THE EXCLUSION FROM INCOME OF SCHOLARSHIPS AND FELLOWSHIP GRANTS: AN EMPIRICAL INVESTIGATION OF TAX COURT DETERMINATIONS

The primary objective of this research was the development of a model that would identify and rank in order of importance the controlling facts used by the Tax Court in determining the taxability or nontaxability of scholarships and fellowship grants. To accomplish this objective data based on factors contained in the regulations of Section 117 and the Tax Court decisions were specified as 14 qualitative variables. A random sample of 89 cases was selected. The decision of the court either in favor of the government or the taxpayer provided the dichotomous dependent variable for assigning each case to one of the two decision groups. Each case was analyzed for the 14 qualitative independent variables. Each variable was coded to indicate the specification of the court as to its presence, absence, or failure to be mentioned. A linear discriminant model containing four variables was determined to be the best model based on a classification accuracy of 90 percent using the Lachenbruch U (jackknife) method. The best (4-variable) model was used to test the null hypothesis to determine if the results of the best model were statistically significant relative to the results expected by chance. The hypothesis was rejected at the five percent significance level. Various sensitivity analyses were performed to evaluate the stability of the model parameters. The structural sensitivity analyses included (1) tests of stability over time, (2) tests of stability between regular Tax Court and Tax Court memorandum decision, and (3) identification of the Tax Court judges for the sample cases. The best (4-variable) model was stable over time and between types of court decisions. Individual judges showed no bias. Data sensitivity analyses included (1) tests for measurement errors and (2) misclassification costs. The best (4-variable) model was stable after introducing a three percent random measurement error for the independent variables. Misclassification costs at a ratio of five to one were necessary to correctly classify all but two of the decisions for the taxpayer