The intention of the present investigation was to demonstrate the causal link between Prior Mathematics/Statistics Experiences and Statistics Self-Perception, using structural equation modeling techniques. This was accomplished by first establishing the discriminant validity of three theoretical constructs, Statistics-Related Self-Efficacy, Statistics-Related Attitudes, and Statistics-Related Anxiety, while also demonstrating that these three factors were significantly related, consistent with prior research.; Since these factors were indeed significantly related, the third-order factor, Statistics Self-Perception, which effectively accounts for the variance shared between the three lower-order factors, was introduced into the model. Although third-order factor structures are theoretically supported in the literature, existing research has been limited to second-order structures.; The number of observed items in the third-order factor model was reduced in an effort to arrive at a model that is more parsimonious, while also measurement invariant and structurally invariant relative to the original model. Item reduction has been attempted in previous research using SEM approaches, however these studies do not present clear evidence that model changes resulted in models that were measurement invariant while maintaining the integrity of the structural model.; The resultant model was used to demonstrate the link between Prior Mathematics/Statistics Experiences and Statistics Self-Perception, as well as the link between Prior Mathematics/Statistics Experiences and the primary factors in this study. Prior research has suggested that a Prior Mathematics/Statistics Experiences is related to the primary factors of Statistics Self-Perception; however, this has not been demonstrated with causal modeling.; Finally, the present investigation demonstrated the benefit of latent factor scores produced in structural modeling. Unlike factor scores produced in factor analysis, latent variables scores computed from a structural model are unbiased estimates, and therefore are error free estimates. Post-modeling, these factor scores provide more accurate data with which to examine the effect of variables, such as participants' gender, age, ethnicity, department, GPA, expected GPA, degree being sought, number of prior college-level mathematics/statistics classes, and hours towards degree on the primary factors used in this investigation. This innovative use of factor scores provided greater flexibility with data produced from structural modeling, not seen in prior research.
展开▼
