Previous: How does EI relate

Up: Frequently Asked Questions

Next: How does EI relate

In what precise statistical sense is EI ``Robust''?

As aggregation bias increases, Goodman's regression becomes biased without limit. The bias in EI, in contrast, has a maximum value that is a function of the bounds. This is easy to see in Figure 9.6, p.180; note how the mean absolute error for the dotted line at the top, for example, is linear in the low aggregation bias region (where $\alpha<0.5$), but for higher levels of aggregation bias it levels off; in contrast, the mean absolute error for Goodman's and any method that does not incorporate the bounds will increase linearly without limit until the correlation of $X$ and $\beta_i^b,\beta_i^w$ is 1.