[R] Heteroscedastic data due to responses at one factor level being too low to measure and assumed to be 0?

Bert Gunter bgunter.4567 at gmail.com
Tue Mar 1 17:05:06 CET 2016

This is a list about the R (statistical) programming language, not a
statistical advice list. That would be (among others)
stats.stackexchange.com, to which you can try posting.

However, recommendation: As you stated, you are well out of your
statistical depth here, and I would strongly advise that you find a
local statistical expert with whom to consult.  For example, this is
an example of (left) censored data, and as you noted, changing all
those lower than detectable limits to 0's may be problematic. There
are packages and functions to deal with this, but you have to know
what you're doing. You don't. So find help.

Personal gratuitous aside (feel free to ignore):  In my over 40 years
of consulting, I have encountered many, perhaps most, engineers, who
have such problems. So it is no dishonor. There just seems to be too
much other important stuff to fit into an engineering education for
there to be time for much statistics training. This comes back to bite
in practice, as you have found.


Bert Gunter

"The trouble with having an open mind is that people keep coming along
and sticking things into it."
-- Opus (aka Berkeley Breathed in his "Bloom County" comic strip )

On Tue, Mar 1, 2016 at 7:01 AM, Dr. Leigh S. Sutherland
<l.sutherland at tecnico.ulisboa.pt> wrote:
> Hello all,
> I have a 3-factor experimental design.
> The response is the load to break an adhesive joint.
> I have a question about Heteroscedastic data ('non-uniform variability',
> Google told me that's what it's called - I'm not a statistician, just an
> engineer)
> *In essence:*
>    - At one level of a factor the response values are so low to make them
>    un-testable and they have to be assumed to be 0
>    - However this gives no variability at that level since all responses
>    are 0, (although in reality they may well have similar variability as at
>    other levels).
>    - This gives heteroscedastic data across the levels of the factor
>    - What implications does this have for stats tests that are sensitive to
>    Heteroscedastic data?
>    - If I need to do anything, what can I do? can I 'superimpose' the same
>    variability on level 1 results to give artificial variable values 'around'
>    0kg? (this would not at all invalidate the data's validity)
> Thanks,
> Leigh
> --
> Leigh Sutherland
> Centre for Marine Technology and Ocean Engineering (CENTEC)
> Instituto Superior Técnico
> Av. Rovisco Pais
> 1049-001 LISBOA
> Tel: +351 218 417 947
>         [[alternative HTML version deleted]]
> ______________________________________________
> R-help at r-project.org mailing list -- To UNSUBSCRIBE and more, see
> https://stat.ethz.ch/mailman/listinfo/r-help
> PLEASE do read the posting guide http://www.R-project.org/posting-guide.html
> and provide commented, minimal, self-contained, reproducible code.

More information about the R-help mailing list