Test Bias: Validity of the Sat for Blacks and Whites in Thirteen Integrated Institutions

This report reviews and summarizes the data of published reports and papers pertaining to the validity of measures commonly used at college entrance. These measures consist of various types of quantitative indices used to describe the high school record and various types of academic tests. The data were categorized by sex and ethnic classifications where possible and comparisons were made. Most of the observations made in previous reviews of this kind were upheld. Both the high school record and standardized academic tests appear to be generally useful at college entrance. Both tend to predict college performance, but they are most appropriately used in combination. For Anglo populations, the high school record usually yields slightly higher predictive correlations. For black populations (except black females), however, the data suggest no superiority of the high school record as a predictor of college performance. The amount of data available for Chicano samples was not sufficient for any generalizations about predictive correlations, but the magnitude of those correlations available were on the average lower than those for Anglo populations. For recent years, correlations for black samples tended to be slightly lower than those for Anglo samples. A primary objective in the review was to consider whether inferences made from the high school record and test scores about different populations tended to be correct inferences. Where inferences tended to be systematically incorrect, there was an interest in the direction of the error. Since the high school record and test scores are used in making inferences about probable college performance, the systematic inferential errors of interest consisted of the degree to which college performance of populations were over or underpredicted when the data used in making the inference were not limited to data for that population. Data from a number of studies show that the college performance of black populations, male and female, have been consistently overpredicted by the traditional academic predictors (the high school record and standardized test scores) when predictions are based on data from white or predominantly white samples. The high school record would appear, from the studies reviewed, to be the principal source of this overprediction. Standardized test scores also have tended to overpredict college performance of black populations, but a combination of the high school record and test scores appears to minimize over prediction. In contrast, published data show women to be consistently under predicted by the traditional predictors when predictions are based on data from male or predominantly male samples. This review indicates a number of topics where further research appears to be needed. These topics include the study of: long‐range criteria, objective criteria, prediction within major‐field, the high school record for different populations, college grades for different populations, neglected populations, factorial consistency of measures across populatio...

Section: From This Perspective One Does Not Validate Tests or Test Itmentioning

confidence: 63%

“…Hoeg (1971) and Temp (1971) dat~using Thorndike model suggests bias against blacks taken as a group. Cleary (1968) and Temp (1971) data suggests bias' against blacks taken as a group.…”

Section: Reanalysis Of Davis and Kern~rmentioning

confidence: 99%

Section: From This Perspective One Does Not Validate Tests or Test Itmentioning

confidence: 99%

Section: From This Perspective One Does Not Validate Tests or Test Itmentioning

confidence: 99%

See 2 more Smart Citations

Population Validity and College Entrance Measures

Breland

1978

“…The special case where P23 = P 12 P 13 will result in equal (1) The slopes of the regression of Y2 on Y l will be the same for both conditional distributions since the covariance matrix V* remains invariant for all values of (2) The intercepts for the conditional distributions will be Although a number of studies have reported parallel regression lines (e.g., Cleary, 1968;Temp, 1971) a recent report by Campbell,Pike,and Flaugher (190) is particularly worth noting because the criterion employed was a carefully constructed, objective, job-knOWledge test. For seven of eight predictors studied the regression lines for minority and majority groups were parallel with lower intercepts for the culturally disadvantaged group.…”

Section: A Note On T-linority Group Test Bias Sttjdies Richard R Reimentioning

confidence: 99%

A Note on Minority Group Test Bias Studies

Reilly

1973

Comparisons of majority and minority group regression lines for purposes of assessing test bias may, under certain conditions, be viewed as comparisons of conditional bivariate distributions. Where these conditions hold, findings should reveal parallel regression lines except for a special case. Given the conditions described, one implication is that even when the test is a parallel form of the criterion, lines with equal slopes but unequal intercepts should be found.

Comparison of a Bayesian and a Least Squares Method of Educational Prediction1

Boldt

1975

The prediction systems under discussion apply where the following conditions obtain: Predictor data are given on the same scale, criterion scores may be given on different scales, and it is necessary to pool data even though criterion scale differences exist. Such a system may be needed for minority group or graduate student prediction where the group sizes are small. Least squares and Bayes methods are used in a cross‐validation study conducted for comparison purposes. Data for the study were taken from the files of the Validity Study Service of the College Entrance Examination Board. A very limited amount of data were supplied by a few American graduate schools. The Bayes method was better, but it was found that both methods yield negative regression weights; when the absolute values of the weights were used, the methods were both improved and yielded results which were very similar in terms of evaluative statistics computed in the cross sample.