Regression with Differential Equation Models
Abstract
Regression analysis normally implies the use of algebraic equations to describe a system; however, some cases would better be modeled by differential equations. This is accomplished by assuming a differential equation model for a given set of data and estimating the values of the unkown parameters within the model. These values are then systematically perturbed to generate particular solutions which are superimposed to yield a better estimate of the unknowns. This process is repeated until a specified accuracy is met. Through an analysis of variance, the statistical characteristics of linear regression can be generated for most nth order differential equations. This provides a basis for evaluating the 'acceptance or rejection' of the regression. The characteristics generated consist of an ANOVA table (uncorrected), general F test on the regression, the R2 value, covariance matrix of the superposition constants, an estimate of the variance about the regression, an estimate of the variance of the parameters, and the confidence intervals on these estimates.
Document Details
- Document Type
- Technical Report
- Publication Date
- Apr 01, 1976
- Accession Number
- ADA026446
Entities
People
- Craig D. Hunter