This study examined the data selection process preceding multivariate analysis for a data set measuring student motivation and self-regulated learning. Data were 138 responses to a questionnaire on motivation and self-regulated learning, adapted for Finnish students. The first goal was to compare the results gained with "gentle" and "Draconian" variable selection. The second goal was to compare traditional linear analysis and Bayesian path analysis. A mechanical comparison of both solutions was carried out with a Bayesian Belief Network evaluator. The main purpose of the comparison is to show that the educational researcher should consider analyzing discrete qualitative (nominal, ordinal scale) variables with nonlinear Bayesian methods instead of linear methods that usually require quantitative (interval, ratio scale) variables. A later stage of the project will attempt to produce sensible path models in confirmation of the theory using structural equation model programs. (Contains 6 figures, 9 tables, and 84 references.) (SLD)