It is well known that size-inextensivity is a major drawback of truncated configuration interaction (CI) methods. In order to ease the problem, many attempts have been made to minimize size-extensivity error of the CI methods through a posteriori corrections. In this study, we investigate performance of various size-extensivity corrections proposed in the literature. While the corrections were basically formulated for the single-reference (SR) CI method with singles and doubles, very soon their straightforward generalizations to a multireference (MR) case started to be used without any deeper theoretical justification. The only correction that is universal enough to be applied in case of any kind of CI method is the coupled-cluster (CC) correction. For the purpose of this study, we have used the closed shell (a) over tilde (1)A(1) singlet excited states of the CH2 and NH2+ molecules. The states are of MR character but still can be quite well described by SR approaches. That allows us to investigate the performance of the corrections in both SR and MR cases. For the first time, the second-order configuration interaction method has been used as a kind of the MR-type CI method. The calculations show that the performance of the CC correction is significantly better than other corrections.