A total of 38 parrots participated in this study: Eight African grey parrots (six females, two males, all 1 year old), eight blue-headed macaws (five females, three males, all 1 year old), 13 blue-throated macaws (one female, twelve males, mean age 2.46, SD = 1.76), and nine great green macaws (eight females, one male, mean age 2.33, SD = 2.69). All parrots were hand-raised and subsequently socialized in parrot groups in the Loro Parque Fundacíon, Tenerife, Spain.
All parrots were housed in aviaries at the Max-Planck Comparative Cognition Research Station in the Loro Parque in Puerto de la Cruz, Tenerife. The blue-throated macaws and the great green macaws were housed in eight aviaries, divided by species and age into five groups of two to eight individuals.
Six of these aviaries were 1.80 × 3.40 × 3 m (width × length × height), and the remaining aviaries were 2 × 3.40 × 3 m and 1.5 × 3.40 × 3 m, respectively. These aviaries were interconnected by 1 m × 1 m windows, which could be closed when desired. The blue-headed macaws were housed together in a separate indoor area (28.61 m2) with access to a smaller outdoor area and the African grey parrots were housed together in another separate outdoor aviary (21.41 m2). All aviaries had at least one side open to the outside, so they followed a natural light schedule and were also kept to ambient outdoor temperature, but they were additionally lit with Arcadia Zoo Bars (Arcadia 54W Freshwater Pro and Arcadia 54W D3 Reptile lamp) to ensure sufficient exposure to UV light. They were also all within the same building as the testing chambers (described below).
Experimental setup and procedures
Training and testing took place in an indoor chamber of 1.5 × 1.5 × 1.5 m (height × width × length) equipped with lamps covering the birds’ full range of visible light (Arcadia 39 W Freshwater Pro and Arcadia 39 W D3 Reptile lamp). The birds were already habituated to moving from aviaries to the testing chambers. The subjects were individually tested in one of the testing chambers with the experimenter in an adjoining room. A sound-buffered one-way glass system permitted zoo visitors to see inside the rooms, but did not allow the birds to see out. All training and testing sessions took place either in the morning or in the afternoon, a minimum of 4 h after the last feeding (or overnight for morning sessions). All birds had free access to water and mineral blocks at all times and were fed fresh fruit and vegetables twice a day. Pieces of walnuts were used as rewards during testing as they are valued by all individuals and were not available outside of testing. The daily amount of nuts and seeds provided to the birds was adjusted according to their intake during testing for weight regulation purposes.
The apparatus consisted of an opaque and a transparent cylinder (Fig. 1). Both cylinders were open at both ends and attached to a wooden base. Following the criterion set by MacLean et al. (2014), the cylinders were long enough so that the birds had to put their heads inside the cylinder to reach the reward in the center, but were not so large they could enter the cylinder entirely. The size of the cylinders was adjusted to the size of each species tested (a length of 15 cm and a diameter of 11 cm for the great green macaws, a length of 12.5 cm and a diameter of 9 cm for the blue-throated macaws, and a length of 10 cm and a diameter of 5 cm for the African grey parrots and the blue-headed macaws). For each species, the size of the opaque and transparent cylinder was identical. Prior to this study, the birds had participated in an extensive physical cognition test battery following the protocol of Herrmann et al. (2007), in which they interacted with humans through holes in a Plexiglas panel on a daily basis for 2 months, and consequently all subjects had experience with transparent surfaces. Although none of the tasks required an active contact with the Plexiglass panel, the birds did explore the panel by touching it with their beak and/or tongue in the course of testing.
In the training phase, the birds learned to retrieve a reward from either side opening of an opaque cylinder. Before each trial, the birds were given a signal (which they had been previously trained on) to wait on a perch at the back of the testing chamber. The experimenter then drew the bird’s attention to the reward (a piece of walnut) by holding it up at eye level and calling their name. The experimenter then placed the food inside the cylinder while the bird was observing, at which point the bird could approach the cylinder. Birds had 360° access to the cylinder. If the bird did not approach the cylinder within 2 min, the reward was removed from the cylinder, and the trial was repeated after a 30-s time-out interval. Hence, such invalid trials did not count toward the maximum of 10 trials per session. Correct responses were scored when the birds retrieved the food without touching the surface of the cylinder, whereas incorrect responses were scored when the birds made contact with the surface of the cylinder. The birds were allowed to retrieve the food after both correct and incorrect responses. When the parrots finished eating the reward, they were again given a signal to return to their perch, and the next trial commenced. To proceed to the testing stage, the birds had to fulfill a criterion of four out of five correct responses on consecutive trials following the criteria from MacLean et al. (2014) and Kabadayi et al. (2016). Birds were given a maximum of 10 trials per session. All birds reached criterion within three sessions, but the majority reached it on their first session.
The testing protocol remained identical to the training procedure, with the exception that the opaque cylinder was now replaced with a transparent cylinder. Ten trials were conducted for each individual, except for two African grey parrots, who participated in one and two trials less, respectively, due to experimenter error. To preclude loss of motivation, the ten trials were divided into two sessions of five trials each, carried out on subsequent days. As in training, a correct response was coded if the birds made a detour to either side of the cylinder and retrieved the food without touching the surface of the cylinder, whereas an incorrect response was coded if the bird made physical contact with the surface of the cylinder before retrieving the food (Online Resource 1). All trials continued until the subject retrieved the reward. The methodology described above followed the one of MacLean et al. (2014) and Kabadayi et al. (2016) for both the training and the testing phase.
For all trials (correct and incorrect), we also measured the duration of time necessary for the birds to obtain the reward from the onset of the trial (response times). The onset of the trial was defined as the moment when the bird crossed a certain boundary line marked on the ground. The change in response times across trials was used in previous detour studies to study learning processes (Lockman and Adams 2001; Wyrwicka 1959). Thus, we analyzed the change in response times across trials within species as well as the difference in the rate of this change between species. Because of the slight between-species differences in distance between the mark on the ground and the cylinder, we did not compare response times between species.
Two variables were analyzed: the number of correct responses (response accuracy) and the response times per individual per trial. To analyze response accuracy, we used a generalized linear mixed-effect regression analysis (GLMM) with trial number, species, and bird age (in years) as fixed effects. Individual birds were included in the models as random effects, and the trial effect was allowed to vary for each individual bird (random slopes). The outcome variable was binary, i.e., the response was either correct or incorrect. To analyze response times, we used a linear mixed effects regression analysis (LMM) with the same fixed and random effects as in the analysis of the response accuracy. The outcome variable was response times in seconds.
We followed the coding criterion from the previous studies (MacLean et al. 2014; Kabadayi et al. 2016; Vernouillet et al. 2016), where all touches of the surface of the cylinder counted as an error, regardless of the location of the touches. However, we observed differences within the errors as some touches did not appear to be directed toward the reward, but could have been the result of exploration or accident. We therefore also provide additional analyses of the patterns of failures in order to potentially differentiate between failures caused by motor self-regulation and those caused by other factors.
In this additional analysis, we coded whether the parrots first touched the cylinder toward or away from the reward. When coding this, the cylinder was divided (on the computer screen) into three equal cross-sections (left periphery, center, right periphery), one of which contained the reward. If the birds’ initial contact with the cylinder was within the same zone as the reward then it was coded as a reach “toward” the reward; otherwise, it was coded as “away” from the reward. The interobserver reliability when coding for the failure patterns was excellent: Cohen’s Kappa = 0.961 (n = 135, z = 11.2, p < 0.001). We then recalculated the scores for all trials, with the following coding criterion: A correct response was coded if the birds made a detour to either side of the cylinder and retrieved the reward without touching the cylinder “toward” the reward, whereas an incorrect response was coded if the bird touched the cylinder “toward” the reward before retrieving the food. We then reran our original model on response accuracy using these new scores. Additionally, we analyzed how many errors were coded as either a reach toward or away from the reward across trials. We analyzed this using a generalized mixed model regression analysis, with error type as the outcome variable (i.e., toward the reward or away from reward), and species, trial number, and age as fixed effects. As in the first analysis, individual birds were included as random effects, and the trial effect was allowed to vary for each individual bird. All statistical analyses were carried in R, version 3.1.3 (R core team 2015).