSCRIPT

  • Principal component and exploratory factor analysis

    Hans Baumgartner

    Penn State University

  • Principal component and exploratory factor analysis

    x1 x2 x3 x4 x5 x6 x7 x8

    What’s the structure underlying

    28 distinct covariances between

    8 observed variables?

  • Principal component and exploratory factor analysis

    The principal component model

    x1 x2 x3 x4 x5 x6 x7 x8

    x1 xn. . . . . . . . . . . . . . .

  • Principal component and exploratory factor analysis

    How many components to retain?

    ▪ Eigenvalue greater than one rule

    (Kaiser’s rule)

    ▪ Scree test (Cattell)

    ▪ Parallel analysis (Horn)

    ▪ Minimum Average Partial (MAP) test

    (Velicer)

    ▪ Theoretical considerations

  • Principal component and exploratory factor analysis

    Factor rotation

    ▪ Orthogonal rotation

    e.g., Varimax

    ▪ Oblique rotation

    e.g., Promax

  • Principal component and exploratory factor analysis

    SAS specification

    DATA coupon;

    INFILE 'd:\ipss\factor.dat';

    INPUT id aa1t1 aa2t1 aa3t1 aa4t1 aa1t2 aa2t2 aa3t2 aa4t2;

    ods graphics on;

    TITLE 'PRINCIPAL COMPONENT ANALYSIS, TWO COMPONENTS'; run;

    PROC FACTOR NFACTORS=2 ROTATE=PROMAX PLOTS=(scree preloadings loadings);

    VAR aa1t1 aa2t1 aa3t1 aa4t1 aa1t2 aa2t2 aa3t2 aa4t2;

    run;

    ods graphics off;

  • Principal component and exploratory factor analysis

    The FACTOR Procedure

    Initial Factor Method: Principal Components

    Prior Communality Estimates: ONE

    Eigenvalues of the Correlation Matrix: Total = 8 Average = 1

    Eigenvalue Difference Proportion Cumulative

    1 5.56628614 4.99193128 0.6958 0.6958

    2 0.57435486 0.05222193 0.0718 0.7676

    3 0.52213293 0.14851325 0.0653 0.8328

    4 0.37361969 0.06185008 0.0467 0.8795

    5 0.31176961 0.05897629 0.0390 0.9185

    6 0.25279331 0.03126256 0.0316 0.9501

    7 0.22153076 0.04401806 0.0277 0.9778

    8 0.17751270 0.0222 1.0000

    2 factors will be retained by the NFACTOR criterion.

  • Principal component and exploratory factor analysis

  • Principal component and exploratory factor analysis

    Parallel Analysis:

    Principal Components

    Specifications for this Run:Ncases 250Nvars 8Ndatsets 100Percent 95

    Random Data Eigenvalues

    Root Means Prcntyle

    1.000000 1.263893 1.3444632.000000 1.163970 1.2300473.000000 1.088389 1.1332354.000000 1.026321 1.0601315.000000 0.967248 1.0070766.000000 0.904000 0.9466257.000000 0.834617 0.8784278.000000 0.751562 0.807438

  • Principal component and exploratory factor analysis

    Velicer's Minimum Average Partial (MAP) Test:

    Eigenvalues

    5.56600.57820.52100.37270.30760.25700.22050.1772

    Average Partial Correlations

    squared power4

    0.0000 0.4280 0.18921.0000 0.0440 0.00412.0000 0.0662 0.01503.0000 0.1058 0.03974.0000 0.1692 0.09125.0000 0.2582 0.12246.0000 0.4869 0.36537.0000 1.0000 1.0000

    The smallest average squared partial correlation is0.0440

    The smallest average 4rth power partial correlation is0.0041

    The Number of Components According to the Original (1976) MAP Test is1.0000

  • Principal component and exploratory factor analysis

    Factor PatternFactor1 Factor2

    aa1t1 0.81547 0.35121aa2t1 0.83341 0.34182aa3t1 0.77536 0.01958aa4t1 0.84339 0.23736aa1t2 0.86051 -0.03867aa2t2 0.86274 -0.23905aa3t2 0.82149 -0.37671aa4t2 0.85709 -0.27731

    Variance Explained by Each Factor

    Factor1 Factor2

    5.5662861 0.5743549

    Final Communality Estimates: Total = 6.140641

    aa1t1 aa2t1 aa3t1 aa4t10.78833603 0.81141733 0.60156808 0.76764683

    aa1t2 aa2t2 aa3t2 aa4t20.74197084 0.80145580 0.81674966 0.81149643

  • Principal component and exploratory factor analysis

    The FACTOR Procedure

    Prerotation Method: Varimax

    Rotated Factor Pattern

    Factor1 Factor2

    aa1t1 0.33675 0.82154

    aa2t1 0.35614 0.82739

    aa3t1 0.54017 0.55658

    aa4t1 0.43637 0.75976

    aa1t2 0.64176 0.57455

    aa2t2 0.78357 0.43298

    aa3t2 0.85044 0.30577

    aa4t2 0.80631 0.40169

  • Principal component and exploratory factor analysis

    The FACTOR ProcedureRotation Method: Promax (power = 3)

    Inter-Factor Correlations

    Factor1 Factor2

    Factor1 1.00000 0.67589Factor2 0.67589 1.00000

    Rotated Factor Pattern (Standardized Regression Coefficients)

    Factor1 Factor2

    aa1t1 0.02338 0.87192aa2t1 0.04504 0.86973aa3t1 0.41062 0.43660aa4t1 0.17970 0.74463aa1t2 0.53035 0.40913aa2t2 0.77916 0.16017aa3t2 0.92612 -0.03361aa4t2 0.82327 0.10941

  • Principal component and exploratory factor analysis

    The exploratory factor model

    d1 d2 d3 d4 d5 d6 d7 d8

    x1 x2 x3 x4 x5 x6 x7 x8

    x1 xn

    q11d q22d q33d q44d q55d q66d q77d q88d

    11

    . . . . . . . . . . . . . . .

  • Principal component and exploratory factor analysis

    +

    =

    8

    7

    6

    5

    4

    3

    2

    1

    2

    1

    8281

    7271

    6261

    5251

    4241

    3231

    2221

    1211

    8

    7

    6

    5

    4

    3

    2

    1

    δ

    δ

    δ

    δ

    δ

    δ

    δ

    δ

    ξ

    ξ

    λλ

    λλ

    λλ

    λλ

    λλ

    λλ

    λλ

    λλ

    x

    x

    x

    x

    x

    x

    x

    x

    =

    1

    1

    21 8811 .... qq

    d Diag=

    Two-factor model

  • Principal component and exploratory factor analysis

    SAS specification

    TITLE 'PRINCIPAL FACTOR ANALYSIS, TWO FACTORS'; run;PROC FACTOR NFACTORS=2 ROTATE=PROMAX PRIORS=SMC;VAR aa1t1 aa2t1 aa3t1 aa4t1 aa1t2 aa2t2 aa3t2 aa4t2;

    run;

    TITLE 'EXPLORATORY MAXIMUM LIKELIHOOD FACTOR ANALYSIS, TWO FACTORS'; run;PROC FACTOR METHOD=ML NFACTORS=2 ROTATE=PROMAX SE COVER=.40 ALPHA=.1;VAR aa1t1 aa2t1 aa3t1 aa4t1 aa1t2 aa2t2 aa3t2 aa4t2;

    run;

  • Principal component and exploratory factor analysis

    The FACTOR ProcedureInitial Factor Method: Principal Factors

    Prior Communality Estimates: SMC

    aa1t1 aa2t1 aa3t1 aa4t1

    0.61896228 0.66565385 0.55622099 0.66929517

    aa1t2 aa2t2 aa3t2 aa4t2

    0.69849794 0.70624991 0.64787954 0.70324282

    Eigenvalues of the Reduced Correlation Matrix:Total = 5.26600251 Average = 0.65825031

    Eigenvalue Difference Proportion Cumulative

    1 5.22802782 4.98990454 0.9928 0.99282 0.23812328 0.10148438 0.0452 1.03803 0.13663890 0.10807443 0.0259 1.06404 0.02856447 0.05139663 0.0054 1.06945 -.02283216 0.05684943 -0.0043 1.06506 -.07968159 0.02020619 -0.0151 1.04997 -.09988779 0.06306265 -0.0190 1.03098 -.16295043 -0.0309 1.0000

  • Principal component and exploratory factor analysis

    Parallel Analysis:

    Principal Axis / Common Factor Analysis

    Specifications for this Run:Ncases 250Nvars 8Ndatsets 100Percent 95

    Random Data Eigenvalues

    Root Means Prcntyle

    1.000000 0.297504 0.3896522.000000 0.193068 0.2688263.000000 0.113040 0.1589434.000000 0.048229 0.0872085.000000 -0.011975 0.0373076.000000 -0.073090 -0.0345227.000000 -0.137988 -0.1000638.000000 -0.213994 -0.163157

  • Principal component and exploratory factor analysis

  • Principal component and exploratory factor analysis

    The FACTOR Procedure

    Rotation Method: Promax (power = 3)

    Inter-Factor Correlations

    Factor1 Factor2

    Factor1 1.00000 0.74679Factor2 0.74679 1.00000

    Rotated Factor Pattern (Standardized Regression Coefficients)

    Factor1 Factor2

    aa1t1 0.15925 0.68215aa2t1 0.12325 0.74482aa3t1 0.35117 0.43689aa4t1 0.21434 0.66378aa1t2 0.51948 0.37873aa2t2 0.71140 0.18860aa3t2 0.72635 0.12017aa4t2 0.74590 0.14729

  • Principal component and exploratory factor analysis

  • Principal component and exploratory factor analysis

    EXPLORATORY MAXIMUM LIKELIHOOD FACTOR ANALYSIS, TWO FACTORS

    Significance Tests Based on 250 Observations

    Pr >Test DF Chi-Square ChiSq

    H0: No common factors 28 1534.0584

  • Principal component and exploratory factor analysis

    The FACTOR Procedure

    Rotation Method: Promax (power = 3)

    Inter-Factor CorrelationsWith 90% confidence limits

    Estimate/StdErr/LowerCL/UpperCL

    Factor1 Factor2

    Factor1 1.00000 0.728480.00000 0.01570. 0.70161. 0.75328

    Factor2 0.72848 1.000000.01570 0.000000.70161 .0.75328 .

  • Principal component and exploratory factor analysis

    Rotated Factor Pattern (Standardized Regression Coefficients)With 90% confidence limits; Cover |*| = 0.4?

    Estimate/StdErr/LowerCL/UpperCL/Coverage Display

    Factor1 Factor2

    aa1t1 0.19857 0.646280.08017 0.077720.06389 0.500040.32616 0.75668

    0[]* 0*[]

    aa2t1 0.06195 0.827830.05207 0.05846-0.02394 0.704230.14693 0.90272

    [0]* 0*[]