Convergence rates for partially splined models (Q1093290): Difference between revisions

Let \(y_ i=\alpha u_ i+g(t_ i)+\epsilon_ i\), \(i=1,...,n\), be a partially spline model, where \(\alpha\) is an unknown regression coefficient, g is a smooth function to be estimated and the \(\epsilon_ i\) are independent random variables with mean 0 and variance \(\sigma^ 2\). As the estimates of \(\alpha\) and g we consider the minimizers of \[ n^{-1}\sum^{n}_{i=1}(y_ i-{\hat \alpha}u_ i-f(t_ i))^ 2+\lambda \int | f^{(m)}(t)|^ 2dt \] where f denotes a natural spline of order m on [0,1], with knots at the points \(t_ i\), \(i=1,...,n\), and the parameter \(\lambda\) belances smoothness of f with fidelity to the data. In this paper, the author obtained concrete expressions of \({\hat \alpha}\) and examined the convergence rates of the variance and bias. The results show that to force the bias to be negligible with respect to the standard error, g has to satisfy some stringent conditions on smoothness.