Patterns of SAT Verbal and SAT Quantitative Scores Associated With College Major and College Success: Results From a Moderated Profile Analysis Technique

• In Event: Profile Analysis via Multivariate Statistics for Continuous and Categorical Data

Abstract

Objective. Park, Lubinski, and Benbow (2006) found that youth whose adult accomplishments were in STEM fields had different patterns of SAT Verbal (SATV) and Quantitative (SATQ) scores at age 13 as compared to youth whose accomplishments were in non-STEM fields. There are two objectives for the present research: (1) Replicate Park, et. al. using college major with SATQ and SATV in a sample spanning a broader range of ability, while using a more rigorous statistical method and (2) develop and provide an example of moderated profile analysis. The example will use profile patterns of SATQ and SATV to differentially predict success (GPA) for STEM versus non-STEM majors.

Theoretical Framework. In research on SAT profiles, individual differences are divided into differences in level as reflected in a composite score (SATV + SATQ) and differences in score patterns. Park, et. al. (2006) found choice of field was related to SAT pattern. Theirs was an unusual sample as subjects were age thirteen and from the top 1% of their peers. If SAT pattern is related to entry in a field, it need not relate to success. We test the relation of pattern and college success. We also test differences in SAT pattern associated with success in STEM versus non-STEM majors using moderated profile analysis. Moderation is increasing in popularity with whole texts devoted to regression interactions (e.g. Aiken and West, 1991). Moderation also forms the basis for regression discontinuity; one of the few statistical techniques acceptable to show causation (Schochet, et al, 2010).

Method. Using the regression approach in Davison & Davenport, 2002, one can take any set of predictors and reduce them to two entities; a level parameter and a pattern parameter. Thus, allowing one to investigate the predictability for choosing STEM versus non-STEM majors for level and/or pattern. This approach generalizes to a moderated profile analysis with the following form:
Y’p= b0+b1L1+b2L2+b3P1+b4P2. Specifically, for our analysis predicting GPAs, b0 is the intercept. L1 is the level value for non-STEM. L2 is the corresponding value for the STEM group. The P’s are the pattern parameters for each group. Moreover, L2 and P2 are moderators, allowing one to test whether the level and pattern parameters are the same for the two groups.

Data source. The 2001 edition of Baccalaureate and Beyond (http://nces.ed.gov/pubsearch/pubsinfo.asp?pubid=2003174) provides the data.

Results. Park et. al. was confirmed. While level was also important, pattern was more so. The moderated profile analysis showed that level was more important for both STEM and non-STEM students in predicting GPA. Moreover, the groups significantly differed for level. Finally, the STEM group had a significant pattern that was different from the non-significant pattern for the non-STEM group.

Scholarly Significance. We confirmed results of previous research with a more diverse sample and rigorous methodology. We developed Moderated Profile Analysis and showed an instance of its use. The final paper will present implications of the findings for methodology and prediction of college major and success.

Authors

• Ernest C. Davenport, Jr., University of Minnesota

• Mark L. Davison, University of Minnesota