Evaluating the Effect of Data Aggregation on Statistical Inference and Loss of Variability Information

Posted: Oct 01, 2023

• Alexa Brooks
• Alexander Bennett
• Amelia Rogers

Abstract

Data aggregation represents one of the most ubiquitous practices in contemporary data analysis, serving as a cornerstone technique for managing large-scale datasets, reducing computational overhead, and facilitating interpretable summaries. The process of combining multiple data points into aggregated representations—whether through averaging, summation, or other summary statistics—has become so ingrained in analytical workflows that its potential consequences for statistical inference are often overlooked. While aggregation undoubtedly offers practical advantages, its methodological implications extend far beyond mere data compression, potentially distorting the very statistical properties that analysts seek to understand. This research addresses a critical gap in the statistical literature: the systematic evaluation of how data aggregation affects statistical inference and the concomitant loss of variability information. Traditional approaches to aggregation have primarily focused on computational efficiency and data reduction ratios, with insufficient attention to the inferential consequences of transforming raw data into aggregated forms. Our work introduces a comprehensive framework for assessing aggregation effects across multiple dimensions of statistical analysis, moving beyond conventional wisdom to provide empirical evidence of aggregation-induced biases. We posit that aggregation is not a neutral transformation but rather an information-processing operation that selectively preserves certain data characteristics while discarding others. The central thesis of this paper is that the loss of variability information through aggregation systematically distorts