Abstract: The presence of pictures of eyes reduces antisocial behaviour in humans. It has been suggested that this ‘watching-eye’ effect is the result of a uniquely human sensitivity to reputation-management cues. However, an alternative explanation is that humans are less likely to carry out risky behaviour in general when they feel like they are being watched. This risk-aversion hypothesis predicts that other animals should also show the watching-eye effect because many animals behave more cautiously when being observed. Dogs are an ideal species to test between these hypotheses because they behave in a risk-averse manner when being watched and attend specifically to eyes when assessing humans’ attentional states. Here, we examined if dogs were slower to steal food in the presence of pictures of eyes compared to flowers. Dogs showed no difference in the latency to steal food between the two conditions. This finding shows that dogs are not sensitive to watching-eyes and is not consistent with a risk-aversion hypothesis for the watching-eye effect.
Analysis
The latency to approach the food was recorded in both the ‘Go’ and ‘Leave’ trials. Latency was timed from the point that the owner gave the command until the dog had eaten the food. An additional coder, blind to condition, coded the approach latency for 40% of the sample. The high intra-class correlation (ICC = 0.99) indicates excellent levels of agreement between coders. To analyse the data, we constructed several mixed-effects Bayesian ANOVA models. The factors included in these models were Trial Type (Leave vs Go), Condition (Eye vs Flower), and a Trial Type*Condition interaction. Due to the repeated-measures aspect of the design (all dogs took part in both a ‘Go’ and ‘Leave’ trial), participant was included as a random effect in all models. Each model was compared to a null model, which only contained participant as a random effect. Additionally, an analysis of effects was carried out to determine the inclusion BF for each individual factor. Inclusion BFs are calculated by comparing the fit of models containing the factor against the fit of models not containing that factor. BFincl > 3 indicate that including a factor substantially increases model fit while BFincl < 0.333 indicates a factor substantially decreases model fit. Each model was constructed with objective priors of prior width (r) = 1 for fixed effects and r = 0.5 for random effects.As the extent to which humans attended to images of eyes appeared to affect their likelihood of showing the watching-eye effect21, we re-ran this analysis but included the proportion of time that the dogs looked at the picture as a covariate for each model. Each model was compared to a null model which contained participant as a random effect and proportion of time looking at the picture as a covariate. Again, models were constructed with objective priors of r = 1 for fixed effects and r = 0.5 for random effects.
Additionally, in order to specifically get at our comparison of interest, we compared the ‘Leave’ latency in both conditions after adjusting for differences in individual dogs’ approach speed. This adjustment was made by subtracting the ‘Go’ latency from the ‘Leave’. If the dogs display the watching-eye effect, we would predict that the adjusted latency would be higher in the eyes condition than in the flowers condition. Comparisons between the adjusted ‘Leave’ latencies were analysed using a Bayesian independent-samples t-test. The prior distribution for the alternative hypothesis was a Cauchy half-distribution, centred on an effect size of 0, with r = 0.707. All analyses were carried out using JASP 0.10.0.0 (JASP team, 2019.) This study design was pre-registered (http://aspredicted.org/blind.php?x=j6er8v). It should be noted using the Go trial as a baseline to adjust the dogs’ Leave latencies meant it was necessary to have the owners give the ‘Go’ command on the same trial. Whilst this means that it is impossible to fully disentangle the effect of the command on the dogs’ latency to approach food from order effect, we concluded that the extreme implausibility that dogs would approach food slower on a 2nd trial after being able to take it without punishment in the previous trials made this a worthwhile trade-off.
Posts
No comments:
Post a Comment