Questionnaires and subjects

A standard questionnaire was developed to investigate the types of training technique used and the prevalence of undesired behaviours in a population of dog owners in the UK. Owners of multiple dogs were asked to only complete a single questionnaire, with respect to their youngest dog. The questionnaire, adapted from Blackwell et al. [2], was refined after piloting using a population of 15 dog owners in the Somerset area. The questionnaire was divided into four sections: owner demographics; dog demographics; information about training classes and training techniques used by owners with the focal dog; and information on the occurrence of a number of commonly reported undesirable behaviours in dogs.

Statistical analysis

Data were checked for coding and input errors, duplicates removed and implausible responses recoded as missing. The frequency and percentage of owners using electronic training methods were calculated. Most data were categorical. Age of dogs in months was not normally distributed but normalised by log10 transformation. Attendance at training classes was reduced to a 0/1 score, by including all dogs reported by owners as having attended the class for at least 4 weeks for all types of classes except for puppy classes, where attendance was scored only where owners reported attending for at least 2 weeks when their dogs were 12 weeks of age or less. The estimate of e-collar use was calculated as a percentage of respondents reporting use of bark activated, remote activated collars, and invisible boundary fence systems. Cases from Wales were excluded because these are no longer likely to be relevant given the recent ban of such devices. Cases from Scotland were also excluded because of small numbers of questionnaires distributed in this area.

In order to identify suitable cases and controls from the full dataset, the numbers of owners using e-collars for different reasons were identified. Electric boundary fence systems were not included due to low numbers reported, and their use in a specific context. Cases using remote and bark activated electric collars were combined for further analyses. Behaviours where remote devices were used in <20 cases were excluded, or combined into single variables were appropriate. Remaining categories where sufficient owners had used e-collars for analysis were recall problems /chasing behaviours combined, and barking. Comparison populations of owners who had specifically recorded using different training techniques for these behaviours were identified. To reduce the number of comparisons, individual training methods in comparison groups were combined into those which involved the application or removal of an aversive stimulus (positive punishment or negative reinforcement), and those which involved the application or removal of a rewarding stimulus (positive reinforcement or negative punishment). Comparison of multiple training methods was not considered appropriate for the questionnaire data since the relative extent to which training devices were used was not explored. It was considered more robust to use mutually exclusive groups in the use of training methods as it is more likely that the methods used were those predominantly chosen by the owner. Categories were made mutually exclusive by removing those cases where multiple techniques were used by owners for the same behaviour. As this left only a small number of cases where owners had exclusively used e-collars for barking problems (N = 14), further analyses were conducted only for different training approaches to recall / chasing behaviour. These cases all used remote activated e-collars.

A multinominal logistic regression analysis was used to investigate prediction of membership in each of the categories of training type (e-collar, other aversives, reward based), based on general factors such as owner and dog characteristics, attendance at training classes, occurrence of undesired behaviour (excluding recall / chasing behaviour) reported by owners, and the total number of behaviour problems shown by each dog calculated as a proportion of total number of behaviours possibly recorded on the questionnaire. Initially all potential risk factors were screened using univariable analyses, and only those showing significant difference between groups at P < 0.2 were included in the model. The contribution of individual components to the model was evaluated using the -2log likelihood. Individual variables were removed in a backward stepwise manner until the change in −2log likelihood in reduced models was less than expected for the associated degrees of freedom of the variable removed. The relative difference in risk as compared to reference categories was expressed as Odd’s Ratios.

Finally, a comparison was made of owner reported success of training using the training technique specified and the type of training used for recall training using cross tabulation and chi-square comparisons of groups.

Estimate of electronic training device use in England

14,566 questionnaires were distributed direct to dog owners, of which 3897 (27%) were returned completed and legible. From those remaining, distributed in England, 3.3% (n = 133) owners reported using remote activate e-collars, 1.4% (n = 54) reported use of bark activated collars, and 0.9% (n = 36) reported using electronic boundary fences.

Factors associated with e-collar use

Description of population

No statistical difference (Kappa Measure of Agreement) was found with respect to distribution between categories in this sub-sample of owners using specific training methods for recall / chasing as compared to the whole population of owners surveyed (n = 3897).

Multinominal logistic regression model for risk factors for training method used by owners for recall / chasing

Initial screening using univariable analysis resulted in the exclusion of: location type for questionnaire distribution, breed category of dog, cross breed or pure breed, age category of owner, origin of dog, attendance at gundog classes, attendance at ring-craft classes, age of dog, owner’s report of their level of dog owning and training experience, and all undesired behaviours except for house-soiling when the owner was out, waking the owner in the night and hiding from unfamiliar people.

Owner reported success of training in different groups

Owner reported success of using the training method was significantly different between groups (χ2 (2, N = 579) = 26.999, P <0.001) with those training using e-collars reporting less success than expected, and those training with rewards reporting greater success than expected (Figure 2).

Estimated prevalence of electronic training device use

The proportion of owners reporting use of electronic training aids is fairly low compared to other training methods, although extrapolation across the estimated UK dog population of 10 million [23] would suggest approximately 560,000 dogs trained with these devices. Should welfare implications arise from their use, therefore, the number of animals affected is considerable. However, some caution should be used in extrapolating these data across the UK. Data from Wales was removed due to the ban implemented in Wales [9] making these cases unlikely to remain relevant, and those from Scotland removed due to low numbers not supporting extrapolation. There were also some (non-significant at P < 0.2) differences between regions with respect to proportion of owners using e-collars. For example, a higher proportion in the East (7.3% cases) and North East of England (6.3% of cases) was found compared to London (1.7% cases), East Midlands (2.4% cases) and the South West (2.5%) cases. Since questionnaire distribution was not even across all regions it is possible that the overall estimate of prevalence may be affected by regional differences in use. Since a higher proportion of questionnaires were distributed in the South West where reported e-collar use was lower, regional effects may mean that the figure here is an under-estimate of overall UK use of devices.

Although questionnaires were distributed in as wide a range of environments as possible, this is not a random sample, and likely to have sampling biases which are difficult to quantify. For example, it is possible that the types of owners choosing to use electronic training devices may be more or less likely to be represented in the populations sampled, or may be more or less likely to complete and return questionnaires. These figures of e-collar use should therefore be regarded as an estimate, although they are analogous with figures published by the Electronic Collar Manufacturers Association who estimate 500,000 collar owners in the UK [24].

Risk factors for use of remote activated e-collars

It is interesting that male owners were more likely than females to use e-collars compared to reward based methods for training their dog for recall or chasing problems. This may relate to gender differences in willingness to admit to e-collar use, or attitudinal differences to training techniques selected. Bennett and Rohlf [25] found that male owners were more likely to report that their dogs were ‘disobedient’ than females and so the increased use of e-collars reported by males in this study may reflect differences in attitudes towards potentially problematic behaviour. It is also possible that dogs show behavioural differences with owners of different genders [26].

In this study, reward based methods were more likely to be used by owners who had not attended agility classes. This may reflect preferences for training method use amongst proponents of this activity. However, further research is needed to investigate causality in this relationship, as it may reflect an attempt to resolve behavioural issues by increasing the dogs structured activities/mental stimulation/exercise by owners who also select to use e-collars.

Although there is no consensus in the literature regarding the influence of attendance at formal training classes on undesirable behaviour, a number of studies have suggested a reduction in problematic behaviour following attendance at obedience classes [27–30] and it seems inevitable that attendance at training classes, recommendations by trainers and observation of training methods used are likely to influence the subsequent selection of training methods by owners. It is therefore important that those running training classes have knowledge of the appropriate use of different training techniques and an understanding of the possible implications of their use.

Both Christiansen et al. [10] and Hansen et al. [31] suggest that different breeds of dogs differ in the extent to which chasing behaviour is motivated, efficacy of this type of training, and the level / number of applications of an aversive stimulus to modify behaviour. Here, breed type did not vary between groups, nor with owner reported success between groups. In the previous research relatively small numbers of animals were used, and of types unrepresented in this survey, making comparison difficult. The dogs selected for testing by Christiansen et al. [10] may not necessarily be representative of breed types reported here. Hence, whilst data here suggest that breed is not a strong factor in the choice of training method used for chasing or recall problems, further research is needed to investigate potential breed differences in response to different training methods.

Christiansen et al. ([10]) also suggest that the number of stimuli given to individual dogs related to a factor derived from tests considered to be related to ‘predatory motivation’. Specific behavioural characteristics such as predatory drive were not measured in this study, although owner reports of the number, occurrence of described undesired behaviours, and whether these were regarded as a ‘problem’ were measured and did not differ between groups.

No difference in age of dogs was found between groups in this study, although it has been reported that more coercive training techniques are used in older search and rescue dogs [32]. Differences between study findings may also reflect geographical differences.

Proportion of total variance explained by model

It is salient that only between 8.4 and 10.1% of the variance between training methods used for recall or chasing problems is explained by the variables measured in this study. Hence, approximately 90% of the difference between categories is due to other factors, not measured here. It is possible that differences are explained by attitudinal factors in owners, their previous experience of different training methods, differences in relative ability to effectively apply and time interventions, and / or specific information or advice received from others, although further research is needed to investigate these factors further. In addition, the severity of the recall / chasing behaviour may have varied between groups. Although owners were asked if they considered the behaviour a ‘problem’, this is subjective, and not necessarily indicative of severity. Indeed, most owners with recall / chasing problems considered this to be ‘problematic’, presumably because behaviour of this type causes interruption of their daily routine, irritation or embarrassment.

Owner reported success of training techniques for recall / chasing problems

A higher proportion of owners who had used reward based methods for recall / chasing problems reported success with their training. Although this may reflect increased efficacy when trainers use reward based methods, there are potential confounding effects in this comparison. For example, the relative training abilities of owners may differ with type of training method, there may be different perceptions of ‘success’ between different groups, or there may be differences in the initial severity of the problem for which different types of training method are selected which could affect outcomes. There is mixed evidence for relative efficacy from previous studies, although there is overall support for the conclusion that efficacy of electronic training devices is no greater than use of other methods.

In a population of owners attending a clinical behaviour service, owners reported the types of training techniques previously used and their perceived success [5]. A higher proportion of owners using reward based methods judged these to have had ‘positive effect’ and fewer ‘negative effect’ than those using more coercive methods. However, relatively few owners had used bark or remote activated e-collars. Eleven owners reported the use of such devices as having a positive effect, 6 a ‘negative effect’ and 6 ‘no effect’, although it is unclear for which behaviours these were used for, and the extent to which these behaviours were comparable to the use of other training techniques. In an observational study, Jones [33] investigated the use of e-collars to train dogs to stop attacking kiwis in a wildlife preservation programme. Thirteen dogs from a local pound underwent training, with the use of a stimulus as they approached a kiwi. However, on subsequent testing in a different context only one dog avoided approaching the kiwi.

Christiansen et al. [34] found a reduced likelihood of attacking sheep in a pen environment in which training with an e-collar had taken place the previous year (only 1 of 13 dogs which needed an intervention the previous year needed further training). However, no difference was found between dogs which had been given a stimulus and those which had not in a subsequent ‘path’ test, where dogs were presented unexpectedly with a lone sheep in a different context. Owners of all dogs in this study reported a reduced inclination to chase sheep, this was not influenced by whether they had received an electrical stimulus or not the previous year. According to owner report, only 1 of the 13 dogs given an electric stimulus a year previously had reduced or no interest in sheep, the rest reported as having no change. These findings may indicate that effects of e-collar training are not necessarily generalised: in other words the dog may respond as trained when in the specific context in which training has taken place, but retain the chasing behaviour in other situations.

In comparing the use of electronic bark collars with those using a citronella spray Juarbe Diaz and Houpt [35] reported that owners found the latter to have greater success at reducing unwanted barking. In addition, owner perception was that the e-collars were less humane to use on their dogs, although this may have had an impact on their evaluation of efficacy.

Christiansen et al. [10] found that dogs which had never seen sheep before had an increased chance of attacking sheep in a confrontation test compared to those which had experienced sheep – this may suggest that more dogs which chase sheep are those that are naïve to sheep rather than being established chasers. Apart from highlighting the importance of preventing such behaviour through careful introduction of puppies to livestock, dogs which chase through novelty / excitement may have their behaviour modified more easily than those with well-established chase responses. Nevertheless, apparently regardless of the extent to which the response is established, CABTSG [15] suggest that other training methods can be successfully used in those situations where e-collars are purported to be of greatest value (e.g. livestock chasing) and that successful resolution is regularly achieved by qualified individuals.

Some literature also compares the perception of owners more widely regarding the relative success of reward based and more coercive methods of training. For example, Loftus et al. [36] reported that across a range of undesired behaviours, owners reported reward based training as ‘more successful’ than other methods. Bussey [37] conducted an investigation of methods used in obedience training at a time when use of reward based training approaches were relatively new in this discipline. She suggested that dogs were no less successful where owners used reward based training rather than more traditional techniques, and that use of a fixed collar rather than choke / check chain had a positive influence on success.