Big data raise questions about the quality of the data and the generalizability of results. In research where data are gathered through the Internet or other media, researchers may have little control over who is providing the data, and information about the background characteristics of participants may be difficult to obtain. This will lead to uncertainty about the population for which the research findings will apply (Hargittai, 2015). Moreover, with big sample sizes, it is a concern that even a small selection bias may lead to a false rejection of the null hypothesis (Ioannidis, 2005). Besides reflecting on data quality, big data call for the development of suitable statistical tools. Conventional statistical methods taught in graduate schools (e.g., Aiken, West, & Millsap, 2008) may not be sufficient in the era of big data. Several issues can be involved when applying conventional statistical methods to big data. First, the datasets may be too large for standard workstations, so that researchers` computers cannot be used to handle the data. Second, the data may involve different types of information such as numerical data, text, sound, and videos. Conventional multivariate techniques such as multiple regression may not be optimal for handling such different types of variables. Third, the large sample sizes may lead to statistically significant results in most statistical analyses, even when the associated effect sizes are practically trivial. If researchers use the significance test as the criterion in, for example, judging the relevance of the predictors, misleading conclusions can be made.
© Copyright 2018 Zeitschrift für Psychologie. Hogrefe. Alle Rechte vorbehalten.