What do educational test scores really measure?

Latent class Poisson count models are used to analyze a sample of Danish test score results from a cohort of individuals born in 1954-55 and tested in 1968. The procedure takes account of unobservable effects as well as excessive zeros in the data. The bulk of unobservable effects are uncorrelated with observable parental attributes and, thus, are environmental rather than genetic in origin. We show that the test scores measure manifest or measured ability as it has evolved over the life of the respondent and is, thus, more a product of the human capital formation process than some latent or fundamental measure of pure cognitive ability. We find that variables which are not closely associated with traditional notions of intelligence explain a significant proportion of the variation in test scores. This adds to the complexity of interpreting test scores and suggests that school culture, attitudes, and possible incentive problems make it more difficult to elicit true values of what the tests measure.