2.2. Data Collection and Analysis

Each of the 10 studies represented an independent study to test to what extent visitors impact the behavior of zoo animals. All 15 researchers were trained and mentored by the same researcher from 2010 to 2018, and all studies used instantaneous scan-sampling methods [63]. Because each study was independent from all other studies, each study involved behavioral data collection by only one or two researchers; when two researchers worked together, one person collected scan sampling data for a species at a particular time. Scan sampling occurred every minute or every two minutes, depending on the specific characteristics of each study (e.g., number of individuals sampled, size of exhibit, species-specific characteristics such as quickness of movement across exhibit; details available in Appendix A). Such time intervals typically result in the same estimates for behaviors [64,65]; therefore, it was not a concern that our 10 studies used a mixture of one-minute or two-minute scan intervals. A typical observation period lasted 120 min (Appendix A). During each scan, the variables associated with human presence (Table 2) and the behavior of each animal in sight (Table 3) were recorded.

Human-focused independent variables addressed in 16 species across 10 studies ¹.

¹ Each study measured one or more of these human-focused variables. Most studies addressed only one human-focused variable. The human variable was recorded at each scan sample. Details are available in Appendix A. ² For species that were studied multiple times, the study identification number (ID) is listed for clarification. ³ The classifications for aquatic visitor abundance were larger than for non-aquatic visitor abundance because the interactive pool allowed for visitors to position themselves around the entire perimeter of the pool. None of the non-aquatic exhibits allowed for visitors to be around the entire perimeter.

Ethogram for the behaviors studied ¹.

¹ Specific ethograms for each of the 10 independent studies of 16 species are provided in Appendix A. The behaviors listed here were all behaviors analyzed. These behaviors were part of the ethograms for multiple species, but these behaviors were not analyzed when the species exhibited them infrequently (<5% of behavioral scans). All ethogram behaviors were analyzed for at least one species, except for aggression. Researchers recorded when instances of aggression occurred, but these occurrences were so infrequent (or did not occur at all) that they did not comprise a sample size large enough for analysis for any of the species (or individuals).

For most of the 16 species, the ethogram (Table 3) was consistent across the studies. Because there were some species-specific behaviors (e.g., swimming was included only for the fish), full details for each of the 10 studies are provided (Appendix A Methods). We focused on the behaviors listed in the ethogram (Table 3) because these behaviors have been often analyzed in previous studies on the effect of zoo visitors on zoo animals (see [21] and the introduction of the current manuscript for a review).

For three studies (Study IDs 7, 8, and 10) representing seven species (white-cheeked gibbon, Sumatran orangutan, cownose ray, southern stingray, bonnethead shark, white-spotted bamboo shark, and brownbanded bamboo shark), the physical locations of the zoo animals were also noted at each scan (details provided Appendix A Methods).

When possible, behavioral scans were conducted on individual animals, based on an individual’s recognizable physical characteristics. Such single-subject analyses can be important because individuals can have varied responses to variables [59]. When individual identities could not be confirmed throughout a study’s duration, group scan sampling was used to record overall group behavior on an interval. Although group sampling does not provide the detailed level of individual behavior that individual scan sampling provides, this methodology is reliable in studies of free-ranging animals in their natural habitat [66].

Each of the species in the 10 studies were analyzed separately because these 10 studies were fully independent of each other. We tested for the presence of a relationship between a human variable (e.g., visitor presence, visitor abundance, visitor proximity, or noise level; Table 2) and zoo animal behavior (Table 3) for each study independently (full explanation of statistical analyses are provided in Appendix A). Most studies addressed one independent variable (Table 2), unless stated otherwise.

Changes in such behaviors do not necessarily indicate a positive or negative welfare status [21]; therefore, our analyses focused on whether or not the human-associated variable was associated with changes in the zoo animals’ behaviors, and to what extent the behavior changed. We defined a change in behavior occurred when p ≤ 0.05 or p ≤ 0.025, based on the number of comparisons made, and we ran Bonferroni-corrected post-hoc tests to determine patterns in behavioral changes (please see the details in Appendix A). Whenever possible, the analyses focused on each individual animal, instead of overall patterns by the group of individuals. We chose this individual-based approach because zoo management plans that focus on individuals, instead of generalized responses by a population, can help focus on the well-being of each individual zoo animal [58].

No statistical analyses focused on identifying predictive variables indicating the likelihood that certain species would respond to humans. Such predictive variables could not be identified because (1) behavioral changes were noted in a large majority of species; and (2) confounding variables existed: for example, the fish were the only species in an exhibit where direct contact with zoo visitors occurred.