Chunk #40 — Reasons to be Concerned about the Published cGxE Literature — Problems with the Recipe: Statistical Concerns in cGxE Research — The importance of scale
First, evidence for interactions can depend on choice of scale as well as choice of statistical model. Despite one’s conviction in the presence or absence of cGxE, interactions are statistical phenomena and only have meaning in the context of a specific statistical and measurement model. However, many behavioral measures have no “true scale.” One might, for instance, argue that a construct like height has a true scale, and, moreover, a ratio scale with a meaningful zero point and equal intervals between data points (Stevens, 1946): a board of two feet is twice as long as a board one foot long. However, many, if not most of our constructs in the behavioral sciences, do not have meaningful zero points and, thus, are scaled somewhat arbitrarily. Quantitative scales without meaningful zero points can vary further as to whether they have meaningful intervals between measurement points or simply reflect differences in relative magnitude. This is true of measures of both behavioral outcomes (e.g., a depression score) and environments (e.g., family function, peer deviance, neighborhood disintegration). The scale of measurement matters profoundly in interaction
