All subjects succeeded in retrieving the food on more Tertiary-2 than Tertiary trials (Figure 4).
In addition, Pierre made significantly more positive swaps than expected in Tertiary trials, and Uék made significantly more positive swaps in all experimental conditions.
In addition, after these blocks of interspersed trials all subjects received an extra set of Tertiary trials (termed 'Tertiary-2'), bringing the total number of these trials to at least 23.
Corbeau used tools that were significantly longer than expected only in his second block of Tertiary trials, which is consistent with this subject starting with Strategy 1 (see above) and then developing selectivity by learning (in Tertiary-2).
Furthermore, the extracted tools first used by Pierre and Corbeau to probe for food were significantly longer in Tertiary-2 compared to Tertiary trials (Mann-Whitney U-test; Corbeau: W = 35.0, p = 0.009; Pierre W == 72.5, p = 0.009).
In Tertiary trials, the longest tool was only reachable with the second longest out-of-reach tool, instead of all other out-of-reach tools as in Experiment 1.
The other two subjects successfully retrieved food under both two-tool conditions, but, although they often extracted more than one tool on Tertiary trials, they never used them in the correct sequence to acquire food.
Food depth was used, rather than 'condition', because we wanted to know whether crows changed their behaviour when they perceived food to be out of reach: in this respect, there is no distinction between Secondary-Long and Tertiary trials.
In contrast, in conditions where the distance to food was greater, they first probed for tools more frequently, with no subjects first probing for food in the final block of Tertiary trials.
Pierre showed tool choice that was most consistent with Strategy 3: tool length did not differ from random expectation in Secondary-Any trials, but he used significantly longer tools in Secondary-Long and Tertiary-2 conditions (tools were also longer than the random expectation in Tertiary trials, but this was not significant at the corrected alpha level; Figure 5).
