Begin with simple RELIABILITY runs (a) on the entire set of 20 items and (b) on each of the three subscales that have been used in the literature (performance self-esteem, social self-esteem, and appearance self-esteem; see .prg file). For each run, show the output lines that contain the average inter-item correlation and alpha value and comment on the degree of reliability of the total scale and each subscale. (c) Is the whole scale unidimensional? Or is it three-dimensional? To aid you in your judgment, run a reliability analysis on the three subscales as items. That means that your three subscale scores now are items in a three-item scale that is the total. Comment on what you see.
Now run a PCA on the 20 items.
(a) Settle, after some
exploration and discussion, on an optimal number of factors. Show the
scree plot and the table of eigenvalues (and percent variance
explained) from your output.
(b) Apply a VARIMAX rotation. If
you want to, you can use the nice
SPSS command that blanks out loadings below a certain value (here, .30):
/format = sort blank (.30)How does the rotated solution differ from the unrotated one? How do you interpret the components of the rotated solution? (Add the loading matrix to your writeup.)
Now run a factor analysis, using principal axis factoring.
(a) Using scree plot and eigenvalue table from your output,
settle on an optimal number of factors.
(b) Note the difference
between the initial communalities and the final communalities.
Why is there a difference? (Provide the portion of the output that you
are referring to.)
(c) Say in a couple of sentences what the
computer does between producing the initial and the final
communalities.
(d) Apply a VARIMAX rotation and
comment on the meaning of the rotated factors. Show the loading
matrix, with the loadings ordered within each factor and
with very small loadings deleted.
(e) In a few sentences, compare the
results and interpretation of this FA with those of the PCA analysis
you conducted earlier.
Now apply an oblique rotation, using OBLIMIN.
(a) Do you
notice differences between loadings in the PATTERN matrix and the
STRUCTURE matrix? If so, explain those differences. (Include the
matrices you are referring to.)
(b) Are the results different
between the two rotations? To answer this question, compare the
FACTOR MATRIX (from the VARIMAX rotation) to the STRUCTURE MATRIX
(from the OBLIMIN rotation) and also examine the FACTOR
INTERCORRELATION MATRIX (which shows how strongly the oblique factors
are allowed to be related to each other). Include the matrices you are
discussing.
(a) Are there any items in the questionnaire you would exclude because they do consistently badly in your analyses (i.e., the reliability analysis, the PCA, and the FA)? (b) Finally, would you recommend using the total self-esteem score or the three subscales in any subsequent research?
Spend considerable time on your summary, which should contain decisions (and justifications) about (a) whether to trust the results from PCA or factor analysis, (b) about how many dimensions the scale contains, and (c) about whether these dimensions are independent (orthogonal) or should be treated as oblique.