1. Livestock guardian dogs (LGDs) are increasingly used to protect livestock from predators, but their effects on the distribution and behaviour of wild predators are mostly unknown. A key question is whether LGDs exclude predators from grazing land, or if predators continue to use areas with LGDs but modify their behaviour in ways that reduce impacts on livestock.
2. We studied effects of LGDs (Maremma sheepdogs) on distribution and behaviour of red foxes Vulpes vulpes in north-eastern Victoria, Australia. We mapped the activity of LGDs across the study areas using GPS tracking and measured fox activity using remote cameras. We also measured risk-sensitive foraging in foxes to test if they reduced feeding time at sites regularly used by LGDs.
3. Foxes occurred throughout areas occupied by LGDs, but their probability of detection was negatively related to probability of LGD presence. Foxes extracted fewer food items from experimental food stations in proportion to the intensity of local activity of LGDs. This indicates that though foxes overlapped with LGDs, they responded to risk of encountering LGDs by allocating less time to foraging. 
4. While LGDs do not necessarily exclude wild predators from areas used for livestock production, they can have strong effects on predator behaviour. Reduction in time allocated to foraging in areas regularly used by LGDs could lead to suppression of hunting behaviour and therefore a reduction in attacks on livestock. The flexible response of predators to LGDs should facilitate coexistence of wild predators with livestock farming, by allowing predators to continue to use areas occupied by livestock while still preventing attacks on those livestock. Our results therefore strengthen the case for use of LGDs in the conservation of predators threatened by conflict with farming. Suppression of hunting behaviour should also mean that prey species experience reduced rates of predation on farmland with LGDs. This effect could be valuable for conservation of threatened species of prey.

For detailed description of the methods, please see our paper: 'Livestock guardian dogs establish a landscape of fear for wild predators: implications for the role of guardian dogs in reducing human-wildlife conflict and suppporting biodiversity conservation', published in Ecological Solutions and Evidence. 

Data was collected on four properties in north-east Victoria, Australia; two were used as experimental sites (Heatherlie and Riversdale) and two were used as controls (Mullenmeah and Wagonbark). The experimental sites ran Maremma sheepdogs, guarding sheep, and the control sites ran livestock without livestock guardian dogs. Each experimental site was paired with its own control site (Riversdale and Wagonbark, 10 km apart; Heatherlie and Mullameah, 45 km apart) due to the fox surveys and experiments taking place in different seasons for each pair of properties; fox activity and behaviour is seasonally highly variable.

GPS tracking collars of two types (Lotek, Havelock North, New Zealand; and Telemetry Solutions, Concord, USA) were fitted on all LGDs between July 2017 and March 2018, the two types being interspersed across properties. One collar failed at Heatherlie and could not be replaced due to the dog’s shyness. Collars took a location every 30 minutes and were fitted a minimum of four weeks before collection of data on foxes. Only locations with a HDOP (horizontal dilution of precision) <10 (Lotek collars) or <4 (Telemetry Solution collars) were retained for analysis. HDOP values were chosen as those that offered the best balance between filtering out inaccurate locations and data retention, based on a pilot study of stationary GPS collars. A mean (± SE) of 3.0% ± 0.4% locations was deleted from the datasets collected by the Lotek collars, resulting in a mean (± SE) HDOP of 2.0 ± 0.05 of the retained sample of locations. A mean (± SE) of 2.4%± 0.8% was deleted from the datasets collected by the Telemetry Solution collars, resulting in a mean (± SE) HDOP of 1.2 ± 0.09 in the retained sample. 

Forty-eight Reconyx PC800 HyperFire Professional IR cameras (Reconyx, Holmen, WI, USA) were distributed over each pair of trial and control properties for a 6-week survey (2,016 trap nights). Cameras were set to take three images in rapid succession when triggered, with a minimum one-minute delay between consecutive triggers to reduce repeat triggers by the same individual. On Riversdale and Wagonbark, this survey ran in August and September 2017; on Heatherlie and Mullameah it ran in December 2017 and January 2018. Half of the 48 cameras were allocated to the control site and were evenly distributed over the chosen area in a grid pattern. The remaining 24 cameras were distributed over each trial site, according to the intensity of use of different areas by the Maremmas (see below).

 A minimum of four weeks of data from the GPS tracking collars were used to calculate a fixed kernel home range (Worton 1989) for each dog group by pooling the locations of all members. We used an ad hoc smoothing parameter designed to prevent under- or over-smoothing, which involved choosing the smallest increment of the reference bandwidth (Href) that resulted in a 95% home-range polygon that was as contiguous as possible (Jacques et al. 2009; Kie et al. 2010). The package ‘adehabitat HR’ (version 0.4.19) in R statistical software (Calenge 2006; R Core Team 2013) was used for all home range calculations. The 10%, 50%, 90%, 95% and 99% isopleths were extracted from this home range calculation, and the area covered by each incremental isopleth was determined. The 24 cameras were distributed in order to equalise camera density for each incremental area, with a higher density in the 10% and 50% isopleth areas to maintain a minimum of two cameras per isopleth zone.

Experimental fox feeding trials exploited the propensity of foxes to dig for food items and were designed to yield a measurement equivalent to a giving-up density (GUD). At each site we dug a 30-cm deep hole approximately 2 m in front of the camera and filled it in while placing a chicken neck at each 5-cm depth interval, the top chicken neck being buried just below the surface. Chicken necks were chosen based on a pilot study that showed they are favoured by wild foxes. The cameras monitoring the foraging behaviour of foxes at trial sites were set to take three images in rapid succession at each trigger, with no delay between consecutive triggers.

When analysing the data in the paper, two measures of Maremma presence were calculated for each camera site. The first was the probability of Maremma occurrence based on home range calculations. We obtained this probability from a long-term (eight-month) home range calculation of each dog group on each property using the Brownian bridge approach of the kernel method (Bullard 1991; Horne et al. 2007). Locations at the control sites were allocated a value of zero. To enable statistical analysis with these values, they were transformed with the following formula: (log10(probability value+1))*10000. The second measure was the number of days on which Maremmas were detected at each camera site during the survey. As Maremmas were rarely detected on camera (N=9), we identified all instances in which a GPS collar from a Maremma logged a location within 30 m of a camera, and added these to the camera detection data (N=76) to calculate ‘days of Maremma detection’. This measure proved to have a greater effect in all models used in our analysis than ‘probability of Maremma occurrence’, so only ‘days of Maremma detection’ was used as an indicator of Maremma presence. At Mullameah, all camera locations that fell within 300 m of their wild dog exclusion fence were excluded from analysis because of poison baiting along the fence just prior to the research. At Riversdale and Heatherlie all camera locations beyond the 95% isopleth were excluded due to continuing lethal fox control on the edge and outside the experimental properties.

The uploaded data includes:

• For each trial/control site combination; the results from the wildlife survey, the two measures of Maremma presence at each camera site (as above) and the covariates for each camera site used in the analysis in the paper
• For each trial/control site combination; the results from the fox feeding trial, with the covariates for each camera site as used in the analysis in the paper.