6Outlier Detection in Nonlinear Regression
Previous chapters have discussed robust methods for fitting nonlinear regression models in null conditions, especially in the presence of outliers. Situations in which classical assumptions did not hold were also discussed. The methods presented reduce the impact of outlier data and lead to more modest estimates in the presence of null values. These estimators are applied to identify the outlier data in this chapter. Applying classical, nonrobust methods results in the false discovery of outliers. For example, Riazoshams et al. (2011) have shown some examples that mix least squares estimators with statistical measures, and which are not successful in identifying outliers.
Outlier detection is essential for practitioners because of the huge potential for misinterpretation in inference. High leverage points together with large errors (outliers) and residuals are responsible for masking and swamping of parameter estimates in linear regressions (see Habshah et al. (2009)). Moreover, masking and swamping breaches the assumptions of classical models. For example, Figure 3.1 shows how outliers not only have a masking and swamping effect on the model fit, but also misrepresent the model assumption properties, such as homogeneity and independence of errors. These facts mean that outlier detection methods need to be applied in cases where the classical assumptions are not also valid.
The outlier detection methods described in this chapter are based ...
Get Robust Nonlinear Regression now with the O’Reilly learning platform.
O’Reilly members experience books, live events, courses curated by job role, and more from O’Reilly and nearly 200 top publishers.