Random Assignment

Causality and Internal Validity

The most important tenet of establishing a causal connection between two variables is that there must be no other plausible explanation for the observed relationship between the variables. That is, to validly make the claim that (independent) variable A causes changes in outcome (or dependent variable or effect) variable B, all other potential causes of changes in B must be ruled out. The many other potential variables that may affect the outcome variable are referred to as confounding variables or nuisance variables. If one can effectively rule out every other explanation for the relationship between variables A and B, the study has good internal validity.

Randomly assigning participants to the various groups (or levels or values or conditions) of the independent variable increases internal validity and thus aids in establishing causation, because it helps to create the “everything else equal” condition: The randomization process roughly equates the groups on every potential confounding variable. Thus, any observed differences between the groups that are beyond what would be expected by chance can be attributed to the independent variable. The beauty of random assignment is that, if the sample size is sufficiently large, it assures that all determinants of B, even unknown and unspecified ones, are largely evenly distributed between the groups.

Consider the following example of a project to evaluate a smartphone-delivered mindfulness meditation (MM) program intended to help students’ critical thinking skills as compared to a sham program. The sham program was also delivered by smartphone and was described as meditation, but it did not provide guidance on how to control awareness of body or breath, critical elements of MM. From a pool of 91 college students, the researchers randomly assigned about half to participate in the true MM program and the other half to receive the sham program. After 6 weeks, the researcher compared the levels of critical thinking in the two groups. The researcher randomly assigned participants to the two groups, using randomizing software (a variety of software packages or macros are available for this purpose, such as Excel, R, SPSS, and randomizer.org). Because of this randomization, any preexisting differences in the levels of critical thinking skills, or any other possible confounding variable, should be approximately equated in these two groups. So, if the difference in the levels of critical thinking skills following treatment exceeds the small amount to be expected after randomization, the researcher could validly claim the difference is due to MM.

...