[R] inquire a statistical terms

S Ellison S.Ellison at LGCGroup.com
Thu Aug 9 03:01:12 CEST 2012


>Is there any standard statistical terminology describing the points beyond
> a confidence region? 

'data'?

Sounds flippant, but even for univariate location estimates (a case where parameter estimates and data can be compared directly)  a lot of data points are usually outside confidence regions, because confidence regions for parameters (or response) are usually a good deal smaller than the dispersion of data used to estimate them. Generally, though, parameters and their confidence regions aren't necessarily in the same dimensions as the data, so it's not meaningful to ask whether data are inside or outside the parameter confidence region. (As a simple example, the gradient from simple linear regression has the units of response/predictor, which is not the same units as the response or predictor making up the data). 

If you mean observations outside a prediction interval for future observations then 'outlier' (with suitable qualification as to the criterion for identifying points as such) is probably as good a description as any because the points are outside the region in which your model predicts as likely. 

*******************************************************************
This email and any attachments are confidential. Any use...{{dropped:8}}



More information about the R-help mailing list