We have five Rasch items with b1 = −0.5, b2 = −0.3, b3 = 0.0, b4 = 0.25, and b5 = 0.5. We want to construct a two-item test that discriminates relatively well at θ1 = −0.5 and at θ2 = 0.5.

Which combination of two items from the item bank with five items is best, given the criterion that the largest of the error variances at θ1 and θ2 should be as small as possible?