Date of Submission
Open Access Senior Thesis
Bachelor of Arts
Regression splines have an established value for producing quality fit at a relatively low-degree polynomial. This paper explores the implications of adopting new methods for knot selection in tandem with established methodology from the current literature. Structural features of generated datasets, as well as residuals collected from sequential iterative models are used to augment the equidistant knot selection process. From analyzing a simulated dataset and an application onto the Racial Animus dataset, I find that a B-spline basis paired with equally-spaced knots remains the best choice when data are evenly distributed, even when structural features of a dataset are known and implementable. However, the residual-based knot selection outperforms both the equidistant knot placement and structural knot placement methods when data are irregularly distributed.
Klein, William, "An Evaluation of Knot Placement Strategies for Spline Regression" (2021). CMC Senior Theses. 2545.