QQ-plot of (squared) Mahalanobis distances vs. scaled F-distribution (or a scaled chisquare distribution). In addition two default cutpoints are proposed.

1 |

`dist` |
a vector of Mahalanobis distances |

`p` |
the number of variables involved in the Mahalanobis distances |

`alpha` |
a probability for cut-off, usually close to 1 |

`chisquare` |
A logical indicating the the chisquare distribution should be used instead of the F-distribution |

Scaling of the F-distribution as
*median(dist)*qf((1:n)/(n+1),p,n-p)/qf(0.5,p ,n-p)*. First default cutpoint is *median(dist)*qf(alpha,p,n-p)/qf(0.5,p,n-p)* and second default cutpoint is the alpha-quantile of the Mahalanobis distances.

`hmed` |
first proposed cutpoint based on F-distribution. |

`halpha` |
second proposed cutpoint (alpha-quantile) |

QQ-plot

Beat Hulliger

Little, R. & Smith, P. (1987) Editing and imputation for quantitative survey data Journal of the American Statistical Association, 82, 58-68

1 2 3 |

Questions? Problems? Suggestions? Tweet to @rdrrHQ or email at ian@mutexlabs.com.

All documentation is copyright its authors; we didn't write any of that.