Bootstrapping for Penalized Spline Regression
read more
Citations
Practical variable selection for generalized additive models
Resampling Methods for Dependent Data
Economic convergence: Policy implications from a heterogeneous agent model
Simultaneous selection of variables and smoothing parameters in structured additive regression models
The Geometry of Nutrient Space–Based Life-History Trade-Offs: Sex-Specific Effects of Macronutrient Intake on the Trade-Off between Encapsulation Ability and Reproductive Effort in Decorated Crickets
References
An introduction to the bootstrap
A practical guide to splines
Spline models for observational data
Generalized Additive Models
Related Papers (5)
Frequently Asked Questions (12)
Q2. How can the authors draw v from the empirical distribution function of the fitted values?
residual bootstrapping can be used by setting u∗ = D̃−1/2 v∗ and drawing v∗ from the empirical distribution function of the fitted values v̂ = D̃1/2û.
Q3. What is the main assumption for penalizing a smooth fit?
Once a basis is selected, a penalized fit is pursued by imposing a penalty on the spline coefficients u and estimating by least squares regression, which results in a ridge regression estimate.
Q4. What is the advantage of the mixed model approach?
One advantage of the mixed model approach, as also noted in Ruppert, Wand, and Carroll (2003, ch.6), is that the bias due to smoothing in the smoothing model becomes a component of variance by treating u as random.
Q5. How can the mixed model bootstrap be used for smoothing?
The authors have demonstrated how the link between penalized spline smoothing and linear mixed models can not only be exploited for smoothing but also for bootstrapping.
Q6. What is the linear unbiased predictor?
Under this model, the estimator µ̂λ can be interpreted as a posterior Bayes estimator or as best linear unbiased predictor (BLUP)5with λ = σ2ε/σ 2 u steering the amount of smoothness.
Q7. What is the smoothing parameter used for the mixed model bootstrap?
The smoothing parameter λP is chosen using REML, which provides an easy and numerically appealing choice (see also Ruppert, Wand, and Carroll, 2003, p.113).
Q8. How is the mixed model bootstrap performed?
The mixed model bootstrap of û is carried out in two ways, first by simply resampling u∗ from û and secondly by accounting for the correlation structure among û as proposed above.
Q9. What is the way to show that the bootstrap distribution is consistent?
It can be shown that the Mean Squared Error based choice of λ has order O(1) (see Kauermann, 2004), so that consistency follows naturally if the smoothing parameter is chosen in a data driven manner, for both pilot and bootstrap versions of λ.
Q10. What is the effect of bathrooms with special features?
The effect of bathrooms with special features is positive but less strong and shows some non-significant behavior based on the mixed model bootstrap.
Q11. What is the main assumption for penalized spline smoothing?
Following the discussion in Ruppert, Wand, and Carroll (2003, ch.6), the authors show here that the bias problem can be circumvented in penalized spline smoothing if a mixed model formulation is used for bootstrapping.
Q12. What is the main concern when bootstrapping smooth models?
A major concern when bootstrapping in smooth models is the bias occurring due to smoothing, which is not accounted for if one applies a naive bootstrap.