This preview has intentionally blurred sections. Sign up to view the full version.
View Full DocumentThis preview has intentionally blurred sections. Sign up to view the full version.
View Full DocumentThis preview has intentionally blurred sections. Sign up to view the full version.
View Full Document
Unformatted text preview: NONLINEAR MODELS I So far the models we have studied this semester have been linear in the sense that our model for the mean has been a linear function of the parameters. I We have assumed E ( y ) = X β I f ( X i , β ) = X i β is said to be linear in the parameters of β because X i β = X i 1 β 1 + X i 2 β 2 + ... + X ip β p is a linear combination of β 1 , β 2 ,..., β p . I f ( X i , β ) = X i β is linear in β even if the predictor variables, the X s are nonlinear functions of other variables. c 2011 Dept. Statistics (Iowa State University) Stat 511 section 25 1 / 40 I For example, if X i 1 = 1 X i 2 = Amount of fertilizer applied to plot i X i 3 = ( Amount of fetrtilizer applied to plot i ) 2 X i 4 = log ( Concentration of fungicide on plot i ) I f ( X i , β ) = X i β = X i 1 β 1 + X i 2 β 2 + X i 3 β 3 + X i 4 β 4 = β 1 + fert i β 2 + fert 2 i β 3 + log ( ( fung ) i ) β 4 is still linear in the parameters β 1 , β 2 , β 3 , β 4 . I Now, we consider nonlinear models for the mean E ( y i ) . I These are models where f ( X i , β ) cannot be written as a linear combination of β 1 , β 2 ,.., β p I Small digression: What about models that can be transformed to be linear in the parameters? c 2011 Dept. Statistics (Iowa State University) Stat 511 section 25 2 / 40 linearizing a nonlinear model I Example: MichaelisMenton enzyme kinetics model v s = v m S S + K m I S is concentration of substrate, v s is reaction rate at S v m is maximum reaction rate, K m is enzyme affinity= S at which v s = v m / 2 I Function is mathematically equivalent to: I LineweaverBurke: 1 v s = 1 v m + K m v m 1 S Linear regression of Y = 1 / v s on X = 1 / S I HanesWoolf: S v s = K m v m + 1 v m S Linear regression of Y = S / v s on X = S I Both are linear regressions c 2011 Dept. Statistics (Iowa State University) Stat 511 section 25 3 / 40 ● ● ● ● ● ● ● ● ● ● ● ● 20 40 60 80 100 0.5 1.0 1.5 2.0 Substrate conc Reaction velocity c 2011 Dept. Statistics (Iowa State University) Stat 511 section 25 4 / 40 I However, the estimators of v m and K m derived from each model are not the same I Illustrate numerically: LS estimates from each model Model ˆ β ˆ β 1 v m ˆ v m K m ˆ K m nonlin 2.05 9.12 LB 0.377 5.64 1 /β 2.65 β 1 /β 14.96 HW: 4.74 0.482 1 /β 1 2.07 β /β 1 9.83 I Why? I Because the statistical model adds a specification of variability to the mathematical model, e.g. v i = v m S i S i + K m + ε i , ε i ∼ ( ,σ 2 ) c 2011 Dept. Statistics (Iowa State University) Stat 511 section 25 5 / 40 I And v i = v m S i S i + K m + ε i , ε i ∼ ( ,σ 2 1 ) (1) I is not the same as 1 v 1 = 1 v m + K m v m 1 S i + i , i ∼ ( ,σ 2 2 ) (2) I If you work out all the details, (2) is equivalent to (1) with unequal variances I The statistical models for MM, LB, and HW are different I Estimates differ because I Different variance models I Leverage of specific observations is not the same c 2011 Dept. Statistics (Iowa State University) Stat 511 section 25 6 / 40 linearizing a nonlinear model: 2nd example...
View
Full Document
 Spring '08
 Staff
 Regression Analysis, Variance, STATE UNIVERSITY, Nonlinear regression

Click to edit the document details