Modelling Data: Knowledge Discovery or Empirical Dexterity?

Abstract

Modelling with data is an important academic activity in almost every area of study. Modelling uses statistically based methodologies to seek out patterns amidst variability in data. Such patterns may be described through theoretical models or es fablished from data analyses. The data can come from tightly controlled and replicable experiments that make for elear statement of the nature of variability. Alternatively, dala may be oblained through observation; in which case the nature of variability may be less elearand often the exploratory nalure of the investigation make the statistical questions less well defined. Although in both cases sound statistical methods may be used, gOingfrom data to the answers requires careful selection, development and testing of appropriate models that seek to quantify and synthesize often complex relationships. The uncertainty in variability makes this process less well-defined and contentious. The merit in statistical modelling in creafing knowledge has been acknawledged ill many areas; for example, it remains an essential slep in I~ droe/opmenl and adoption (u sually based on experimental dala) and roen discontinuance (often based all observational data) of medical drugs. On the other hand, it led 10 tI,. accusation lhat economelrics (the application of slalistics in the field of economics) is little more than "junk science" in reference to diametrically opposed cone/usions about tire efficacy of gun control laws by two groups of researchers working on the same set of data! Does modelling lead to true knowledge creation or does it merely reflect empirical dexterity on the part of the researcher? This lecture considers Ihe debate and highlights the need for a proper modelling procedure to get from data to knawledge.