Another issue that has to be mentioned is exposure control. When maximum
Fisher information is applied for selecting the next item, only those items that
perform optimally with respect to this criterion will be selected. Typically,
20 percent of the items in the bank are selected for administration, while 80
percent are not selected at all. The same pattern can be found when any of
the other selection rules that maximize some kind of information criterion
is applied. The best items in the pool are over exposed, while the other items
are hardly exposed during test administration. Van der Linden and Veldkamp
(2007) studied this phenomenon more into detail and they found that for an
operational item bank, only a few items will be maximally informative over
the whole ability range. We repeated their analyses for an operational bank
of 499 items. The items bank was part of an intelligence testing battery. To
calibrate these items, they had been administered to a pre-testing sample of
3000 candidates. These candidates represented the Dutch labor force. Each of
the candidates answered a subset of items in the bank. Bilog MG (ZIMOWSKI
et al.,1996) had been applied to calibrate the items with the 2PLM. The curves
of the most informative and second most informative items for every ability
level are shown in Figure 1.