Understanding the Hypercorrection Effect: Why High-Confidence Errors are More Likely to be Corrected