Using Database Approach, With Big Data And Unsupervised Machine Learning To Model Tax Behavior In The Expatriate Community

Abstract

The researchers interpreted survey results from the Moral Obligation of Paying Just Taxes Survey to identify a range of potential constructs, 18 constructs were identified. Using a big-data strategy, data was collected on these 18 constructs. Utilizing effective targeting, 2090 pages of data were collected with 377,783 words, on the tax evasion/tax avoidance topic. Data was scrubbed for pre-processing for analysis with KH Coder. Interpretation of the data indicated the 18 constructs could be collapsed into seven overarching constructs. It was upon these seven ‘master’ constructs that the literature review was prepared. Moving to data analysis, with KH coder, for text analysis and using machine learning output from a Co-occurrence Network, Co-occurrence Network idealized a triangular structure with three extremities. Correspondence analysis, supported this trend, while Multi-Dimensional Scaling with a stress factor of .170, helped specify a new model for tax avoidance and tax evasion. 1)Taxing the rich, 2) Enforcement strategies, 3) Tax planning for business, 4) Capital Gains tax, 5) Wealth and power inequality, 6) Economic effects of taxation, 7) Audits and materiality