Influential ObservationsInfluential observations arise in two fundamentally distinct ways. First, they may be the result of measurement or recording errors. In that case, they are just bad data, detrimental to model estimation. On the other hand, they may reflect the true distribution of the innovations process, exhibiting heteroscedasticity, skewness, or leptokurtosis for which the model fails to account. Such observations may contain abnormal sample information that is nevertheless essential to accurate model estimation. Determining the type of influential observation is difficult when looking at data alone. The best clues are often found in the data-model interactions that produce the residual series. We investigate these further in the example Time Series Regression VI: Residual Diagnostics.