Because some component of a test score is due to error ( as opposed to true score ), extreme scores on one test are likely to be closer to the mean on a second test, thus posing a threat to the validity of an experiment in which extreme scores are selected; the amount of regression will be greater for less reliable tests.