Message-ID: <24328.24844.630465.11132@stat.math.ethz.ch>
Date: 2020-07-10T12:37:32Z
From: Martin Maechler
Subject: [RsR] A question about robust estimation of a mean from multiple instruments
In-Reply-To: <CA++P4r1tLhQjWJXEBCa3r982VZ3p6za_mfeP5aurmdOfyx1R+Q@mail.gmail.com>
>>>>> Ajay Shah
>>>>> on Fri, 10 Jul 2020 16:37:47 +0530 writes:
> I have a question about a robust location estimator when faced with the
> following situation:
> There are n instruments, and each instrument has a different measurement
> standard error.
> I can just treat the data x as i.i.d and our standard concepts of lmRob()
> would do a lot.
> But in truth, in this problem, we know more. We have a standard error
> estimate of each instrument. We should be able to utilise this, e.g. when
> we see a weird value from an untrusted instrument we should be more willing
> to disregard it.
> How would one go about this? I would greatly appreciate pointers. I am
> happy to place sample 25-observation datasets here if it's fun.
> --
> Ajay Shah
> ajayshah at mayin.org
> http://www.mayin.org/ajayshah
Dear Ajay,
this is good question (even better would be *not* to use
econmetric language but stay with probability and statistics).
Yes, do provide the dput() (or other simple reproducible R code to *create*)
a sample data set, so we can use reproducible code here to
explore possible answers.
Best regards,
Martin
--
Martin Maechler
ETH Zurich and R Core Team