Test validity defined as d-connection between target and measured attribute: Expanding the causal definition of Borsboom et al. (2004)

This article proposes a modification of the test validity definition put forward by Borsboom, Mellenbergh and van Heerden (2004). According to the original definition, a test is only valid if test outcomes are caused by variation in the target attribute. However, given that the goal of measurement is to derive a procedure for estimating true values of an unmeasured attribute, the requirement of causal influence of attribute upon measurement put forward by Borsboom et al. (2004) can be relaxed to a requirement of d-connection between the test and attribute. According to the d-connection definition of test validity, a test is valid for measuring an attribute if (a) the attribute exists, and (b) variation in the attribute is d-connected to variation in the measurement outcomes. In other words, a test is valid whenever test outcomes inform us either about what has happened to the target attribute in the past, or about what will happen to the target attribute in the future. Thus, the d-connection definition expands the number of scenarios in which a test can be considered valid. Defining test validity as d-connection between target and measured attribute situates the validity concept squarely within the structural causal modeling framework of Pearl (2009).