Preprocessing influential observati

Preprocessing influential observations has three components: identification, influence assessment, and accommodation. In econometric settings, identification and influence assessment are usually based on regression statistics. Accommodation, if there is any, is usually a choice between deleting data, which requires making assumptions about the DGP, or else implementing a suitably robust estimation procedure, with the potential to obscure abnormal, but possibly important, information.Time series data differ from cross-sectional data in that deleting observations leaves "holes" in the time base of the sample. Standard methods for imputing replacement values, such as smoothing, violate the CLM assumption of strict exogeneity. If time series data exhibit serial correlation, as they often do in economic settings, deleting observations will alter estimated autocorrelations. The ability to diagnose departures from model specification, through residual analysis, is compromised. As a result, the modeling process must cycle between diagnostics and respecification until acceptable coefficient estimates produce an acceptable series of residuals.