Inference in a linear functional relationship with replications (Q2244847)

This paper studies statistical inference for the error in variables functional linear regression model. Methods based on availability of replicated observations are proposed. Normality is assumed. A maximum likelihood method in the presence of incidental parameters and an unbiased estimating equations approach are developed. To obtain the ML estimators a simple EM type algorithm is implemented. The two approaches lead to explicit expressions for the asymptotic covariance matrices of the estimators of the model parameters. A Wald test is proposed to test linear hypotheses of interest. A scaled Mahalanobis distance is used to detect outliers. A diagnostic measure is proposed to analyze the sensitivity of the regression coefficients. Q-Q plots of the transformed distances are implemented. To assess the empirical behavior of the estimators and of the Wald statistic the simulation study is performed. The methodology is illustrated with an application to a real data set of women blood glucose levels.