Separating power and speed components of standardized intelligence measures

Abstract

Gulliksen (1950) established the distinction between pure power and pure speed intelligence tests. Most standardized measures combine power and speed requirements. The speediness component affects test scores' reliability and validity, since it involves variance not due to the mental ability of interest. Here we propose the use of the Stafford's Speediness Quotient (SQ, 1971) for identifying items biased by the speed component. We developed two converging methods based on Structural Equation Modeling (SEM) to assess the validity of the SQ index. The methods concurrently identify items substantially affected by speededness in three standardized fluid intelligence tests with different speed requirements. Basing on the SQ at the item level, a simple strategy for separating the power and speed components of mental ability tests applied under time constraints is proposed. This strategy allows an estimation of the respondents' level uncontaminated by the speed unwanted variance. This procedure only requires right/wrong responses (e.g., does not need any information external to the test, such as response times) and it is appropriate for medium-small sized samples. A rule of thumb is suggested for identifying items affected by speediness. The simplicity of the proposed procedure allows its use in applied settings for detecting and controlling speed-related variance in tests' scores.