Awhile back, I came across these two videos:
The videos depict a wild female* raccoon ('Fiver'), being trained to give 'high fives' in exchange for dog kibble. In the first night, in response to Fiver spontaneously tapping the woman's hand while getting kibble, she decides to start telling Fiver to 'give me five' and rewarding pats to the hand with kibble. Fiver's variable number of pats and seeming confusion and hesitation lead the woman to wonder if Fiver may have been expecting more pats to lead to more food. On the second night, therefore, the woman starts giving Fiver the same number of kibble pieces as Fiver's pats, to see if Fiver learns the association.
I decided to tabulate the results of these two videos and run some statistical analyses on Fiver's behavior.
If Fiver does know how to count, and can tell that more pats leads to more kibble, what should we expect?
Assuming she wants to maximize her kibble reward, Fiver should react to the link between her pats and the kibble she gets by increasing the number of pats over time. Therefore, she should provide more pats per trial on the second night than on the first night. In addition, over the second night, she should increase her number of pats as she learns the contingency.
I coded Fiver's pats based on carefully watching the video, as well as listening to the woman's comments. I also coded the amount of kibble Fiver received. In addition, I recorded whether the woman commented on the number of pats Fiver gave her. Each exchange of pats & treats was considered a single trial.
On two occasions on night 2, Fiver gets interrupted by another raccoon coming near (the first time, the other raccoon actually displaces Fiver and gets some kibble). Since Fiver seemed to take a while to get back into the flow of things after these interruptions, I decided to consider each interruption as the end of a 'session'. Therefore, on the two nights, Fiver had four sessions - one session of 10 trials on night one and three sessions of 25, 58 and 24 trials on night 2 - for a total of 117 trials.
I analyzed correlations between pats, trial number and kibble received for each session individually, and also compared the same variables across nights and sessions.
Over both nights, Fiver gave a single pat 51% of the time, two pats 28% of the time, and 3-7 pats on the remaining trials. The trial with 7 pats (on the first night) appears to be an outlier, with the next highest number of pats in one trial being 5. She received 1 kibble in 59% of trials, two kibbles in 25% of trials, and 3-5 kibbles in the remaining trials.
Apart from the 7-pat trial, when she received 4 kibbles, all other trials on the first night resulted in 1 kibble per trial, despite her giving 1-4 pats (1 pat in 2 trials, 2 pats in 3 trials, 3 pats in 1 trial and 4 pats in 3 trials). As a result, once the outlier was excluded, there was no possible correlation between number of pats and number of kibbles. In addition, the woman never commented on the number of pats Fiver gave her. (Though she commented on the quality of the pats at times.) There was no correlation between trial number and number of pats Fiver gave, suggesting that Fiver did not show a trend towards increasing or decreasing pats over time on this night.
On the second night, Fiver received 1-5 kibbles for 1-5 pats, with a correlation of .965 between pats and kibbles. This correlation did not vary appreciably between sessions (.933-.991). Despite this, Fiver showed a negative correlation between trial number and number of pats (-.256), suggesting that she gave fewer pats as the night went on - the opposite of the predicted result. Within session, the negative correlation was stronger in session 4 (-.442) and nonsignificant in sessions 2 and 3 (.030 and .184). The woman gave frequent verbal feedback on trials, saying a single number in 45% of trials (92% of which were 1-pat trials, and the rest were 2-pat trials) and counting aloud in 38% of trials. In all cases where the woman commented on the number of pats, she gave the same number of kibble pieces as her comment indicated.
A T-Test comparing the two nights revealed that Fiver showed significantly greater variance in number of pats on the first night than on the second night (SD 1.83 vs .94). In addition, contrary to the hypothesis that Fiver would produce more pats on the second night, there was an almost significant difference in the opposite direction (p = .053), with Fiver producing an average of 3 pats per trial on the first night and 1.71 pats on the second night.
An ANOVA revealed a significant difference between sessions (P <.001). Post-hoc tests revealed that Fiver produced significantly more pats in session 1 (on the first night) than in sessions 3 (p = .003) and 4 (p <.001). In addition, Fiver produced significantly more pats per trial in session 2 (at the start of the second night) than in session 4 (p = .022). All other comparisons were non-significant.
Overall, the results found were the opposite of what the hypothesis predicted. Rather than increasing her number of pats per trial, Fiver actually decreased
the number of pats she produced over the second night.
These results could indicate that Fiver was unable to perceive a link between the number of pats and the number of kibbles she received. Alternately, Fiver may not have wanted to receive multiple kibbles at one time. The presence of rival raccoons may have made her wary of potential theft, leading to a strategy of requesting only as many kibbles as she could eat in one bite. In addition, since Fiver performed 107 trials on the second night, the decline in number of pats over time could be attributed either to fatigue or satiation. Certainly, Fiver seems to be less motivated as the night goes on, although part of this is undoubtedly because she's watching out for her rivals.
With the current data, we can't determine whether Fiver's behavior represented a deliberate strategy or a simple lack of understanding.
Anecdotally, Fiver does seem to respond more promptly and with less coaxing in the second night than in the first. However, this is not necessarily due to the change in contingency, since there are several other factors that might influence her behavior.
Firstly, from what I can gather, the session on the first night started after Fiver had already been given a lot of kibble noncontingently. The sudden shift between noncontingent to contingent reward may have confused and/or frustrated Fiver. In contrast, on the second night, the reward was contingent to begin with, and Fiver had already had several trials of contingent reward the night before.
Secondly, Fiver is a wild raccoon, and the woman who posted the videos describes her as 'new' in the first video. This suggests that Fiver was unfamiliar with the woman, and may have been a bit afraid of her. (Fiver often backed away between trials, supporting the fear interpretation.) Once again, on the second night, Fiver knows better what to expect - this woman will feed her kibble and not try to hurt her. Wild animals are often neophobic, especially towards humans. Though city-dwelling animals often have a lower flight distance than rural animals, coming close enough to eat from her hand may have still been a bit outside of Fiver's comfort zone.
Thirdly, miscommunication may have played a part. On three occasions on the first night, immediately after accepting the reward, Fiver did a very fast, very light pat, which the woman refused to reward her for. After each of those quick pats, Fiver paused and acted as if she was waiting to be rewarded, and seemed to get confused when the reward didn't come. On the second night, I saw only one such unrewarded pat, which occurred early on in session 2. It's unclear whether Fiver learnt to give better pats or the woman became more consistent in rewarding all of her pats, but the proportion of unrewarded pats was certainly far lower on the second night.
* The woman mistakenly refers to Fiver as male in the video, but Fiver has since had pups, making it clear that she's actually female.