Proceedings of the Annual Meeting of the Cognitive Science Society

The Give-a-Number task has become a gold standard ofchildren’s number word comprehension and has beenincreasingly used to organize debate in developmentalpsychology. In this task, the experimenter asks children togive specific numbers of objects (e.g., 1 to 6), and based ontheir pattern of responses, children are classified into stagesthat can be readily related to other developmental milestones.The increasing popularity of Give-a-Number raises thequestion of how reliable it is, since the size of a correlationbetween two different tasks cannot reliably exceed the test-retest reliability of either measure taken individually. InExperiment 1, 2- to 4-year-old children were tested twice in asingle session with Wynn’s (1992) version of the Give-a-Number task, which features a titrated design. In Experiment2, we tested a second group of children with an alternativeversion that uses a larger number of trials in a non-titrateddesign. We found that in both cases the task was highlyreliable in differentiating children who could accurately countfrom those who could not, but that reliability differed forspecific numbers, and was more reliable for very smallnumbers (i.e., “one” and “two”) than for slightly larger ones(i.e., “three” and “four”). We discuss practical implications ofthese results for researchers studying numeracy and discussfurther directions to assess the validity of the task.