Reinforcement bias performance in the midsession reversal task