Saturday, June 19, 2010

Is the SAT biased?

Jay Mathews reports that a controversial study found African Americans at a given score (e.g. 370) performed better on harder questions than Whites with the same score. He suggests this is evidence that smarter blacks are harmed by a bias in the easy question. The researcher has a implausible theory for why this might be the case. Commentators provide a more reasonable theory--the whites are more familiar with the material the hard questions test and are thus susceptile to fall into the "traps" the SAT test-makers design into the hard questions.

I think an important point left out of the debate is that Blacks with a given score tend to have lower college GPAs than Whites with the same scores. This means that, if we think that GPAs are the best available measure of academic achievement, the SAT is biased in favor of Blacks.

For race/ethnicity, the pattern of results supports previous findings. Specifically, American Indian, African American, and Hispanic students are overpredicted for all measures and combinations of measures. African American students’ FYGPAs tend to be the most overpredicted, with mean standardized residuals ranging from -0.32 to -0.17.

The previous finding summarized:

Ramist et al. found Asian American (mean residual = 0.08) and white (mean residual = 0.01) students were underpredicted while American Indian (mean residual = -0.29), black (mean residual = -0.23), and Hispanic (mean residual = -0.13) students were overpredicted.

I wouldn't necessarily interpret this as evidence the test is biased in favor of Blacks. Blacks with a given score tend to go to more selective colleges because of Affirmative action and thus lower GPAs would be expected since they compete with smarter peers.

Source: Differential Validity and Prediction of the SAT

No comments:

Post a Comment