Abstract
More data does not necessarily mean more information: this is the big data paradox, and it should be more widely disseminated as the essential point is that caution should be exercised regarding inferences made in areas where large amounts of data are routinely used, such as data mining and artificial intelligence. In this article, the assumptions underpinning methods of statistical inference are considered, and whilst departures from these assumptions may have little effect when the amount of data (sample size) is relatively small, there can be substantial cumulative effects from increasing sample size. Consideration is also given to assessing any practical significance from the effect size relative to variation in the population to which any inferences might apply. Emphasizing assumptions and exploring their effects could (should?) be incorporated as adjunct material in lectures/classes on statistical inference.