Download presentation
Presentation is loading. Please wait.
1
Performance Tests in Guide Dogs
Factor Analysis Performance Tests in Guide Dogs R. Leotta, B. Voltini, M. Mele, M.C. Curadi, M. Orlandi, P. Secchiari (2006). ``Latent Vaiable Models on Performance Tests of Guide Dogs. 1. Factor Analysis,’’ Italian Journal of Animal Sciences, Vol. 5, pp
2
Data Description Data collected on n = 143 guide dogs on p = 11 variables: X1 ≡ Coming on Recall X2 ≡ Retrieving X3 ≡ Dominance X4 ≡ Following Aptitude X5 ≡ Curiosity to big windowed box in center of room X6 ≡ Reaction to Noise X7 ≡ Walking on anomalous surface as grilled ground X8 ≡ Passage on a weighing plank X9 ≡ Umbrella Test X10 ≡ Reaction to rolling Trolley X11 ≡ Reaction to a big puppet dog Variables scored on 1-5 scale, with 1 being unfavorable for working dog
3
Correlation Matrix, Means, Standard Deviations
Due to the scaling of the variables, very little difference between factor analyses of the covariance matrix (S) or Correlation matrix ® would occur. We will use R.
4
Orthogonal Factor Model - I
5
Orthogonal Factor Model - II
6
Estimation – Principal Factor Method - I
7
Estimation – Principal Factor Method - II
8
Eigenvalues and Eigenvectors of R
First two eigenvalues account for 43% of total variation, first four account for 62% To be able to test for adequacy of m factors, need 0.5((p-m)2 – (p+m)) = 0.5((11-m)2 – (11+m)) > 0 => m ≤ 6 Authors used m = 2 in their analysis
9
Principal Factor Estimates of L, Communalities and Y for m = 2, 4
The two columns of L2 are the unrotated factor loadings for the 2-factor model. The first may interpreted as an overall sum of the scores for the dog (similar positive coefficients for all 11 variables) The second may be interpreted as a contrast between variables X1:X5 and X6:X10 The 4 factor model has two additional sets of factor loadings.
10
Maximum Likelihood Estimates of L, Communalities and Y for m = 2, 4
The two columns of L2 are the unrotated factor loadings for the 2-factor model. The first may interpreted as an overall sum of the scores for the dog (similar positive coefficients for all 11 variables) The second may be interpreted as a contrast between variables X1:X5 and X6:X10 The 4 factor model has two additional sets of factor loadings. Note that for the ML estimates, the first 2 sets of factor loadings differ for the 2- and 4-factor models
11
Large-Sample Test for # of Common Factors (m)
Test Stat DF X2(.05) P-value m= m=
12
Plot of Factor Loadings for 2 Factor Model
Some variables have high loadings on both factors. Factor 2 is “bi-polar” with positive and negative loadings Will rotate axes so that there are possibly clearer interpretations for the 2 factors. Varimax rotation
13
Rotated Factor Loadings
Variables 5:10 Load High (> .35) on Factor 1, Variables 1:5 Load High on Factor 2 Less overlap and bi-polarity removed
14
Estimating Factor Scores – Weighted Least Squares
15
Guide Dog Example
Similar presentations
© 2025 SlidePlayer.com. Inc.
All rights reserved.