Test score differences

Most of the evidence of intelligence differences between racial and ethnic groups is based on studies of intelligence test scores. Intelligence tests measure many important abilities, such as verbal and quantitative reasoning, and can predict socially-relevant outcomes such as academic performance and occupational outcomes. However, intelligence test scores do not reflect all of the intricacies of the everyday meaning of intelligence, so researchers take care to distinguish between IQ test results and intelligence.^[6][11]
Some studies of intelligence tests use statistical methods to extract so-called latent variables from the IQ test scores. One such variable is the general intelligence factor, or g, which accounts for most of the differences in IQ test scores between individuals. There are other latent variables in addition to g, and IQ tests vary in their ability to measure these latent variables, if they measure them at all. IQ tests scores, while often summarized as a single overall number, are actually multidimensional in nature. Transforming IQ test scores into latent variables is an attempt to find one or more dimensions on which to compare IQ test scores.
Latent variables are also sometimes called factors or constructs. The construct validity of an IQ test score is a key criteria for judging whether the IQ test score differences are meaningful. Tests which do not measure difference in latent variables for some group are said to have measurement bias. The construct validity of most commonly used IQ tests has been fairly well established within multiple racial-ethnic groups in developing countries such as the United States. That is, test score difference within each racial-ethnic group are valid indicators of differences in latent variables such as g. A related question is whether test-score differences between groups are valid. There is a consensus that test score differences between Black and White people in the United States have predictive validity (also called predictive invariance), meaning that test scores predict the same socially-relevant outcomes regardless of the race of the person being tested. To further address this question, three studies using sophisticated statistical techniques have shown that Black-White differences in IQ test scores are not a result of measurement bias (a criteria called measurement invariance).^[12][13][14] These studies imply that Black-White IQ differences reflect very general differences in some underlying latent variables, but they are unable to differentiate precisely which latent variables differ under a variety of models. These studies were performed in response to previous investigations which suggested that Black-White IQ differences are primarily differences in g in particular.^[15]