“Actually that is exactly what the study says. Both are predictors, but GPA is a better one.”
- I do wonder if that's what the study is actually saying. In a regression analysis, "better predictor" refers to the statistical validity of the variable, not whether it leads to a certain estimate of the dependent variable (in this case, odds of finishing college in 6 years). If ACT is better than GPA at nailing the odds, then ACT is a "better predictor" than GPA even if the expected odds are lower. Simply because there is more confidence surrounding the odds estimate.
It’s important to point out that the CPS data, including the 11th grade required ACT, is a treasure trove for a social scientist; however, it’s not surprising to see that predictive abilities would be consistent across high schools in the same district. Whether that conclusion can be generalized will depend on similar analyses using data from other districts, even those with different characteristics (ie suburban or rural).