Essays on data analytics

Abstract

Essay 1. Data preprocessing evaluation: an extended role of Benford’s Law Sophisticated anti-fraud systems for the healthcare sector have been built based on several statistical methods such as neural networks, decision trees, and genetic algorithms. However, these algorithms consume considerable time and cost, and lack a theoretical basis to handle large-scale data. Hence, this study proposes an efficient data analysis solution in terms of scalability and speed for handling large volumes of healthcare data. We first extend mathematical theory to demonstrate the manner in which large-scale data conform to Benford’s Law: the aggregated data from various sources of distribution follow the Benford distribution. Then, we test its applicability empirically using actual large-scale healthcare data from Korea’s Health Insurance Review and Assessment (HIRA)-ational Patient Sample (NPS) of symptoms, treatment, and medical costs for patients. Given that government health departments and private insurance firms have rapidly aggregated and digitized large and complex datasets, Benford’s Law can be used as a suitable instrument for detecting irregularities in payment for medical services from large-scale healthcare data. Thereby, it can also help them reduce administrative time and expenses, with unnecessary medical expenses kept under control. Essay 2. Quantitative marketing model for context-dependent consumption preferences Marketers have long been interested in understanding how, and the extent to which, consumer choices may be influenced by the context in which the product is consumed. In this paper, we develop a parsimonious context-dependent multidimensional unfolding (CDMDU) model that can accommodate consumers’ context-specific ideal points in multi-attribute space along with brand locations in that space. The specification allows for unobserved heterogeneity via a normal distribution on attribute weights and a discrete distribution on brand locations and ideal points. The CDMDU model is flexible and reduces to a factor structure random coefficients brand choice model when there is only one consumption context. We also demonstrate how the CDMDU model can be used to derive a firm’s optimal direction of brand re-positioning given its competitive landscape in the various consumption contexts and provide an empirical illustration using panel data from consumers in the U.S. beer market. A key observation when repositioning a brand is that consumer preferences can be correlated across contexts

so a movement towards the ideal point in one particular context does not necessarily improve the firm’s market competitiveness in other consumption contexts and can therefore hurt its overall performance in the market.