What is the Cramer-Rao Lower Bound?
The Cramer-Rao Lower Bound (CRLB) gives a lower estimate for the variance of an unbiased estimator. Estimators that are close to the CLRB are more unbiased (i.e. more preferable to use) than estimators further away.
The Cramer-Rao Lower bound is theoretical; Sometimes a perfectly unbiased estimator (i.e. one that meets the CRLB) doesn’t exist. Additionally, the CRLB is difficult to calculate unless you have a very simple scenario. Easier, general, alternatives for finding the best estimator do exist. You may want to consider running a more practical alternative for point estimation, like the Method of Moments.
The CLRB can be used for a variety of reasons, including:
- Creating a benchmark for a best possible measure — against which all other estimators are measured. If you have several estimators to choose from, this can be very useful.
- Feasibility studies to find out if it’s possible to meet specifications (e.g. sensor usefulness).
- Can occasionally provide form for MVUE.
There are a couple of different ways you can calculate the CRLB. The most common form, which uses Fisher information is:
Let X1, X2,…Xn be a random sample with pdf f (x,Θ). If is an unbiased estimator for Θ, then:
Where:
Is the Fisher Information.
You can find examples of hand calculations here.