Simultaneous Error Bars
In Section 5.1.1 we mentioned a basic technique to compute asymptotic confidence intervals for m(x). However, this technique ignores the bias of the Nadaraya-Watson estimate and employs plug-in estimation of f(x) and σ2(x). In this section we present a bootstrap technique which does not need explicit estimation of functional of f, σ2, or m. The bootstrap is a resampling technique that prescribes taking “bootstrap samples» using the same random mechanism that generated the data. A bootstrap procedure automatically incorporating the bias term is the so called golden section bootstrap, introduced by Härdle and Mammen (1988).
KeywordsConfidence Limit Bootstrap Sample Coverage Probability Simultaneous Confidence Interval Bootstrap Residual
Unable to display preview. Download preview PDF.