Perceived Competence: A Common Core for Self-Efficacy and Self-Concept?

Abstract

This study uses Bandura's Multidimensional Scales of Perceived Self-Efficacy (MSPSE; Bandura, 1990) and Harter's Self-Perception Profile for Adolescents (SPPA; Harter, 1988) to examine the extent to which self-efficacy and competency-related elements of the self-concept are independent constructs. Factor analysis of data provided by 778 high school students revealed that when measured using domain-general measures such as the MSPSE and SPPA, self-efficacy and competency self-concept do not represent totally separate, distinct constructs. Overlap of dimensions occurs at both the first- and second-order levels of analysis. The practical and theoretical implications of these findings are discussed.

Notes

Of the two opposing statements the most positive statement combined with an “always like you” response = 4; most positive statement/sometimes like you = 3; least positive statement/sometimes like you = 2; least positive statement/always like you = 1.

Principal factors analysis (PFA; often called common factors analysis or principal axis factoring) does not explain all the variance in a matrix and factors are less likely to be contaminated by error. Principal components analysis (PCA) explains all the variance in the correlation matrix to which it is applied and because all correlations contain error, components must therefore be contaminated by error. Theoretically, this is a disadvantage because it is unlikely that factors could explain all the variance in a given matrix (Kline, 1994, 2000; Tabachnick & Fidell, 2001). Therefore, to achieve a solution uncontaminated by unique and error variability, PFA was conducted here. Note that PCA was also conducted, with similar findings. This is consistent with assertions that when the number of variables and the sample size are large, as they are here, the differences between PFA and PCA extractions are negligible (Harman, 1976; Nunnally & Bernstein, 1994; Thompson, 2004).

Delta = 0 is the default value in SPSS and was used here to ensure that very high and very low correlations between factors were not allowed (see Field, 2009). The default was also used as it has been argued that manipulating Delta from the default value introduces unnecessary complexity for interpretation of results (Costello & Osborne, 2005).

Parallel analysis compares prerotational eigenvalues from the real data with corresponding eigenvalues from a factor analysis of a similarly proportioned set of randomly generated data. Where a real data eigenvalue exceeds the associated eigenvalue for the random data then the associated factor is not expected to have occurred by chance and can be extracted. Parallel analysis with a 95% criterion allows us to determine whether there is a less than 5% chance that the eigenvalues could have occurred if there are in reality no factors in the data set.

The proportion of variance assigned to individual factors is to some extent ambiguous. This is because in oblique rotations factors are correlated and share overlapping variability (Tabachnick & Fidell, 2001).

MSPSE items excluded = 1, 2, 4, 10, 18, 24, 29, 33, 34, 41, 46, 47. SPPA items excluded = 1, 10, 32. All item numbers are consistent with those given by the original authors.

Parallel analysis using O’Connor's (2000) PFA syntax indicated 21 factors with real eigenvalues greater than the random number equivalents, which seemed excessive. However, parallel analyses of adjusted correlation matrices (as is the case in PFA) commonly produce real eigenvalues that surpass corresponding random data eigenvalues for trivial, negligible factors; that is, for those that explain a very limited amount of variance (Buja & Eyuboglu, 1992; O’Connor, 2000). These authors suggest that additional procedures be used to trim any trivial factors. Parallel analysis using PCA syntax was therefore used to do this. Using PCA syntax, eigenvalues for the real data exceeded the eigenvalues for random data in 10 cases.

Note that two additional first-order factor structures were inspected—a nine-factor structure and an 11-factor structure. Both of these provided a less theoretically meaningful interpretation than did the 10-factor solution, therefore the 10-factor solution was retained. The factor structures were also examined using a.40 cutoff criterion. However, the .30 criterion was retained as it was thought important to keep this consistent with previous factor analyses undertaken on the MSPSE and the SPPA.

MSPSE items = 3, 44, 45, 54, 55, 56, 57; SPPA items = 15, 29, 33, 42.

Note that two additional second-order structures were inspected—a two-factor structure and a three-factor structure. Both provided a less theoretically meaningful interpretation than did the four-factor solution, with a number of cross-factor loadings. The three-factor solution also contained a one-item factor. Therefore the four-factor solution was retained.