3
$\begingroup$

This is the rather confusing go-to internet definition for robust data: Robust data is data that is constructed to survive and function in multiple settings. It's reusable. It can be updated.

Anyone want to take a stab at elaborating on this in plain English? I can't think of any example data that wouldn't be reusable or updatable by simply adding additional observations to it.

The crux of the matter is that I have a large dataset with some extreme outliers, and I therefore trust the median much more than the mean as the 'center' of my data. Do I have robust or non-robust data, (or does mean vs median have nothing to do with the concept)?

Improve this question
$\endgroup$
5
  • $\begingroup$ can you cite the source? $\endgroup$
    – user603
    Commented Nov 30, 2014 at 1:04
  • $\begingroup$ xark.typepad.com/my_weblog/robust-data.html: not the most official of sources, but it seems to be the most prolific one out there, and has been copied to numerous other forums. $\endgroup$
    – bubbalouie
    Commented Nov 30, 2014 at 1:11
  • 1
    $\begingroup$ The sense of "robust" in this source is very different than the sense usually used in the statistical literature! The last paragraph of this question uses "robust" in the statistical sense. $\endgroup$
    – whuber ♦
    Commented Nov 30, 2014 at 1:51
  • 2
    $\begingroup$ "Robust data" isn't a standard term in statistics and the link is clearly not using the word "robust" in the statistical sense. $\endgroup$
    – Glen_b
    Commented Nov 30, 2014 at 6:04
  • $\begingroup$ Robust is usually a word that applies to an algorithm's ability to return correct and useful output (or at least mitigate damages and fail gracefully) in the face of hostile input such as missing values, evolving distributions of feature inputs, increasing/decreasing dimensionality and to some extent misinformation and environments in which it couldn't have been expected to succeed. The opposite of robustness might be brittleness, like a pane of glass that shatters when struck. The holy grail of machine learning is a totally Robust algorithm that learns under any possible circumstance. $\endgroup$
    – Eric Leschinski
    Commented May 2, 2017 at 15:20

1 Answer 1

Reset to default
1
$\begingroup$

Robust data in the sense your source uses it is not a statistical concept, and have nothing to do with the way robustness is used in statistics. Robustness in statistics is used to qualify methods, estimators, ... not data.

Your data is what it is, but your data, as you describe it with many outliers, might be in need of robust methods.

Improve this answer
$\endgroup$

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service and acknowledge you have read our privacy policy.

Not the answer you're looking for? Browse other questions tagged or ask your own question.