2

Yesterday I was at a medical conference in which a lot of plots of Point Clouds with linear fits were shown. In many cases the fit seemed (at least to me and colleagues) to be influenced mostly by some outliers, while most of the points seemed to be pretty much random point clouds. I think there is a name for that fallcy but I can't remember. Can you help me? That would help me talk about the problem with the participants.

Also what do you typically recommend to deal with outliers?

kjetil b halvorsen
  • 63,378
  • 26
  • 142
  • 467
geo
  • 362
  • 3
  • 9
  • I think my page of animated curve fitting problems might be useful, it includes the effect of various outliers on the statistical confidence in models: http://zunzun.com/CommonProblems/ When fitting to the lowest sum of *squared* errors, outliers are far from the other data points so the *square* of the error becomes large and they dominate the regression. Fitting to the lowest sum of [absolute value of error] is one of the ways to ave a "robust" regression less sensitive to outliers, but this means that Gauss' linear algebra method cannot be used and non-linear fitting techniques are needed. – James Phillips Oct 14 '18 at 13:56
  • thanks, that is indeed helpful. I thought there was a specific name for these kind of plots, but I could have remembered wrong. – geo Oct 16 '18 at 11:20
  • As I understand the technical statistical term most commonly used in reference to these specific plot types, it is "totally awesome". – James Phillips Oct 16 '18 at 11:39
  • @JamesPhillips I think the medical term is "highly significant" – geo Oct 16 '18 at 17:49

0 Answers0