This paper is from Session 8J: Evidence-based statistical practice
Full topic list
which comes under Topic 8: Research in statistics education

(Monday 12th, 14:00-16:00)

The influence of presentation on the interpretation of inferential results




Confidence intervals (CIs) have frequently been presented as an alternative for null-hypothesis significance testing (NHST). Earlier, it was shown that the frequency of misinterpretations for results presented by means of CIs are lower than those for data presented by means of NHST outcomes. Little is known, however, about whether the subjective estimates that arguably play an implicit role in most interpretations of results differ for results presented by means of CIs or NHST outcomes. In the present study, participants were asked to interpret outcomes of fictitious studies. For significant outcomes, participants tend to be more certain about the existence of a population effect and about replicability of their results when the results are presented by means of NHST than by means of CIs. Such a difference could not be found for clearly non-significant findings. Apparently, a significant finding presented by means of a p-value is more convincing than the same effect presented by means of a CI.