GPA alone did not explain 20.4% of the variance in the UC study. Model 1 did explain 20.4% of the variance, but included the additional variables of parental education, family income, and school API rank. Those last three variables, while having little correlation to unweighted HSGPA, do have significant correlations to SAT scores. If you removed those variables, and looked at unweighted HSGPA alone, you would get a lower percent of the variance explained for GPA.