5
$\begingroup$

This is a follow-up to this question: Do M-estimators and L-estimators overlap?. In particular, the answers to that question suggest that there are L-estimators which are not M-estimators, but do not give such an example.

More concretely, suppose I have an L-estimator $\ell(x_1,...,x_n)$ which is a linear combination of order-statistics of $(x_1,...,x_n)$. Can I always write

$\ell(x_1,...,x_n) = \operatorname{argmin}_{\theta} \sum_{i=1}^n \rho(x_i,\theta)$ for some function $\rho$?

$\endgroup$
5
$\begingroup$

A classic example would be a trimmed mean.

For concreteness consider a 25% trimmed mean, where we average the middle half of the data.

That's an L-estimator, but not an M-estimator. It can in a sense be approximated* by a Huber-type M-estimator but they're not the same.

* (perhaps 'analogy' would be a better term than 'approximation' -- they might not always be very close together -- if the distribution is quite skew for example. In symmetric cases they're very alike.)

While many L-estimators can also be M-estimators, there are a great many that aren't.

$\endgroup$

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.