0
$\begingroup$

To sum up the question: is there some recommended set of datasets or should I put the list together myself from articles on robust estimators?

If I wanted to test performance of different robust estimators, what are the various datasets that are frequently used? I am interested in both simulated and real datasets that I could use. It seems impractical to have to simulate various scenarios on my own and also it might not be very useful to pick real datasets at random.

I am interested in a set of datasets which would allow me to empirically find scenarios under which various robust estimators (LTS, LWS, WLS, S estimators, S-weighted estimators) break down.

$\endgroup$
0
$\begingroup$

I think if you want to compare performances of different estimators, you should simulate your datasets. This way, you are in control of the settings, and you can create harder/easier scenarios to compare the performance of the different estimators. Furthermore, you'll know what a good estimate should look like, since you simulated the data and chose the true parameter value.

There a few ways you can simulate datasets with outliers and extreme observations. 1) Simulate from a normal distribution with large variance. 2) Simulate from a t-distribution with low degrees of freedom. 3) Simulate from any distribution under "normal" settings and artificially add in extreme observations. You can also look into other heavy-tailed distributions such as Laplace distribution (also known as double exponential). You can always plot a histogram of the data to get a quick visual on how spread out the data is and how extreme are the outliers.

$\endgroup$

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

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