It has traditionally been assumed that responding after an error is slowed because participants try to improve their accuracy by increasing the amount of evidence required for subsequent decisions. However, recent work suggests a more varied picture of post-error effects, with instances of post-error speeding, and also decreases or no change in accuracy as well as post-error improvements. Further, the causal role of errors in these effects has been questioned due to confounds of slow fluctuations in attention caused by factors such as fatigue and boredom. We investigate both post-error speeding and slowing associated with instructions emphasising the speed versus accuracy of responding. In order to identify the causes of post-error effects we fit this data with evidence accumulation models using a method of measuring post-error effects that is robust to confounds from slow fluctuations. When the response-to-stimulus interval between trials was short there were no post-error effect on accuracy, and speeding and slowing were caused by differences in non-decision time (i.e., the time to encode choice stimuli and generate responses). In contrast, when the interval was longer, due to participants providing a confidence rating in their choice, there were also effects on the rate of evidence accumulation and the amount of evidence required for a decision. We discuss the implications of our methods and results for post-error effect research.