Limitations in Sample-to-Population Generalizing
Generalizability of research fi ndings is one of the key criterion researchers use to determine the value and usefulness of research. Often research that utilizes larger samples of data is associated with greater degrees of generalizability. Research that uses large-scale data bases and statistical approaches to analyses (typically referred to as “quantitative”) is associated with higher levels of generalizability than research that utilizes interpretive approaches with small samples (typically referred to as “qualitative”). In a recent article, my colleague Wolff-Michael Roth and I argued that generalizability of research fi ndings cannot be judged based on sample size (Ercikan & Roth, 2006). We provided examples of research which use small samples, such as in the case of phenomenology, yet may identify universal relationships that have great degrees of generalizability. As well as examples of research that may use very large samples but focus on overall group results and have very little generalizability for key sub-groups. To demonstrate this point, we used research on gender differences in mathematics as an example:
Gender differences based on typical statistical analyses of group results are, in fact, differences in group means. These differences refl ected in the group means may have very different patterns for groups of students. Some students may have different mathematical skills, while others may have been taught different curricula using different instructional methods and their results do not provide any information about individual students. Therefore, the fi ndings are neither informative about individual students, nor do they provide insights about subgroups of students (e.g., high performing students who may have differential patterns of gender differences) that are not the focus of statistical comparisons. They are only relevant for the overall group the analyses focused on. (p. 21)
Generalizing from education research, typically, refers to the extent to which research fi ndings are applicable to a broader set of contexts.