Introduction

Behavioural and ecological studies typically assume that sampled individuals are representative of the broader population and the violation of this assumption would introduce sampling biases that can distort conclusions and even affect conservation efforts [1,2]. Since many research questions entail the need for capturing animals to gather otherwise inaccessible data on behaviour, genetics and physiology [3–6], one often-overlooked source of bias arises from consistent individual differences in behaviour, possibly influencing an individual’s likelihood of being captured. Such behaviour-dependent biases can systematically skew research findings not only about behaviour itself, but also about population structure and ecology, because behaviour is often correlated (functionally and/or genetically) with physiological and life-history traits. As a result, non-random sampling of behavioural phenotypes can bias demographic inferences, including population size estimates, and systematically shift estimates of trait means and distributions [1].

Previous work has mainly addressed this issue in the framework of animal personality and behavioural syndromes [7], testing whether certain personality traits are disproportionately represented in captured samples [8–10]. However, because capture bias is driven by behavioural outcomes that affect capture probability rather than by their ultimate causes, it follows that any behavioural pattern consistently expressed within individuals can generate bias if it alters exposure to trapping or other risks. Risk-prone and more active individuals, typically more likely to undergo capture procedures, can indeed become overrepresented in captured samples [2,9]. The extent of this bias typically depends on the capture method employed: passive methods, such as baited traps, rely on the animal’s voluntary entry and often oversample those individuals that are less fearful of artificial apparatuses [11,12]. Conversely, active methods, which involve direct human action to physically restrain or chemically immobilise an animal, are generally expected to depend less on the animal’s immediate behaviour at the moment of capture and may therefore provide a more representative sample of the population [13]. For instance, hand nets captured individuals exhibiting greater variance in boldness in three-spined sticklebacks (Gasterosteus aculeatus) compared to passive traps, which were biased towards bolder and more sociable individuals [13]. However, capture bias has been reported for both active and passive methods [9,10], whereas other work found no significant biases regardless of the method used [8]. This mixed evidence underscores the relevance of explicitly testing behavioural representativeness across capture techniques and contexts. Beyond methodological challenges, other intrinsic factors such as sex and age can influence the probability of being trapped, with males and juveniles often being more exploratory and thus more likely to be captured by passive trapping than females and adults, respectively [14–16].

While these studies provide valuable insights into capture bias, fully assessing its extent remains inherently challenging, as direct comparisons between captured and non-captured individuals would be necessary to accurately assess its effects. However, behavioural data often depend on capture itself, either to facilitate behavioural assessments under controlled conditions (e.g., [17,18]), mark and deploy tracking devices (e.g., [5,19,20]), or, rarely, both [21]. To circumvent this issue, most studies compare individuals captured using different methods [8,9,13,22] or analyse recapture probabilities in capture-recapture frameworks [10,16,23]. Yet, these approaches still rely exclusively on the trappable portion of the population. Direct comparisons between captured and non-captured individuals are rare and typically confined to controlled environments (and notably, still among pre-captured samples of individuals, e.g., [12,15]), leaving substantial gaps in our understanding of capture bias under natural conditions (but, for naturalised fish populations, see [1]).

Behavioural capture bias has been investigated across a broad taxonomic range with studies testing whether particular behavioural patterns are over-represented in captured samples of birds, reptiles, fish and small-to-medium-sized mammals [8,9,12–15]. In contrast, despite its potential importance for large-bodied species, capture bias has seldom been explicitly tested in ungulates and other large mammals (but see Reale et al., [10]), particularly under field conditions and across capture methods, likely owing to their wide home ranges, the difficulty of reliably recognising individuals, and the logistical challenges of assessing consistent behavioural differences in the wild [24]. Given the frequent use of capture techniques in large mammal studies for ecological research, evaluating potential sampling biases is crucial for ensuring robust ecological conclusions. Moreover, because such biases may affect the efficiency of capture-based interventions such as removals and translocations, their assessment is also fundamental for improving wildlife management and conservation strategies.

This study investigates whether two active capture methods, such as chemical immobilisation and vertical drop nets, used to capture Northern chamois (Rupicapra rupicapra), yield representative samples of the population with respect to the ecologically relevant behaviours. To test the hypothesis of capture procedures introducing behavioural biases in the sample of captured individuals, we compared behavioural patterns of 25 captured chamois with their non-captured conspecifics by means of direct observation of their spontaneous behaviours, focusing on those related with the exposure to risks, using intra-group comparisons to account for unrecognizable individuals and controlling for demographic and environmental confounding effects. First, to confirm that the selected behaviours were suitable for evaluating inter-individual behavioural variation, we tested whether the recorded behavioural patterns varied significantly and consistently among captured individuals. Second, we tested whether the behavioural patterns of the captured individuals significantly differed from those of their non-captured counterparts. If capture methods introduce bias, we predicted that collared individuals should exhibit distinct behavioural patterns compared to non-collared individuals. Beyond assessing the presence of capture bias in our case study, we aim to provide a practical methodological framework for evaluating the occurrence and potential consequences of capture-related sampling biases in field studies of wild animals. By combining non-invasive direct observations with within-group comparisons in free-ranging individuals, this approach allows researchers to assess whether capture procedures yield a representative sample of behavioural variation, thereby strengthening the validity of behavioural and ecological inferences used in research, conservation, and management.

Methods

Study area

This study was conducted in the Monte Grappa Massif (Venetian Prealps, North-East Italy; 45.873103N, 11.808166E, Fig 1), a 400 km² mountainous complex reaching a maximum elevation of 1 775 m a.s.l.. The massif is geographically distinct, bordered by the Piave and Brenta rivers and spanning the provinces of Treviso, Belluno, and Vicenza. The environment transitions from mixed deciduous forests at lower elevations (600–1 300 m) to a mosaic of beech-conifer forests and open areas between 1 300 and 1 650 m. Above 1 650 m, the landscape is dominated by alpine meadows and rocky outcrops (Fig 1). The climate is suboceanic pre-Alpine, with snowy winters, mild summers, and frequent fog formation on southern slopes. Human presence in the area is relatively high during spring and summer due to mountain sports such as hiking, climbing, biking, and paragliding, but is not negligible even throughout autumn and winter. This is facilitated by a relatively well-developed paved road network, which allows year-round access to the mountain top by car. Meanwhile, permanent human settlements are virtually absent.

Download:

• PNG
  larger image
• TIFF
  original image

Fig 1. Study area on the Monte Grappa Massif, showing land cover and the GPS locations recorded during the behavioural observation period (May–June 2024) by collars deployed on the 27 captured Northern chamois.

To improve readability, only locations recorded at 12:00 each day are shown. The inset shows the location of the Monte Grappa Massif in Europe. Land-cover data: Regione del Veneto, Banca Dati della Copertura del Suolo (Italian Open Data Licence, IODL 2.0). Hillshade (inset): European Environment Agency (EEA), Hillshade Europe DEM (CC BY 4.0).

https://doi.org/10.1371/journal.pone.0351124.g001

After their extinction in historical times, Northern chamois were reintroduced to the area between 1995 and 2006 using wild individuals translocated from other regions of the Italian Alps (Maritime and Dolomite Alps). The historical extirpation of chamois from the massif is plausibly linked to overexploitation during the period when the World War I front crossed this area. Chamois primarily inhabit higher elevations above 1 300 m. Since 2016, they face predation from grey wolf (Canis lupus), following the species’ natural recolonization of the Alpine region after historical declines caused mainly by persecution linked to livestock conflicts.

Behavioural observations

Behavioural observations were conducted in May and June 2024, allowing at least two months for chamois to recover from capture and ensuring that recorded behavioural patterns were not influenced by capture-related stress [5,19]. Observations were restricted to a relatively homogeneous 5 km² area at the top of the massif, where all collared chamois were residing. Data collection was conducted exclusively in open meadows and during daylight hours, with most observations occurring from 11.00 AM to 08.00 PM, when weather conditions, particularly the absence of fog, facilitated data collection.

Observations were conducted on groups containing at least one collared and one non-collared adult chamois. Collared individuals (hereafter “focal”) were initially listed in capture order, with selection following a rotating priority system to prevent operator bias. Two operators used GPS data and, occasionally, a Very High Frequency (VHF) beacon to locate focal individuals. Once identified by their unique collar and ear-tag colour combination, observations commenced if the focal was accompanied by at least one uncollared adult of the same sex in the same group (as defined by the 50 m chain rule, where each chamois in the group is within 50 m of any other chamois). The location and start time of each observation session were recorded. The coordinates of the observation were later used to quantify the proximity to the nearest hiking trail, as the Euclidean distance (m) between each observation location and the closest segment of the official trail network provided by the Veneto Region (https://idt2.regione.veneto.it/idt/webgis/viewer?webgisId=216). Groups were observed from a distance of at least 100 m using binoculars for a minimum of 30 minutes. If another collared individual was in sight at the end of this period, it became the new focal, and a new observation session began. Otherwise, the initial focal was observed for an additional 30 minutes before concluding, for a maximum observation duration of 60 minutes per focal individual. After an observation session ended, observers could either initiate a new session on another collared individual or terminate observations.

Behavioural data were collected via instantaneous scan sampling at 3-minute intervals, recording mutually exclusive behavioural states and intra-group spatial positioning (see Table 1 for a detailed description) of all adult chamois of the group. Relative intra-group spatial positioning was recorded live and visually estimated using a virtual grid, then classified into a four-level group centrality index ranging from 1 (peripheral) to 4 (central; see S2 File for details). The collared status (yes or no) of each individual, its sex, and its identity (for collared chamois only) was also noted. Of the seven recorded behaviours, only relative intra-group positioning, rest, eat, stand, walk, and non-confrontational aggressive behaviour were analysed, as confrontational aggressive behaviours were too rare to model (<20 occurrences), resulting in insufficient data for reliable parameter estimation and inference. To ensure coding reliability, 70% of the behavioural records from the primary observer were compared against a continuous live coding session performed on the focal individual by a second observer using the BORIS v. 8.25.4 mobile application. Cohen’s kappa values summarising inter-observer agreement are reported in Table 1, with values of 0.40 to 0.60, 0.60 to 0.75, and > 0.75 representing fair, good, and excellent agreement, respectively [26].

Download:

• PNG
  larger image
• TIFF
  original image

Table 1. Description of the analysed behaviours in Northern chamois, inter-observer agreement indices, relationship with the exposure to risks, and supporting references.

https://doi.org/10.1371/journal.pone.0351124.t001

A total of 69 independent observation sessions were conducted on 25 collared chamois (11 females, 14 males) across 22 observation days, as one female and one male were never located in open areas. As multiple collared individuals were occasionally observed within the same group and could be monitored simultaneously, this amounted to 135 individual observation sessions. Each collared individual was observed in an average of 5.40 ± 2.89 independent sessions, accumulating a mean total of 3.68 ± 1.91 hours per chamois (mean ± SD). The observed groups comprised an average of 3.84 ± 2.20 adult chamois per group, with collared individuals accounting for a mean proportion of 0.45 ± 0.20 of the total group size.

Results

Repeatability of collared chamois’ behaviours

We detected a significant repeatability for eat, stand, rest, and group centrality index. The repeatability significantly differed from 0 for eat (R = 0.10 ± 0.03, mean ± SE, P < 0.001), stand (R = 0.05 ± 0.02, P < 0.001), rest (R = 0.15 ± 0.04, P < 0.001), and group centrality index (R = 0.15 ± 0.05, P < 0.001). The models’ summaries also revealed significant effects of the control predictors, with a decrease of eat (β_{time from noon} = 0.69, P = 0.02) and an increase of rest (β_{time from noon} = −1.04, P = 0.008) in the central hours of the day, where ‘time from noon’ is the absolute number of hours from 12:00. Moreover, a significant difference between sexes emerged from the group centrality index model, with males occupying more peripheral positions compared to females (β_male = −0.66, P = 0.006). Conversely, the repeatability tests for walk (df = 1, χ² < 0.001, P = 1) and non-confrontational aggressive behaviour (df = 1, χ² < 0.001, P = 0.50) did not reveal significant individual repeatability.

The Pearson correlation analysis revealed significant negative relationships between rest and eat individual scores (r_rest-eat = −0.55, N = 25, P = 0.004), indicating that individuals resting more frequently tended to be less likely to eat. No significant correlations were found among the other individual scores. The overall distribution of the 25 individual chamois’ behavioural scores is represented in Fig 2.

Download:

• PNG
  larger image
• TIFF
  original image

Fig 2. Individual behavioural scores for 25 Northern chamois captured, equipped with collars, and observed in Northeast Italy in 2023-2024.

The x-axis represents the rest score, while the y-axis represents the eat score. Dot size corresponds to the stand score, dot colour represents the group centrality score, and dot shape indicates sexes. Individual scores for rest, eat, stand, and group centrality were obtained by means individual random intercepts of Mixed Models (see section Repeatability of collared chamois’ behaviours of the main text for more details).

https://doi.org/10.1371/journal.pone.0351124.g002

Capture bias testing

The likelihood ratio tests revealed no significant differences among the chamois being captured with chemical immobilisation, drop nets, and not captured at all, on any of the considered behaviours, including eat (χ² = 2.54, df = 4, P = 0.64), stand (χ² = 0.63, df = 4, P = 0.96), rest (χ² = 6.19, df = 4, P = 0.19), and group centrality (χ² = 4.69, df = 4, P = 0.32) (Fig 3).

Download:

• PNG
  larger image
• TIFF
  original image

Fig 3. Variability in the behaviours of chamois (a) eating, (b) standing, (c) resting, and (d) group centrality, as predicted by the full models described in section Capture bias testing of the main text.

Predictions are shown across the chamois captured using chemical immobilisation, using drop nets, and non-captured, separately for males and females. Vertical bars represent 95% confidence intervals (CI) obtained through bootstrapping, while grey dots represent the observed scan-sampled raw data. Rectangular areas enclosed by dashed lines indicate a break in the y-axis.

https://doi.org/10.1371/journal.pone.0351124.g003

Similarly, the Bayesian analyses provided an overwhelming evidence against a difference among individuals captured with different methods, or not captured at all, in the likelihood of exhibiting these behaviours, with BFs strongly supporting the null hypothesis for eat (BF = 0.0017 ± 2.84%), for stand (BF = 0.0009 ± 3.95%), for rest (BF = 0.006 ± 19.10%) and for group centrality (BF = 0.01 ± 27.11%).

Discussion

Our results revealed significant and repeatable individual variability in four of the investigated behaviours within the study population, while also demonstrating that the two evaluated capture methods provided a reliable representation of this existing behavioural diversity.

The significant repeatability of eating, resting, standing, and group centrality index indicates that these behaviours were consistently expressed within individuals and significantly varied among them [39], supporting their suitability as indicators of consistent inter-individual behavioural variation in the population. In contrast, walking and non-confrontational aggressive behaviours did not exhibit significant within-individual repeatability, likely due to their relatively low frequency, at least when recorded through scan sampling. The negative correlation between resting and eating, along with their significant variation throughout the day (consistent with a summer bimodal activity rhythm, [41]), suggests their association with activity levels, although their relevance in characterizing individual differences in the activity behavioural trait would have required an independent confirmation. By contrast, we found no evidence of individual-level covariation between standing and group centrality (Table 1). This pattern suggests that these behaviours may not reflect a single underlying behavioural trait or syndrome but could instead represent individually repeatable trade-offs between different forms of risk-taking, as observed in other wild ungulates [24].

Regardless of their relationships with underlying behavioural traits, overall activity (time spent eating versus resting), standing behaviour, and group centrality remain key components of individual risk–resource trade-offs. More active individuals acquire more resources at the cost of increased energetic expenditure and exposure to risks [32,33,55]; standing entails a trade-off between foraging efficiency and vigilance [29–31]; and centrality provides safety at the cost of increased intraspecific competition [36,37]. Given their ecological relevance and consistent within-individual repeatability, investigating whether capture methods selectively sample individuals differing in these behaviours remains crucial to understanding potential biases in behavioural ecology studies on individuals and populations.

Contrary to most studies investigating capture bias in other systems [2,9,11,12], our comparison of captured and non-captured chamois, while accounting for sex differences, demonstrated the absence of such bias, in accordance with Michelangeli et al. [8], who used active capture methods on delicate skink (Lampropholis delicata) despite not explicitly comparing captured and non-captured individuals. Similarly, our findings indicating the absence of sampling bias may be attributed to the use of capture methods that are entirely active, meaning they do not require animals to engage in exploratory behaviour toward artificial objects, such as traps or other capture devices [8]. Nonetheless, our capture procedures still involved behavioural responses to the approach of human operators, which could have introduced biases based on individual differences in risk perception [2,9]. For drop nets, capture success was primarily driven by chamois flight responses, leading them to run towards the nets, regardless their tolerance to human operators, making the lack of a behavioural trait-skewed sampling unsurprising. Conversely, chemical immobilisation required operators to approach individuals at relatively short distances, raising the possibility of bias based on inter-individual variation in flight initiation distance. In this context, the longer flight distances of females compared with males (e.g., 46 and 108 m in males and females, respectively [56]) contributed to making them unsuitable targets for this method, which we deliberately applied only to males. This is consistent with the stronger risk avoidance generally shown by female chamois, in line with their sex-specific life-history strategy [57], and with our finding that females occupied safer, inner positions in mixed groups. However, female behaviour and flight distance were only part of the constraints preventing chemical immobilisation in females in this study: pregnancy and lactation limited the period in which anaesthetic drugs could be safely administered, while during the remaining suitable months females tended to remain near steep rocky areas, where sedation would have been unsafe. Therefore, although behavioural differences may have contributed to the impracticability of chemically immobilising females, our study system did not allow us to directly evaluate this potential source of sex-skewed samples.

However, our results showed that, within males, chemical immobilisation did not introduce any sampling bias, despite the 30–40 m darting distance being similar to or shorter than the male flight distances reported in other studies (46 m in Hamr [56], 103 m in Gander & Ingold [58]). A likely explanation is the relatively high and widespread tolerance to human proximity observed in our study population, likely facilitated by frequent exposure to recreational activities. Such exposure is known to habituate chamois to non-threatening human presence and has been shown to substantially reduce flight distances to approaching humans [59]. Accordingly, caution is warranted when extrapolating these results to other species, different ecological contexts, or study areas where chamois exhibit greater wariness toward human presence. Nonetheless, by explicitly testing behavioural representativeness across two active capture methods and by directly comparing captured with non-captured individuals in a free-ranging large mammal, our results suggest that while capture bias remains a concern that warrants consideration in large mammal ecological studies, its occurrence is not unavoidable, even when using methods that rely on individual tolerance to a human operator.

Our study entails some limitations. First, behavioural assessments were based on passive observations of spontaneously exhibited behaviours in natural conditions rather than controlled experimental setups. Since this approach may limit cross-study comparability and make relationships between observed behaviours and underlying individual traits uncertain [60], we acknowledge that our results cannot exclude possible biases in behavioural syndromes or traits. On the other hand, since the observation of spontaneous behaviours in non-manipulated contexts provides a more ecologically valid representation of behavioural variation within the population [61], our approach was well-suited for assessing potential capture bias in behavioural ecology patterns, particularly those related to the individual exposure to risks. Furthermore, although experimental approaches such as flight initiation distance trials could offer an alternative method to detect capture bias in large mammals, we deliberately avoided setups that might artificially alter chamois spatial movements, which could in turn bias the behavioural ecology analyses planned within the project (as shown by reduced use of open areas after flight distance experiments, [58]). Additionally, because we evaluated only two active capture methods, we could not directly compare active versus passive sampling approaches; such comparisons would be valuable to more broadly validate behavioural sampling bias within populations. Moreover, because chemical immobilisation was not applied to females, we were unable to assess potential capture bias for this method in them. Consequently, our conclusion that chemical immobilisation does not introduce behavioural bias applies only to males. Finally, our study focused exclusively on adult individuals of a relatively human tolerant population and was restricted to a specific period of the year. While this may limit the generalisability of our findings across life stages, ecological contexts and seasons, it also helped minimise potential confounding effects, allowing for a clearer interpretation of individual behavioural variation and its implications for capture bias.

This study provides a rare explicit assessment of capture bias in a wild, free-ranging large mammal, and, to our knowledge, the first to quantify it by directly comparing captured and non-captured individuals while also evaluating multiple active capture methods. We show that the two methods used for Northern chamois yielded a sample that accurately reflected the population’s behavioural diversity, particularly in behaviours linked to risk exposure and resource-acquisition trade-offs. However, to ensure robust conclusions about individual and population-level behavioural and ecological patterns, potential capture biases should be more systematically considered in behavioural ecology studies. Given its non-invasiveness for animals and practical applicability for researchers, we propose our protocol as a robust and transferable framework for behavioural studies involving the capture of wild animals, at least for species and environmental contexts where it is feasible to repeatedly observe the spontaneous behaviour of both captured individuals and their non-captured conspecifics.

Supporting information

Acknowledgments

We sincerely thank all the operators involved in the chamois captures, with special appreciation for Liliana Costanzi, Michele Zanni, Silvia Cavazza, Celeste Buelli, Siriano Luccarini, and the officers of the Provincial Police Corps of Treviso, Vicenza, and Belluno for their invaluable assistance. We also extend our gratitude to the administrations of these provinces for their logistical support throughout the study.