Figure 1.
Experiment 1
(A) Experimental design. During the familiarization phase (four trials), the infant saw the experimenter hide a toy in one of two opaque boxes. The experimenter then asked her to point to indicate where she remembered the toy to be. As soon as the infant produced a pointing response, the selected box was pushed forward to allow retrieving the toy. The test phase (eight trials) was similar except for two aspects. First, a variable memorization delay (3, 6, 9, or 12 s) was introduced, during which a curtain was closed to occlude the boxes from the infant’s view. Second, unbeknownst to the child, the toys were now out of reach during the searching period (i.e., hidden within an unreachable pocket inside the box), so as to measure infants’ persistence. As soon as infants gave up searching, the experimenter recovered the toy from the box and showed it to the infant before either (1) replacing it in the box and letting the infant recover it after a correct response, or (2) starting a new trial when no response or an incorrect response was given by the infant.
(B) Mean type 2 ROC curve. Individual type 2 ROC curves were constructed by plotting the probability of searching for a certain amount of time for correct trials against the probability of searching for an equivalent amount of time for incorrect trials, cumulated across persistence time (PT) terciles. The gray area between the ROC curve and the diagonal is an estimate of the extent to which infants searched longer for correct versus incorrect trials: a type 2 ROC curve departing upward from the diagonal indicates that participants were more likely to search longer for correct over incorrect trials. The dashed line represents the minor diagonal; the solid thick line corresponds to the ROC curve; and the thin solid line corresponds to the major diagonal. Area under the ROC curve is shown in gray.
(C) Relationship between PTs and accuracy depending on task difficulty. PTs were averaged separately for correct and incorrect trials for each level of difficulty.
(D) Mean accuracy as a function of task difficulty, computed separately for short (<median) and long (>median) PTs. Although PTs did not significantly predict accuracy when all conditions were collapsed (χ2 = 2.54; p = 0.1), there was a significant interaction between PTs and delay (χ2 = 6.34; p < 0.02). In the 3-s condition, accuracy was lower for short PTs as compared to long PTs (t(25) = 2.85; p < 0.01). Importantly, the accuracy for short PTs did not differ from chance level (t(25) = 0.38; p > 0.7), whereas the accuracy for long PTs was greatly above chance (t(25) = 4.79; p < 0.001). Mean accuracy did not differ from chance level even when dividing PTs in deciles and considering only the tenth-shortest PTs (t(25) = 0.4; p > 0.7).
Error bars show SEMs. See also Figure S1.