RLR
Purpose of Regression Fit data to model Known model based on physics P* = exp[A - B/(T+C)] Antoine eq. Assumed correlation y = a + b*x1+c*x2 Use model Interpolate Extrapolate (use extreme caution) Identify outliers Identify trends in data
Linear Regression There are two classes of regressions Linear Non-linear “Linear” refers to the parameters Sensitivity coefficients of linear models contain no model parameters.
Which of these models are linear?
Example: Surface Tension Model
Issue 1: Nonlinear vs. Linear Regression Nonlinear model Linearized model
Nonlinear Regression: Mathcad - GENFIT
Nonlinear Regression Results
Linear Regression: Mathcad - Linfit Does the linear regression Redefine the dependent variable Defines the independent variables
Linear Regression Results
Comparison nonlinear linear
Issue 2: How many parameters? Linear regressions with 2, 3,4, and 5 parameters
Straight Line Model as Example
Fit a Line Through This Data
Least Squares
How “Good” is the Fit? 1. What is the R 2 value Useful statistic, but not definitive Does tell you how well model fits the data Does not tell you that the model is correct Tells you how much of the distribution about the mean is described by the model
Problems with R 2
How “Good” is the Fit? 2. Are residuals random
Residuals Should Be Normally Distributed
How “Good” is the Fit? 3. Find Confidence Interval
Parameter Confidence Level
Confidence Level of y
Multiple Linear Regression: Mathcad - Regress
Mathcad Regress Function
Results on Ycalc vs Y Plot
Residuals
R 2 Statistic
Confidence Level for Parameters n is number of points, kk is number of independent variables
Confidence Level for Ycalc