Wu, Chengzhang. The use of cloud data of IRS form 990 to advance research in not-for-profit institutions. Retrieved from https://doi.org/doi:10.7282/t3-xwmx-8q66
DescriptionThis dissertation consists of three chapters based on IRS Form 990, the return of organization exempted from income tax. Chapter one introduces building a new database that includes all the e-filed IRS Form 990 starting from 2011. The research data of the other two chapters is based on this new database. IRS Form 990 provides a rich set of information about not-for-profit organizations. Historically, these returns were publicly available in PDF format or through information aggregators that provided only part of the data. Beginning in 2011, forms that were electronically filed with the IRS have been available to the public in an XML format. The first chapter of this dissertation follows the design science paradigm to describe accessing the forms from AWS (Amazon Web Services), examining XML structures, and creating an updatable database. The resulting database was then used to demonstrate that the process provides information accessible for simple and complex inquiries. This database provides researchers access to data that has been out-of-reach and enables extensive data analysis.
The second chapter of this dissertation focuses on the not-for-profit’s governance level, which is measured by several responses in Form 990. Three hypotheses are proposed and tested. Specifically, this essay examines the quantitative association between both the level and the nature of asset diversion and corperate governance, as well as evaluates governance in the period after the diversion was reported. The research data used in the second essay is derived from the database built in the first study to demonstrate the usefulness of the Form 990 Database developed in chapter one. The second chapter of this dissertation examines the qualitative association between the level of asset diversion and the governance level.
The third chapter of this dissertation focuses on the dissolution of not-for-profit organizations and attempts to use several supervised machine learning algorithms to predict the discontinued operation of not-for-profit organizations. The accurate prediction would allow the donors to allocate their economic resources more rationally. The predicting factors are primarily important financial factors derived from prior studies. Three research questions are proposed and developed in this chapter. The prediction performances of Logistic Regression, Decision Tree, Random Forest, Neural Network, Support Vector Machine, and Bayes Net are compared.