I know that it is not desirable for behavioural data to show a (within-subjects) speed-accuracy trade-off, as it can lead to e.g. the wrong interpretation of a single subject's decreased RTs as meaning improved performance, when in fact if one looks at error rates, one can see the increased speed came at the cost of increased error rates.

However, what I don't understand is why, assuming that subjects are instructed to respond slowly enough so that their error rates approach zero (so not the usual speeded responses paradigm), why would it not be correct to then analyse the group RTs, which will probably show wide between-subjects variations and can thus offer meaningful interpretations? This way, it would not matter that all subjects will have traded speed for accuracy, as it is precisely the amount of "slowing" that becomes the dependent variable.

Hope the question makes sense, thanks!


You may be able to reduce errors by telling participants to react slowly. However, by adding this instruction, you may also dilute the response time differences that your are trying to find (whether between or within subjects). If you are lucky, you are just adding random noise, however it is more likely that you will add confounds or tap into different mental processes (occuring at a later time).

Another option to tackle this problem is to hold the speed more or less constant, by giving participants a very narrow time window to respond (which can also be tailored to their speed). As a result, they will produce more errors, which now become the dependent variable (see Draine & Greenwald, 1998, who have introduced this technique).


Draine, S. C., & Greenwald, A. G. (1998). Replicable unconscious semantic priming. Journal of Experimental Psychology. General, 127, 286–303.

  • $\begingroup$ Thanks, this makes sense. But why is it that instructions that emphasise accuracy at the expense of speed lead to more noise (or to other more complex confounds) in the dependent variable (RTs) as opposed to the instructions that emphasise speed at the expense of accuracy (the latter being the dep variable)? $\endgroup$ – z8080 Jun 21 '15 at 9:10
  • 1
    $\begingroup$ As an example, imagine a semantic priming task (if "doctor" is presented before "nurse", participants are faster in recognizing "nurse" as a word). If you tell participants to make 100% sure that they have the right answer, but that they can take any time, it may be less likely to find a priming effect. Participants may consciously double-check their answer, and in the meantime, the effect of the prime may be gone (participants may have thought of something else now/you are tapping into the double checking more than into priming). Essentially, the paradigm is not about reaction times any more. $\endgroup$ – user7759 Jun 22 '15 at 10:59
  • $\begingroup$ That's a useful comparison. Thanks again! $\endgroup$ – z8080 Jun 23 '15 at 11:00

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

Not the answer you're looking for? Browse other questions tagged or ask your own question.