Despite the foregoing cautions about the function and legitimate
warrants of correlational studies (i.e., studies lacking a researchercontrolled
treatment component), some authors of such studies are
nonetheless inclined to impute causal connections between methods/
treatments and outcomes. For example, a recent article in one
of the educational psychology journals reported the results of an
international study of the relationship between mathematics selfconcept
and achievement (Wilkins, 2004). The author “used univariate,
bivariate, and multivariate statistical techniques to investigate
the relationships between self-concept, achievement, gender,
and age” (p. 335). No variables were experimentally manipulated
and the analysis was strictly correlational (hierarchical linear modeling),
yet causal statements related to the findings and recommendations
such as the following appeared in the discussion section (p.
344):