No open-access location is known for this record yet.
No open-access location is known for this record yet.
Test bias research that examines the predictive accuracy of the test for different groups typically is based on the assumption that the outcome variable (e.g., (first-year averages]) is not biased. Research to test the accuracy of that assumption is lacking.2 - Linda F. Wightman I. Introduction The ability to predict first-year law school grades is the raison d'etre for using the Law School Admission Test (LSAT) to select law school students.3 Therefore, any analysis of gender or racial/ethnic bias on the LSAT must confront the issue of how test bias (in a predictive sense) is predominantly defined and measured. To point out problematic assumptions with conventional wisdom regarding the LSAT, I will attempt to reveal how unfairly the standard definition of test bias treats a broad range of groups. In this article, I argue that the conventional definition of test bias-the linear regression model-can actually mask bias by relying on a criterion (law school grades) that also may be contaminated by bias against women, people of color, and a host of other outsider groups. This critique moves back and forth between satire4 and commentary and analysis.5 This format highlights the ways in which conventional wisdom about test bias logically leads to shocking and unacceptable social consequences. In addition, I will argue that the regression model rests upon internally contradictory and socially decontextualized assumptions. While this article involves a number of technical issues, I hope that the vehicle of satire will facilitate an inclusive audience by allowing those readers with little or no background in psychometrics to still get an intuitive sense of the high-stakes social justice concerns imbedded in seemingly neutral law school admission policy decisions.6 I will argue that the Law School Admission Council (LSAC) relies too heavily upon a criterion-related definition of test bias-while at the same time failing to validate the fairness of that criterion. Consequently, I argue that it is far from established that the LSAT validly measures the construct of potential for the study of law. Samuel Messick, a senior researcher at the Educational Testing Service, suggests that the method by which one can transcend the circularity of criterion-related validity7 is to evaluate (and inform) expert judgment on the basis of other evidence about the structure of the behavioral domain under scrutiny.8 With Messick's suggestion as a starting point, I review decades of legal education research to show how the environment of law school can systematically disadvantage outsiders based on their gender, race, ethnicity, personality type, commitment to public interest work, and socioeconomic background-even when entering credentials (LSAT scores and UGPAs) are held constant. From this finding that most groups tend to be overpredicted-that is, perform less well than others with equal credentials-I argue that the behavioral domain of legal education is itself contaminated by bias. I therefore argue that the correlation between law school grades and LSAT scores-which is the basic justification for using the test-is, by implication, also questionable. This article is an attempt to use the tools9 of educational measurement if not to dismantle the master's house of the LSAT, to at least reveal how the foundational assumptions that support the conventional wisdom about test bias and predictive validity are untenable, socially regressive, and result in systematic blaming of the victim. II. The Chronicle of the Discriminatory Law School A. The Regression Model of Test Bias Undoubtedly, the most frequent response to the criticism that the LSAT is biased against women or people of color is the assertion that the LSAT has equal if not better predictive validity for women compared to men or people of color compared to whites. In other words, when measured against the criterion variable of first-year law school grades, the LSAT does an equally good (if not better) job of predicting the performance of women and people of color. …
Loading…
Imported from OpenAlex (CC0) — nothing here is generated.