Psychometrics

Origins

The origins of psychometrics can be traced to the work of experimental psychologists in the nineteenth century who sought to bring the rigor of “scientific methods” to the measurement of mental attributes. Following the traditions of physicists and other natural [Page 828]scientists, German psychologists sought to quantify the relationship between the intensity of a mental sensation and the magnitude of the stimulus needed to invoke the sensation. Similar efforts to quantify psychological attributes were also being developed in England in the mid-nineteenth century. Influenced by the work of Charles Darwin, his cousin, Francis Galton, sought to develop scientific measures and procedures for studying the relationship between intelligence and other characteristics of individuals. Because of his development of mental tests, use of the normal distribution and work on the statistical techniques of correlation and regression, he is considered by many to be a key figure in the development of modern psychometrics. However, it is Charles Spearman who is credited with development of the classical model of mental test scores, introducing the term reliability coefficient, and creating factor analysis and the two-factor theory of intelligence during the first decade of the twentieth century.

Psychometrics is a relatively young science but one that has seen tremendous advances in the past 100 years. When Spearman introduced the concept of the reliability of a measure, his intent was to show how measurement errors adversely affected the measurement of an attribute and the resulting correlation of that attribute with any other. His definition relied on the possibility of measuring a single trait with two similar measures. Since his initial work, the concept of reliability has changed tremendously. Researchers have expanded the concept from a focus on two measures to many and shown that as the number of similar measures increases, reliability increases. The traditional view or model of reliability did not permit researchers to differentiate sources of measurement errors. The emergence of generalizability theory not only permitted researchers to isolate the source of measurement errors, but equally important, it made it possible to plan a measurement process so as to minimize specific sources of error. Recent developments in reliability have emphasized the fact that reliability coefficients reflect the influence of measurement error on test scores and can vary from one group to another.

There have also been significant developments in mathematical models of test scores. The simple classical model guided test development and use for most of the early part of the twentieth century. However, in the 1950s, Fred Lord introduced a series of models that made strong assumptions about the relationship between observed test performance and the underlying trait measured by the test but yielded many advantages over the classical model. The so-called item response theory models explicitly modeled the probability that an examinee would respond correctly to an item as a function of his or her ability. The shift in focus permitted test developers to develop tests that did not rely for their properties on the specific group of examinees on which they were field tested or normed and the measurement of an examinee's ability independent of the specific group of items which he or she happened to take. This latter result facilitated the development of computerized adaptive tests that measure examinees with considerable accuracy in less time by targeting their ability level based on their response patterns to selected items. Item response theory models have become the dominant theory behind most large-scale testing programs. They have developed significantly over the past 50 years. They now include multivariate models, models that permit ordinal or nominal test items, and even models that generate ability estimates for multiple aggregates such as classes or schools.

...