Announcement

Technical note:

Best means best in terms of something called quadratic loss (or squared-error loss). The name says what it is. Suppose there is a true value you wish to estimate and you are required to pay a penalty depending on how far off you are. In particular, suppose the penalty is

(estimated value - true value)2 ,
that is, the square of the difference between your estimated value and the true value. You do this repeatedly and are free to use whatever estimate you choose. For many of these problems, statistical theory shows that you'll lose the least amount of money if your estimate is the sample mean. For example, suppose you have to estimate a population mean. You might estimate it by using the first observation alone, or the mean of the largest and smallest observations, or by the median. However, statistical theory says if you use anything other than the sample mean, you'll pay more in the long run.