Improving Reasoning Performance in Large Language Models via Representation Engineering