In successive cancellation list (SCL) decoding, the tree pruning operation retains the L best paths with respect to metric at every decoding step. However, the correct path might be among the L worst paths due to imposed penalties. In this case, the correct path is pruned and the decoding process fails. In this work, we propose a scheme for additional decoding attempts when decoding fails, in which the pruning window does not necessarily select the L best paths, but this window is shifted between positions 1 and 2L in the sorted list. In the simplest form, the L worst paths are selected at the decoding step where the probability of elimination of the correct paths is high. Additionally, we generalize the scheme and propose a number of variants such as constrained shifting, nested shifting and shifting under segmented decoding, aiming to reduce the computational complexity. The numerical results for polar codes of length 512 with code rates 0.5 and 0.8 and list sizes L=2, 8, 32, show that the shifted-pruning scheme can provide 0.25-0.5 dB gain in error correction performance, while the average computational complexity approaches the conventional list decoding complexity at practical FER ranges.