Individual variation in tolerance of human activity by urban dark-eyed juncos (Junco hyemalis) 2021 Stansell et al data file

An important goal of urban ecology is determining what differentiates urban-tolerant populations of birds from their non-urban ancestors and urban-intolerant species. One key to urban success may be reacting appropriately to human activity, and the degree to which birds view humans as threats can be quantified by their escape behavior.  Understanding individual-level plasticity, however, requires the tracking of known individuals. We compared flight-initiation distances (FID) and distances fled (DF) from approaches by a human between an urban and a non-urban population of individually-marked Dark-eyed Juncos (Junco hyemalis) in southern California. The urban population is more tolerant to people as evidenced by attenuated FIDs and DFs relative to non-urban birds.  Although individual urban birds either habituated or sensitized to repeated approaches, there was no significant pattern at the population level.  Overall, the behavioral patterns exhibited by this urban population of juncos is more supportive of in situ evolution than either being a biased sample from an ancestral non-urban population or intrinsic behavioral plasticity that produces a uniform adjustment to urban life.

Methods
All encounters were recorded by the same individual (HMS) following the protocol in commonly used to study FID (Cooper and Blumstein 2015). An encounter began when a bird was observed foraging or stationary, and not alarm calling or otherwise visibly agitated. Because juncos are territorial, most encounters with a given bird occurred near where it was initially captured and banded. Birds were always approached in a straight line at a practiced pace (approximating 0.5m/sec) when on the ground and exposed from vegetation, with no obstacles or other juncos between the observer and the focal bird. This ensured consistent, readily-detectable approaches to each individual (Frid and Dill 2002, Tätte et al. 2018). A colored marker was dropped at the location where the experimental approach began (the starting distance, SD), a second at the observer location when the focal subject fled (the FID), and a third at the location from where the focal subject fled, later converting paces to meters (0.825m/step). The distance fled (DF) was recorded by visually estimating the horizontal and vertical distance travelled in meters, then converting to a Euclidian distance. In some cases, it was not possible to collect data on DF because the bird left the immediate area.  Such occurrences were arbitrarily recorded as a distance fled of 50m.

We also recorded the time of day, presence/absence of conspecifics within a 5 m radius of the focal bird, distance to nearest cover (either bushes and trees, or structures such as walls and outdoor furniture), and pedestrian density.  Distance to cover was measured in paces from the targeted bird’s location when the approach began. Pedestrian density (only at UCLA – there were almost never pedestrians at the James) was recorded categorically as low (defined as <5 people per minute crossing a 10m sample transect in the immediate vicinity of the approach) or moderate to high (≥5 people/min).

At UCLA, we collected repeated measures for 22 series on both a short time scale (4 attempted approaches to the same bird on the same day) and on a longer time scale (over a consecutive 4-day period).  Not every series had the complete sequence of 16 approaches (6 had one missing value in terms of a missed approach on one day, 4 had two missing values, and 1 had four missing values because of a missed day).  The median time between sequential approaches within a day across 198 approaches was 3 minutes, with 183 times being 30 minutes or less. Only 6 approaches were separated by more than an hour. Birds were approached during the 2017 breeding season while rearing chicks (February to July at UCLA; June to July at the James) in order to be consistent across sites relative to foraging needs (i.e., risk-taking can differ in individuals when breeding versus non-breeding: Mikula et al. 2018).  In total, we collected 404 approaches across 31 individuals at UCLA, and 104 approaches across 22 individuals at the James Reserve.  At both sites, the majority of data were collected between 08:00 and 13:00 h.

Statistical tests on the multiple approaches towards marked birds in UCLA followed standardized analytical methodology (Pezner et al. 2017, Dehaudt et al. 2019, Andrade and Blumstein, 2020). All analyses used R 3.4.2 (R Core Team 2017), on code modeled after Pezner et al. (2017), which had an identical experimental design: 4 trials per day, repeated across 4 consecutive days. We used linear (FID) and logistic (DF, with distances ≤2 m as near and >2 m as far) mixed-effects models for individual responses to repeated approaches on the UCLA campus. Effects within UCLA and across the James population values were compared by regression analysis, paired t-tests and two-sample, unpaired t-tests, as appropriate.

We fitted models using the R package "lme4" v1.1-14 (Bates et al. 2015) (supporting package "car" v2.1-6: Fox and Weisberg 2011). A null model used individual bird as a random intercept, then iteratively incorporated fixed effects (contextual variables), with stepwise selection to find the combination of fixed effects with the lowest AIC value. Each predictor variable was added to the model and then selectively removed depending on their effect on AIC relative to the null model. In cases where a fixed effect resulted in only a non-significant decrease in AIC, likelihood ratio tests used "lmerTest" v2.0-33 (Kuznetsova et al. 2016) to measure significance. Fixed effects without significant model improvement were discarded.

After selecting a model via this process, a likelihood ratio test evaluated whether either the inclusion of trial iteration as a fixed effect or as a random slope significantly improved the explanatory power over the model containing contextual predictor variables and the random intercept. Best mixed-models were compared against their fixed effects-only counterparts via likelihood ratio test using "RLRsim" v3.1-3 (Scheipl et al. 2008), supporting packages "MASS" v7.3-47 and "arm" v1.9-3 (Gelman and Su 2016) to determine whether individual differences among birds explained a significant portion of behavioral variation. Where individual was a significant random effect in models, adjusted repeatability was calculated using code provided by Jean-Nicolas Audet, modified from "rptR" v0.9.21 (Stoffel et al. 2017). Analyses were based on 334 individual approaches.