Can We Feel Confident in How We Measure College Confidence?

Abstract

Two studies were conducted to examine validity evidence for the College Self-Efficacy Inventory, by investigating dimensionality and theoretically based relationships with external criteria. Modifications to the scale yielded an adequately fitting three-factor model, and most hypothesized relationships were empirically supported. However, continued refinement of the instrument is recommended owing to construct underrepresentation.

• self-efficacy
• validity
• confirmatory factor analysis

1. 1. This model, with a second-order College Self-Efficacy factor predicting the four first-order factors, is more parsimonious than the correlated four-factor model and should thus be estimated only if the more complex, correlated four-factor model has adequate model—data fit. Moreover, a model that specifies a second-order factor predicting three first-order factors would fit equivalently to the correlated three-factor model and thus was not estimated.

2. 2. One might question whether the differences in the number and magnitude of the standardized residuals were due to differences in sample sizes rather than to differences in the level of control, given that the standardized covariance residuals used to examine misfit are computed by dividing the covariance residuals by the standard error. Because standard errors are affected by the sample size—that is, smaller samples tend to yield larger standard errors and may in turn lead to smaller standardized covariance residuals)—it is possible that the large residuals in Samples 1 and 2 were due to their large sample size and that the small residuals in Sample 3 were due its small sample size . To ensure that this was not a plausible explanation for the pattern of results, all analyses were conducted a second time, using the correlation matrix as input (i.e., the standardized covariances); when conducted in this manner, correlation residuals are computed, which are not affected by the standard error and, consequently, the sample size. The results indicated that the correlation residuals followed a similar pattern and had similar relative magnitudes, thereby providing evidence that the differences in the number and magnitude of standardized covariance residuals across the samples were a function of the testing condition rather than the sample size.

Additional information

Notes on contributors

Carol L. Barry

Carol L. Barry is a doctoral student in the Assessment and Measurement program at James Madison University. Her research interests involve applications of structural equation modeling, mixture modeling, and growth modeling. Her substantive interests include noncognitive constructs such as self-efficacy, helpseeking, goal orientation, and test-taking motivation.

Sara J. Finney

Sara J. Finney is an associate professor in the Department of Graduate Psychology at James Madison University. The majority of her research involves the use of structural equation modeling to gather construct validity evidence for various self-report measures. Her other research interests include mixture modeling and practical issues in the application of structural equation models.