Testing the Information Matrix Equality with Robust Estimators

Figure 1: Asymmetric contamination: non-centraiity parameter

Note: right panel uses 104 as measurement unit on vertical axis

Figure 2: Symmetric contamination: non-centraiity parameter

Note: right panel uses 104 as measurement unit on vertical axis

We see in both figures that the non-centrality parameter corresponding to
the ML estimator is uniformly smaller than those corresponding to the other
estimators, as shown. The non-centrality parameter associated with the
MAD estimator is discontinuous at x = Φ^-1(3/4) = 0.6745, where also ρ_c(■ )
is discontinuous. Figure 3 gives the power curves of 5%-level IM tests with
level e contamination at x = 10, a clear outlier relative to the N (0, 1) distri-

13

<<   <   12   13   14   15   16   17   18   >   >>