Advances in biologging technology have led to the increased use of animal-borne tri-axial accelerometers to study domesticated and free-living animals [1,2,3]. This technology enables detailed information to be recorded about an individual’s orientation and body movement [4]. Devices typically record acceleration data at high frequencies (e.g. 20–50 Hz) in three axes which enables dynamic body acceleration to be calculated, such as vectoral dynamic body acceleration (VeDBA) [5, 6]. Measured acceleration can be used to identify patterns of behaviour through analyses of body movement and orientation [7,8,9,10]. Body acceleration has also been shown to correlate with oxygen consumption, and along with other movement-related metrics, can be used as an index of energy expenditure [11,12,13]. Furthermore, it can be included with other recorded data such as magnetic compass heading to infer movement paths of individuals by dead-reckoning [14].

The importance of correct attachment and positioning of biologging devices is broadly accepted, both for animal welfare considerations [15,16,17] and application purposes such as tag position to answer specific movement-related questions [5, 18]. Generally, devices are attached to the subject’s body, as close to the centre of mass as possible, and aligned so that the direction of measured axes is aligned with the organism’s anatomical axes [19]. Such measures are easier to apply when attaching devices directly to the animal’s hair or feathers using tape or glue, as is the case, for example, when attaching logging devices to marine mammals [20,21,22]. However, many studies utilise collar-attached systems, most commonly for deployment on quadrupedal terrestrial mammals [8, 23]. This moves the device away from the centre of mass, and unlike directly attached devices, collar-attached devices are not fixed to the body and may therefore move independently to the subject [19]. This may result in less accurate measurements of acceleration of the animal, and reduce the ability of accelerometry to determine behavioural patterns and body orientation.

Previous studies have proposed that in the interests of animal welfare, collar weight should not exceed 2% of body weight and collar size should be of appropriate tightness, dependent on the study species [17, 24]. A few studies have evaluated the effect of collar weight on the subjects’ behaviour [17, 25] and investigated the welfare consequences of collar tightness [26]. However, the effect of collar size and weight on acceleration measured by a collar-attached device has not yet been quantified. These determinations are important because device weight is related to battery size and, therefore, to the period over which a device can record. Researchers may only have one opportunity to deploy a device, and thus, it is important to ensure that collar size and weight are optimal for accurate measurement of body acceleration.

We used African pygmy goats (Capra aegagrus hircus) as a model quadruped species to measure body acceleration of walking, measured via VeDBA over a second, at different speeds on a treadmill, which acted as a controlled environment. The main aims of this study were (1) to assess the effect of different collar sizes, ranging from 3 to 10 cm larger than the subjects’ neck circumferences (S1–S5), and weights, ranging from 0.4 to 1.2% of the subjects’ body weights (W1–W5), on the measured acceleration of a collar-mounted device (Table 1); (2) to investigate the acceleration of a device acting as a control, positioned on the body of the animal using a harness so that the device is not subject to collar roll; and (3) to assess the effect of collar sizes and weights on the difference between acceleration measured on the collar and the harness. Selection amongst candidate general linear models (GLMs) used an information theoretic approach based on Aikake’s Information Criteria (corrected for small sample size) (AICc) [27]. Using predictions from the GLM output, we developed guidelines for relative collar sizes and weights, in order for future studies to measure body acceleration as accurately as possible.

Table 1 Each measured combination of collar size (S1–S5), ranging from 1 to 9 cm greater than the animals’ neck circumferences, and collar weight (W1–W5), ranging from 0.4 to 1.2% of the animals’ masses, for the different goats (G01–G04)


Effect of collar size and weight on VeDBA

The interaction between collar size and weight was retained in the model, positively affecting mean VeDBA (ΔAIC = 110.7 relative to the best model that excluded the interaction) with higher mean values and more variation recorded on looser and heavier collars (Fig. 1; Additional file 1). For heavy and looser collars (S5, W5), VeDBA was predicted to increase twice as fast as that for tighter and lighter collars (S1, W1 and W2). When the collar was loose but light (S5, W1 and W2), VeDBA was predicted to increase by more than two-thirds than if the collar was tight but heavy (S1, W5).

Fig. 1
figure 1

Mean VeDBA (g) as measured on the collar-attached device for each combination of collar size and weight: a collar weights with each collar size; b collar sizes with each collar weight. Boxes represent the mean and interquartile range, and the kernel density plots represent the probability density and the variance of the data

Collar parameters had an important effect on the relationship between VeDBA and animal travel speed (Fig. 2; Table 2; Additional file 1). There was a positive relationship between speed and mean VeDBA recorded from the collar-attached device. There were also positive interactions between collar size and speed and collar weight and speed retained in the model. There was no interaction between speed and goat ID; however, goat ID was retained in the model.

Fig. 2
figure 2

The predicted relationship using the linear model between mean VeDBA (g) and walking speed for a different collar sizes and b different collar weights. Fitted lines represent the predicted relationship and error bars represent predicted standard error

Table 2 The effect of collar size and weight on measured VeDBA, including the relationship between speed and goat ID as a fixed factor

VeDBA measured by the harness

We investigated the relationship of harness-measured acceleration with the other variables to confirm that the relationship between acceleration and speed measured on the harness did not vary with different collar weights or sizes. There was a positive relationship between speed and VeDBA measured using the harness-attached device (ΔAIC = 578.6 relative to the best model that excluded the interaction; Table 2) and goat ID was retained in the model. The effects of collar size and weight were retained. However, the interactions between speed and collar weight, or speed and collar size were not retained in the model (ΔAIC = 3.3 and ΔAIC = 1.1, respectively, relative to the best model that excluded the interaction). Showing that despite retaining collar size and weight in the model, it did not influence the relationship between acceleration and speed.

Difference between collar and harness

The interaction between collar size and collar weight was retained in the model, positively affecting difVeDBA (the difference between collar- and harness-measured acceleration), with a larger difference and more variation recorded when the collar was looser and heavier (ΔAIC = 8.99 relative to the best model that excluded the interaction; Table 2; Fig. 3). When collar size increased, the predicted increase in difVeDBA (g) per km/h was up to four times greater for heavier collars than for lighter collars, and when collar weight increased, the predicted increase was up to three times larger for looser collars than tighter collars (Table 3; Additional file 1).

Fig. 3
figure 3

The predicted relationship using the model output between the differences in mean difVeDBA (g), calculated as the difference between VeDBA measured on the collar and the harness, and animal walking speed for a collar sizes and b collar weights. Fitted lines represent the predicted relationship and error bars the predicted standard errors

Table 3 Predicted measured increase difVeDBA (g) per km/h for each collar size (S1–S5) and weight (W1–W5) combination with predicted standard error

The extent of agreement between collar- and harness-derived estimates of VeDBA was negatively affected by animal travel speed as difVeDBA increased with speed (ΔAIC = 260.1 relative to the best model that excluded the interaction). There were significant interactions between speed and collar size (Fig. 3a) and speed and collar weight (Fig. 3b) on measured VeDBA. There was also a significant interaction between goat ID and speed.


Our results indicate that the size and weight of a collar are likely to affect acceleration measured in animal-borne tags. To ensure ethical deployment of collars onto subject animals, researchers must demonstrate that the best available techniques are being used that do not cause harm [16]. Therefore, in the context of animal welfare, it is also important to consider the effect of collar size and weight on the ability to measure the subject’s body movement accurately.

Collars that are too tight may cause discomfort or injury and not account for potential animal growth, whilst collars that are too loose may ensnare the animal or be prematurely removed [28]. In line with ethical recommendations, we did not test collar sizes that are predicted to cause the animal discomfort [15]. Our results show, however, that the size and weight of the collar have a significant effect on the measured acceleration when the subject is walking. Tighter and lighter collars are more likely to provide more accurate measures of body movement, with smaller differences in effect size and variation between collar sizes S1 to S3 and collar weights W1 to W3. These collar parameters are also more likely to fit with ethical constraints. In contrast, looser and heavier collars are more likely to yield larger acceleration values and variation, reducing accuracy of measured acceleration. Collar sizes S4 and S5 with collar weight W5 were likely to have an effect size five times larger than the collar combinations S1/2 and W1/2.

This study also measured the interaction between collar size and weight. We found that if the collar is fitted tightly (neck + 5 cm or 16% of neck circumference), collar weight will have minimal effects on measured acceleration, see Table 2. As the increase in VeDBA is larger for heavier collars, it is recommended that when deploying collars heavier than 0.8% of body weight, collar size should be kept to within 5 cm or 16% of the subject’s neck size. If the collar is looser (> neck + 5 cm), then weight will have a larger effect; therefore, if this size is used, researchers should consider using collars within 0.6% of the subject’s body weight. The fact that the effect of collar size was smaller than that of weight demonstrates that size is important to consider but, within 16% of neck circumference, is less likely to affect measured acceleration significantly. Changes in body mass must be considered when deploying collars, as many ungulates undergo substantial body mass changes seasonally, affecting collar tightness with the potential to cause harm to the subject [26, 29]. Therefore, collars should be deployed that are neither too tight nor too loose in proportion to the neck size of the subject to satisfy ethics and provide accurate measurement of relative body acceleration. Of course, this study does not address body mass changes; considerations in relation to this must be made for any given study species.

This study measured the effect of collar weight within 1.2% of the subject’s body weight. The commonly suggested ‘5% (or 3%) rule’ has been suggested to be arbitrary [24]. However, studies investigating the effect of logger weight have mainly focused on flying animals [30, 31]. On terrestrial mammals, it has been suggested that 2% is more appropriate for collar-attached devices [15], and even an increase from 0.4 to 0.6% in collar weight has been shown to affect the behaviour of a large ungulate [25]. This study shows that when using collar-attached devices to measure body movement, weight has an effect on measured acceleration. It is important to consider device weight for both measurement accuracy as well as considering its potential influence on behaviour.

The effect of collar size and weight also varied with walking speed: at low speeds, there was little variation between collar sizes and weights, but at higher speeds, variation increased [19]. Wild animals are likely to display a wider range of behaviours and larger accelerations than in our controlled experimental setting, depending on the species’ body size and characteristics of their body movement [10, 23]. The larger variation in these behaviours may be exaggerated by collars that are loose and/or heavy; therefore, in a normal setting, the observed effects of collar size and weight might be larger and more variable. It is, therefore, important to consider that this effect may be different in behaviours other than walking. Furthermore, when applying these recommendations to other species, it is important to consider other factors such as body shape and differences in movement, making deployment decisions based on the study species in question.


This study highlights that both the size and weight of the collar affect measured body acceleration in a quadrupedal terrestrial mammal. The importance of correctly fitting a collar for deployment of devices that record body movement, to ensure the highest standards of both welfare and scientific precision, is highlighted. We recommend that collar tightness be kept to + 16% of neck size and collar weight to be kept within 0.8% of the subjects’ body mass. If the collar is tighter than this + 16% guideline, collar weight will have less effect, and thus, heavier collars will not have a negative effect. We suggest consideration of these recommendations when deploying collar-mounted devices and that researchers should report the size and weight in research outputs to demonstrate best practise and enable similar studies to employ comparable and reproducible methods.


Study animals

The study took place in June 2018 at Belfast zoo, within the subjects’ normal housing area. Four female adult African pygmy goats were used (body weight: 13.5–23.1 kg; Table 4), as they are easily trained domesticated caprids amenable to the use of collars and harnesses. Using positive reinforcement, consisting of the subject’s normal daily food, individuals were trained to walk on a treadmill. Individuals were habituated to the equipment and then trained in stages over a period of 1 month. Each individual walked on a treadmill for a period of up to 10 min, at different speeds, wearing both a collar (Dog collar, Pets at Home, UK; 53 g) and a harness (Adjustable nylon dog harness, Bunty Pet Products, UK; 128 g).

Table 4 Details of each individual goat (G01–G04) used in the study, including body weight, neck size and age

Acceleration data collection

We used tri-axial accelerometers (Daily Diary tag, Wildbyte Technologies; [19]) that measured acceleration in three orthogonal axes at a sampling rate of 40 Hz. The devices were powered by a Li-Po rechargeable battery (Cameron Sino, HK, 31.54 × 25.13 × 3.95 mm; 5.5 g), enclosed in plastic casing (35 × 45 × 12 mm; combined mass 15.2 g) and attached using waterproof tape (Tesa® tape 4651, Tesa, Germany). One accelerometer was placed on the collar in the ventral position on the subject’s neck (Fig. 4); using different combinations of collar sizes, from approximately 3 to 30% (+ 1 to + 9 cm) larger than individuals’ neck sizes, and weights, from 0.4 to 1.2% (88 to 285 g) of individuals’ body masses (Table 4). Weight was added adjacent to the accelerometer using metal nuts taped to the collar, with the aim of resembling larger battery weight. A second accelerometer was placed on the harness in a ventral position, to act as a control in the same position as the first device. The harness provided a more rigid placement, minimising movement of the accelerometer beyond that of the study animal (Fig. 4). Collar combinations measured were each collar size with the smallest and largest weight, and each collar weight with the tightest and loosest collar (Table 1). After each individual was trained and acclimated to the equipment, each collar combination was tested on each individual once. Collar combinations that tested the largest and smallest collar size with the lightest and heaviest collar weight were measured. Goats walked at speeds between 0.8 and 3.0 km/h at increments of 0.2 km/h and walking speeds were recorded in a random order, for a minimum duration of 1 min. Each individual was video-recorded during data collection to remove data obtained during behaviours other than walking (Fujifilm XP120, Fujifilm, Japan; 59.94 frame/s). Sections, where the speed of the treadmill was changing (~ 15 s), were also removed.

Fig. 4
figure 4

Goat (G03) with an accelerometer attached to the collar (a) and to the harness (b)

Data were processed and analysed in R version 3.6.2 [32]. Data were labelled for the corresponding speed, collar size and collar weight. To calculate VeDBA, raw acceleration for each axis was used to derive static acceleration over a running mean of 2 s and thus the vectoral sum of dynamic body acceleration was calculated (VeDBA) [33]. Mean VeDBA measured over a 1s interval was calculated for the collar- and harness-attached devices. To determine how acceleration measured by the collar changes, the difference between the collar and harness attached devices was also calculated \(\left( {{\text{difVeDBA}} = \overline{{ {\text{VeDBA}}}}_{{{\text{collar}} }} - \overline{\text{VeDBA}}_{\text{harness}} } \right)\).

Data analysis

To understand the relationship between collar size and weight on measured acceleration (VeDBA), three separate general linear fixed effect models were used to (1) investigate the VeDBA measured on the collar-attached device; (2) investigate the VeDBA measured on the harness-attached device; and (3) investigate the relationship with the difference in VeDBA measured from the two devices (difVeDBA). The fixed variables included in the models were goat ID, walking speed and collar size and weight as categorical variables. A two-way interaction between collar size and weight was included. The two-way interactions between speed and the variables, goat ID, collar size and collar weight, were also included. The global models were simplified using the “dredge” function in the R package MuMin [34] which uses AICc to assess the best fit model [27]. Models within ΔAICc ≤ 6 were retained for inference and the simplest model was selected (Table 2 and Additional file 1) [35]. Model residuals were checked for a normal distribution.