Even when using a provably secure voting protocol, an election authority cannot argue convincingly that no attack that changed the election outcome has occurred, unless the voters are able to use the voting protocol correctly. We describe one statistical method that, if the assumptions underlying the protocol’s security proof hold, could provide convincing evidence that no attack occurred for the Norwegian Internet voting protocol (or other similar voting protocols). To determine the statistical power of this method, we need to estimate the rate at which voters detect possible attacks against the voting protocol. We designed and carried out an experiment to estimate this rate. We describe the experiment and results in full. Based on the results, we estimate upper and lower bounds for the detection rate. We also discuss some limitations of the practical experiment.