Of the many techniques proposed for mitigating errors on quantum computers, error detection is one of the simplest protocols. While error detection has been used in several small-scale experiments, the overhead and scaling on real quantum computers is understudied. In this work we provide a clear exposition of error detection which unifies multiple formulations in literature, and we present results from error detection experiments on real and simulated noisy quantum computers. Our experimental results show that the accuracy of expectation values calculated via error detection can improve exponentially in the number of physical qubits, but reveals the important practical considerations that must be met in order to achieve this, including overhead from compilation, fault-tolerant gates, and sample complexity.