Re: [R] Siegel nonparametric regression / mblm package

Thanks a lot! Il giorno lun 11 feb 2019 alle ore 14:39 Roger Koenker < rkoen...@illinois.edu> ha scritto: > A quick look at the code for Siegel in mblm reveals that it is extremely > inefficient, but it seems to be correct. > One “explanation” for this behavior, presuming that we haven’t overloo

Re: [R] Siegel nonparametric regression / mblm package

A quick look at the code for Siegel in mblm reveals that it is extremely inefficient, but it seems to be correct. One “explanation” for this behavior, presuming that we haven’t overlooked something more basic, is that such high breakdown estimates sacrifice some efficiency, that is to say, they a

Re: [R] Siegel nonparametric regression / mblm package

My first thought was also that this was an artifact of the ties, but dithering the data n <- length(child) child <- child + runif(n,-.5,.5) parent <- parent + runif(n,-.5,.5) and rerunning yields the same discrepancy between the Siegel and other fits. Curiously, both lmsreg and ltsreg from MASS