A sea change in statistics: A reconsideration of what is important in the age of big data

Abstract

The development of information technology (IT) in the last decade has ushered us into the ‘big data’ era. Whereas sampling error has been of high interest in such studies over the last century, extremely large datasets will virtually rule out the possibility that statistics generated from samples are not also true of the population. The critical discussion must move on to how much the exogenous and endogenous variables explain ultimate outcomes. The stress will have to move from generalizability to explanatory power. By examining the three common statistical tests – the T-value, the F-value and the beta coefficient (or ‘b’ in the sample) – this paper aims to show how important effect size analysis is with the trend toward larger and larger samples and big data. In the end, the paper points out the gap of what the acceptable ranges of effects that appears in our sundry scientific literatures are. We call for studies in this regard so that social sciences can move forward and strengthen weak theory bases, and recognize strong theories when they do appear.