A nonparametric model for unbalanced longitudinal data with application to geophysical data (Q1965987)

Consider a set of \(J\) individuals for which two quantities \(X\) and \(Y\) are measured. So we have a data set \((X_i^j,Y_i^j)\), \(i=1,\dots,n_j\), \(j=1,\dots,J\). If it is reasonable to consider a possible common structure, the authors propose to model the link between the random variables \(Y^j\) and \(X\) as \(Y^j=m_1(x)+m_2^j(x)+\varepsilon\) and use a two-stage procedure to estimate \(m_1(x)\) and \(m_2^j(x).\) The mean curve \(m_1(x)\) represents the common effect, while \(m_2^j(x)\) describes individual effects. \(m_1(x)\) is estimated in the first stage by a nonparametric regression performed on the whole data set. \(m_2^j(x)\) is estimated in the second stage by a nonparametric regression performed only over the \(n_j\) observations of the \(j\)th individual. A technique based on Watson-Nadaraya kernel estimators is introduced to deal with nonparametric regression models. Real geophysical data are examined as an example of application of the proposed model.