1. Introduction
Domestic dogs are thought to have evolved from wolves around 25,000 to 40,000 years ago [
1]. This makes dogs the earliest domesticated animal, predating cattle, pigs, and sheep who are thought to have been domesticated around 10,500 years ago [
2]. A recent large-scale comparison of DNA in ancient and modern dogs reveals a complex history of divergence and admixture in the late Pleistocene to early Holocene; according to the authors, these complex migratory dynamics obscure a clear locus of origin for domestic dogs, but nevertheless they suggest that dogs likely evolved, originally, from a single wolf population or, less likely, multiple closely related wolf populations [
1].
Dogs are currently the most common household pet, with 470 million worldwide [
3], but they have also been used for working purposes since they were first domesticated, from initially helping humans with hunts to now working as search and rescue, police, herding, and assistance dogs. This means that dogs play an important social role and spend a disproportionately large amount of time around humans when compared with other species. Dogs are thought by some to have developed such human-like social skills as understanding the referential intent of deictic gestures (e.g., pointing) via the process of domestication. This was initially hypothesised by Hare and Tomasello [
4], who noted the tens of thousands of years that dogs and humans have spent together, which allows substantial opportunity for co-evolutionary processes to take place. The process of domestication has been shown on a small scale by the ”The Silver Fox Experiment” (reviewed by Dugatkin [
5]). Researchers selected a group of foxes (
Vulpes vulpes) with non-aggressive traits and found that within six generations, the foxes showed signs of a domestication syndrome: including neotenous morphological changes such as floppy ears, and shorter, more rounded snouts, plus behavioural changes such as reduced levels of stress hormones [
5], but see [
6], for an argument rejecting the characterisation of these foxes as ”domesticated”. These foxes also outperformed non-“domesticated” foxes in following human communicative gestures [
7].
The Domestication Hypothesis holds that dogs are born with the ability to understand human communication gestures. This was supported by Riedel and her colleagues [
8] who found that dogs at 6, 8, and 16 weeks of age could all follow human communicative cues; because the puppies at 6 weeks were still being raised by their mothers, this suggests an innate ability provided through a history of domestication. Further support for the Domestication Hypothesis is that wolves (
Canis lupus), despite having the same ancestors as dogs, do not show the same ability when it comes to understanding human communicative gestures (but see [
9] and below). Ujfalussy and their colleagues [
10] compared hand-raised and mother-raised dogs to hand-raised wolves—both groups of dogs performed better than the wolves despite having similar socialisation. Similarly, Lazzaroni and her colleagues [
11] compared pet dogs, free-ranging dogs, and research dogs to hand-raised wolves (the latter two groups housed and raised in a research facility), finding that dogs approached humans more for food and a cuddle (i.e., food was not the only motivator); wolves approached only for food and were slower to do so, despite being raised in the same environment as some of the dogs tested. Lampe, Bräuer, Kaminski, and Virányi [
12] compared similarly raised wolves and dogs living in packs to pet dogs and found all groups of dogs outperformed wolves on a variety of cognitive tasks. They hypothesised that pet dogs rely more on humans to make decisions as opposed to their wild counterparts, therefore, domestication has allowed them to develop an understanding of non-verbal human communicative gestures to make this easier. Dogs are willing to wait longer for rewards than wolves [
13].
A common test of animals’ abilities to follow human cues is the object-choice task. In these studies, an experimenter hides a reward inside an opaque container and the subject is then presented with the container with the reward and one or more empty containers. The subject can only choose one container and will receive the reward if they choose the correct box. To indicate the correct container, a communicative gesture is used. The gesture is most commonly a type of point, however, gazing cues are also used. Anderson, Sallaberry, and Barbier [
14] were the among the first to use the object-choice task with non-human primates. They found that capuchin monkeys (
Cebus capucinus) could follow the cue only when pointing was used and failed on a gazing-only condition. Shortly later, Miklósi, Polgárdi, Topál, and Csányi [
15] used the object-choice task on the first set of dog subjects; they found that dogs were able to reliably follow a point to find the rewarded container. Domestic dogs have also been found to follow human cues even when the cue is counterproductive, as some dogs chose empty cups, over cups with rewards [
16]. Dogs have very powerful olfactory senses, and they will use this sense to find rewards in object-choice tasks [
17]; however, when human cues are provided, dogs have been reported to privilege these cues over their own senses [
18,
19].
Dogs are not the only domestic animal to exhibit this trait as when tested, both dogs and cats (
Felis catus), chose a baited container over an empty one if no cue was given but would choose the container, they saw a human approach, indicating that they are influenced by human actions [
20]. However, even though dogs may show trust in humans and will initially follow counterproductive cues, they have been found to be less likely to follow a point once a human has shown to be misleading [
21].
It is frequently claimed that dogs out-perform apes in paradigms such as the object-choice task, and this has been attributed to their different selective histories [
22,
23], but, as noted by Clark, Elsherif, and Leavens [
9,
24,
25,
26,
27,
28], to date, dogs and non-human primates have not been matched in testing contexts, task preparation, life history stage, or either quality or quantity of experience with humans. Therefore, as Udell and her colleagues noted [
28] (p. 717), “[d]ogs’ sensitivity to human gestures may not be entirely explained by phylogenetic variables; rather, the interactions between genetic, developmental and experiential variables must be considered”.
Numerous studies have reported facilitating effects of training on dogs’ performances in a variety of object-choice tasks, demonstrating, among other things, that dogs can rapidly adapt to novel pointing postures [
28,
29]. McKinley and Sambrook [
30] compared actively working gundogs to pet dogs and found superior performance by the highly trained dogs. Interestingly, they found, on average, gundogs of traditional gundog breeds outperformed gundogs from non-gundog breeds, this may suggest that the traditional breeds are aided by a genetic predisposition to perform better. Wobber, Hare, Koler-Matznick, Wrangham, and Tomasello [
31] reported that working breeds are more skilled at using pointing gestures than non-working breeds. They also found that primitive breeds with little human contact and not bred to be pets or to perform working roles, such as the New Guinea Singing Dog, were able to follow human cues but were outperformed by pet dogs on the object-choice task. Udell and her colleagues [
32] compared point-following performance on hunting, herding, and livestock-guarding dog breeds, using Airedale Terriers, Border Collies, and Anatolian Shepherds, respectively, reporting that Border Collies and Airedale Terriers performed above chance level when following a momentary distal point, and that, after training, Anatolian Shepherds were also able to perform above chance level, which is not typical of the breed. Thus, life history factors seem to matter in dogs’ performances in human-canine interactions, and we sought to examine the broad effects of training by comparing actively working dogs with dogs of both non-working breeds and working breeds (see
Table 1 notes for definitions).
The types of interaction dogs have with a cue-giver is a potential factor in dogs’ performances in formal assessments. For example, dogs prefer receiving petting to vocal praise and do not distinguish between owners and strangers during these activities [
33]. A similar finding was reported by Hegedüs, Bálint, Miklósi, and Pongrácz [
34], who asked owners to encourage their dogs to select the correct box when an experimenter was pointing to incorrect one; they found that owners had no effect and suggested this was due to the experimenter calling the dog’s name and therefore, the experimenter and dog were engaged in communication. Thus, the effects of familiarity of interactant seem to be moderated by situational factors, such as the quality of interaction.
We were interested in whether actively working dogs were advantaged in the use of human cues by the cue-giver’s familiarity (owner/trainer vs. experimenter) and breed/working classification. We therefore compared an actively working group of dogs with pet dogs from working-dog breeds and pet dogs not from working dog breeds on an object-choice task, in which the owner/trainer or the experimenter provided cues to toy locations. We reasoned that if actively working dogs displayed an advantage over pet dogs from working dog breeds, that this would implicate a facilitatory effect of experience over breeding, and that if both of these groups of dogs outperformed pet dogs from non-working breeds, this would implicate an influence of breeding on performance. Finally, we elected to use familiar and unfamiliar toys in this study, for two reasons: (a) to avoid the potentially distracting effects of food, e.g., Ref. [
35] and (b) because of the relative ease in manipulating toy familiarity, compared with food familiarity, during pandemic lockdown conditions.
We set out to answer three questions: (1) Are there systematic differences between dogs of different life histories in performance in the object-choice task? (2) Are there systematic differences between dogs of different life histories in their responses to familiar and unfamiliar cue-givers in the object-choice task? (3) Are there systematic differences between dogs of different life histories in their responses to familiar and unfamiliar toys in the object-choice task?
Table 1.
List of breeds, age, sex, and working role for each subject.
Table 1.
List of breeds, age, sex, and working role for each subject.
Subj. ID | Breed | Sex | Age (M) | Working Role |
---|
Actively Working a |
A1 | Jack Russell Terrier X Spaniel | M | 36 | Search and Rescue |
A2 | English Springer Spaniel | F | 96 | Search and Rescue |
A3 | Scottish Terrier | F | 23 | Therapy dog |
A4 | German Shepherd | F | 12 | General purpose police dog |
A5 | Labrador | F | 13 | General purpose police dog |
A6 | Labrador | F | 7 | General purpose police dog |
A7 | Cocker Spaniel | M | 84 | Active drug dog |
A8 | Springer Spaniel | M | 36 | Active explosives dog |
A9 | Working Cocker Spaniel | M | 72 | Beating and agility |
A10 | Springer Spaniel | M | 37 | Victim recovery |
A11 | Labrador X Springer Spaniel | M | 10 | Assistance dog |
A12 | Cockapoo | F | 36 | Assistance dog |
A13 | Labrador | M | 108 | Assistance dog |
A14 | German Shepherd | F | 8 | General purpose police dog |
A15 | Working Cocker Spaniel | M | 120 | Agility |
A16 | Golden Retriever | F | 120 | Agility |
A17 | Working Cocker Spaniel | F | 36 | Agility |
A18 | Border Collie X Welsh Sheepdog | F | 96 | Scent work |
A19 | Working Cocker Spaniel | M | 60 | Agility |
Non-working b |
B1 | Pug | F | 24 | N/A |
B2 | Unknown mix | M | 72 | N/A |
B3 | French Bulldog x Pug | F | 36 | N/A |
B4 | Unknown mix | F | 36 | N/A |
B5 | Maltese | F | 65 | N/A |
B6 | Unknown mix | M | 12 | N/A |
B7 | Unknown mix | F | 7 | N/A |
B8 | Cavalier King Charles Spaniel | F | 120 | N/A |
B9 | Shih Tzu X | M | 8 | N/A |
B10 | Shih Tzu X | F | 72 | N/A |
B11 | Unknown mix | M | 144 | N/A |
B12 | Shih Tzu | M | 96 | N/A |
B13 | Chihuahua | F | 12 | N/A |
B14 | Toy Spaniel X Toy Poodle | M | 11 | N/A |
B15 | Bichon Frise X Shih Tzu | F | 96 | N/A |
B16 | Unknown mix | F | 96 | N/A |
Working breeds c |
C1 | Cocker Spaniel X Jack Russell | M | 12 | N/A |
C3 | Cockapoo | F | 60 | N/A |
C4 | Whippet | F | 10 | N/A |
C5 | Show Cocker Spaniel | F | 36 | N/A |
C6 | English Springer Spaniel | M | 36 | N/A |
C7 | Working Cocker Spaniel | M | 48 | N/A |
C9 | Bull Lurcher | M | 16 | N/A |
C10 | Shar-pei X Cane Corso | M | 24 | N/A |
C11 | Labradoodle | F | 6 | N/A |
C12 | Cockapoo | M | 36 | N/A |
C13 | Rottweiler X German Shepherd | M | 81 | N/A |
C14 | Cockapoo | F | 96 | N/A |
C15 | Cane Corso | F | 16 | N/A |
C16 | Dachshund | M | 72 | N/A |
C17 | Border Collie | M | 11 | N/A |
C18 | Staffordshire Bull Terrier | M | 9 | N/A |
C19 | Hungarian Vizsla | M | 108 | N/A |
C20 | Hungarian Vizsla | F | 22 | N/A |
C21 | Dachshund | F | 8 | N/A |
2. Materials and Methods
2.1. Participants
A total of 56 domestic dogs (
Canis familiaris) served as subjects in this study. The subjects were divided into three groups, these were “Actively working”, “Non-working breeds”, and “Working breeds”. For the statistical analyses, 54 out of 56 dogs tested were analysed. Two dogs were excluded from the “Working breed” group (one male and one female) due to experimenter error: an unequal number of points were displayed to the familiar and unfamiliar items. This left the “Working breed” group with 19 subjects to be analysed (see
Table 1 and notes for elaboration of dog categorisations).
The “Actively Working” dogs were breeds recognised by the American Kennel Club (AKC; [
36]), and these were dogs which had been trained for a specific purpose. There were 19 subjects in this group, this included 9 males, 10 females with the age ranging from 7 months to 10 years (mean = 3.75 years).
“Non-working breeds” were dogs who were kept only as pets but not breeds belonging to any of the working groups. These breeds belonged to the toy dog group as defined by the AKC [
36]; this group also contained (a) mixed breed dogs, where at least one of the dogs’ parents was from the toy group and the other of unknown breed, and (b) dogs of unknown breed. There were 16 subjects in this group, this included 6 males, 10 females with the age ranging from 7 months to 12 years (mean = 4.72 years).
“Working breed” dogs were also kept only as pets and included dog breeds from the herding, hound, sporting, non-sporting, terrier, and working group as defined by the AKC. This group also included mixed breeds when both breeds were from working groups. These dogs had not been trained to work a specific role. There were 19 subjects in this group which included 12 males, 9 females with an age range of 6 months to 9 years (mean = 3.10 years).
All dogs used in this study were over 6 months of age and had lived with their owners for a minimum of 2 months. For the full list of breeds, see
Table 1. No pre-training was given to the dogs, and food rewards were not used. The participants were given a bag of dog treats for taking part in the experiment after the event.
2.2. Materials
The dogs were tested in Pevensey Dog Laboratory (2.5 m in length and 4.55 m in width), to ensure standardisation in the testing context (see
Figure 1). The subjects were filmed on a Sony HDR-PJ410 Handycam video camera that was set up prior to the subjects entering the room. A questionnaire was used to assess the subjects’ social experience, age, lifestyle habits, and training level to see if this affected the results (see
Appendix A). This involved questions such as: On a scale of 1 to 10 how well does your dog follow commands?
2.3. Design
A repeated measures design was used, with two independent variables: the familiarity of the cue giver (familiar, unfamiliar) and the familiarity of a toy (familiar, unfamiliar). The first author served as the unfamiliar cue-giver and the dog owner served as the familiar cue-giver. We asked the owners to bring in a familiar item for the dogs (i.e., favourite tennis balls/toys). The unfamiliar items were a novel stuffed toy or tennis ball. The dependent variable was whether the subject (a) chose the correct referent, (b) chose the incorrect referent, or (c) made no choice. Each subject took part in all conditions of this study.
2.4. Procedure
The dogs were allowed off lead in the laboratory, with the owner and experimenter present to allow the dog to feel comfortable in the room. To ensure the dogs were not under stress, they were required to willingly approach the experimenter and eat from their hand. If the dogs apparently did not want to do this, then they would not take part in the experiment (following the protocol in Ref. [
37]). However, no subjects were excluded on this basis. Once the dog met the inclusion criteria, the study began. Both the owner and the experimenter served as cue-givers, in counterbalanced order across the dogs. The cue-giver (either owner or experimenter) guided the dog to the starting position, which was marked on the floor of the laboratory. The human not acting as the cue-giver stood silently in a central position behind the dog and held the dog in place until the cue-giver arrived in the correct position. Commands such as ”sit” and ”stay” were used to keep the dog in the starting position. Once the dog and cue-giver were in their starting positions, ostensive cues such as calling the dogs’ name to attract their attention were used. The cue-giver pointed to one of two items (see
Figure 1,
Figure 2 and
Figure 3); this was either an item that was familiar to the dog, or an item that was unfamiliar.
Each dog received eight trials, four with the owner as the cue giver and four with the experimenter as the cue-giver, so that each cue-giver pointed to each object once. The order in which the cues were given (left or right) were counterbalanced to reduce a response pattern bias forming, as well as the order in which the familiar and non-familiar cue giver gave the cue. All subjects received the same type of cue, the cue-giver used an ipsilateral static point, whereby the pointing hand on the same side as the intended target was in place before the participant viewed the cue and remained so until the participant chose one of the items. The cue giver kept their head directed centrally, and body position facing forward, and the cue-giver looked at the dog whilst giving the cue, to indicate the cue is meant for the dog.
If the dog did not make a choice or approached one of the two items after 1 min this was classed as a no-choice trial (as seen in similar protocols: [
29,
37]). A successful trial was when the dog chose the correct referent by touching it, retrieving it, or coming within 10 cm of it. An unsuccessful trial was when the subjects approached the incorrect item. During the experiment, our plan was to exclude a dog if they exhibited urination, defecation, destructive behaviour, or excessive vocalisation (these are signs of anxiety), however, no dog was excluded on these grounds.
2.5. Spontaneous Variations on the Form of the Pointing Cue
During the initial coding, we noticed that some owners/trainers did not point with their index fingers, as had been displayed by the experimenter, and this non-canonical pointing seemed to be more frequent in the trained working dog group; therefore we added a code for conformity with an index-finger point to determine the distribution of these points among the groups.
2.6. Statistical Analyses
Tests of normality showed the performance data were not normally distributed and so we elected to test our statistical hypotheses with nonparametric statistics. Kruskal–Wallis H tests were used to compare the three groups on the dependent variable and Mann–Whitney U tests were used for the pairwise comparisons between groups. Spearman’s rho was used to assess the relationship between overall performance and the results of the survey, and Cochran Q tests were used to measure the frequency of no-choice trials, and Wilcoxon signed ranks tests were used to compare two related samples.
2.7. Interobserver Reliability
All eight trials from eight subjects were randomly selected from the corpus and independently coded for performance by a naïve coder, who was blind to the hypotheses under test; this represents ~15% of the corpus. Reliability for dogs’ performances was Cohen’s kappa = 0.74, indicating excellent inter-observer reliability.
2.8. Ethical Note
All subjects were volunteered by their owners and were recruited through posts on social media. The dogs remained with their owners throughout this study, and the owners gave written consent for their dog to partake and to be filmed prior to this study; they were also reminded that should they wish to they could withdraw their participation at any time and leave. This study was conducted under ethical approval from the Animal Welfare and Ethical Review Body (AWERB) of the University of Sussex.
2.9. Risk Assessment
This study took place during the COVID-19 pandemic, therefore we needed to ensure the safety of the participants during testing. All participants wore a mask and remained a social distance of two metres when possible. The laboratory was sanitised between each participant and spare gloves, masks, and antibacterial hand gel were available. The experimenter was tested regularly for COVID-19 and all participants were invited to watch a video detailing the university’s COVID procedure.
4. Discussion
Our results show that actively working dogs and working breeds perform better than non-working breeds on the object-choice task. These findings are consistent with those of McKinley and Sambrook [
30] and Wobber and her colleagues [
31], who found that actively working dogs perform better than pet dogs. These findings are also consistent with those of Udell and her colleagues [
32], who found that traditional working breeds perform better in the object-choice task. Similarly to McKinley and Sambrook [
30], we compared working breed dogs kept as pets to non-working breed dogs kept as pets and found that working breeds perform better; however, unlike McKinley and Sambrook, we did not find a statistically significant difference between actively working dogs and working breeds kept as pets. Therefore, our findings are consistent with the idea of a breeding-related advantage for working dogs, although it should be acknowledged that we did not equate the pet working dogs and the pet dogs from working dog breeds on the quality or quantity of their pre-experimental interactions with their owners. Because there was no statistical difference in performance between the “Actively working” and “Working breed” groups, and both of these groups outperformed the pet dogs from non-working dog breeds, we conclude that breeding may play a role in performance in this task environment, given our sampling sizes and sampling environment. However, as emphasised by an anonymous reviewer, social experiences with humans are also clearly relevant to performance in understanding human communicative cues (e.g., [
28,
37,
38]).
We also found that the familiarity of the cue giver has a significant effect on overall performance, such that actively working and working breed dogs respond better to cues given by their owners than by a stranger. However, no such effect was found in non-working breeds kept as pets. These findings are only partially congruent with those of Cunningham and Ramos [
39], who reported that all the dogs they tested preferred their owners over a stranger; however, the breed sampling in their study differed from ours, as all dogs in their pet and highly trained categories were of working breeds, with no non-working breeds used.
The present results also indicate that previous studies conducted on dogs without their owners present may not have reflected the dogs’ true abilities to follow human cues, as suggested by Burani, Barnard, Wells, Pelosi, and Valsecchi [
40]. Our results also support the findings of previous studies that have found that dogs largely react better to their owners than a stranger [
17,
41]. However, our results are not consistent with those of Scandurra et al. [
42] and Lazarowski et al. [
43], who found that actively working dogs responded just as well to a stranger as they did to their owners. In fact, we found the opposite results: in our study, non-working breed dogs kept as pets responded to strangers and owners equally. The difference in findings in this case may lie in the types of actively working dogs tested. Scandurra et al. [
42] exclusively used water rescue dogs and Lazarowski and her colleagues [
43] only used detection dogs in their studies; these are both types of jobs which require the dog to act on their own and not rely on a human to guide them; whereas, in our study, we used a wide range of dogs which included detection and search and rescue dogs, but also included assistance and therapy dogs who rely much more on owner interaction for their working role. We also wanted to find out whether the familiarity of the object being pointed to would affect the performance of the dogs. However, we found no significant effect of referent familiarity for all three groups on overall performance.
We then moved on to analyse the results of the survey given to the owners prior to the experiment. We found no significant correlation between the degree to which owners said their dogs get on with other dogs or the frequency the dogs play with toys and overall performance. However, we did find significant positive correlations between the degree to which owners say their dogs are well-trained and performance in this object-choice task, and we found significant positive correlations between the degree to which owners say their dogs follow commands and performance in this object-choice task. This suggests that dogs who are perceived by their owners to be well-trained do perform better on the object-choice task as these dogs may have received more training and could be perceiving the point as a command to be followed. Interestingly, we found a positive correlation between the degree to which dogs were food motivated and overall performance; because we did not use food rewards in this study, this was a surprising result. However, because most dogs are trained with food rewards, those who are more food motivated may have expected a treat once they followed the point.
The survey was measured on a 10-point scale for most of the questions and an ordinal scale for others (Yes/Sometimes/Usually/No). Because the owners were answering the survey to characterise their dogs, it could have been subjected to a variety of response biases. For example, the question “How well trained would you say your dogs is?” could be interpreted in a number of different ways pertaining to the dog’s general demeanor. Still, several items did positively correlate with the dogs’ actual performance, suggesting that owners do have a fairly good grasp of their dogs’ performance potential.
We then examined the no-choice trials in terms of frequency and changes over time. We compared the overall performance for each group and found they did differ significantly, by group classification, in the number of no-choice trials. We found that “Non-working breeds” had significantly more no-choice trials than “Actively working” and “Working breeds”, and that “Actively working” and “Working breeds” did not differ from each other significantly in the number of no-choice trials. This shows that “Actively working” and “Working breeds” are more motivated to participate in the point-following task. When we look at the exact number of no-choice trials, we can see that “Actively working” dogs nearly always made a choice, with only three no-choice trials out of 152 total trials. Therefore, we can conclude again that “Actively working” and “Working breeds” may benefit from a genetic predisposition to not only perform better in the object-choice task but to be more motivated than non-working breeds.
We wanted to see if no-choice trials were increasing over time, to indicate whether the dogs were either losing interest over time (i.e., if these no-choice trials were increasing) or, alternatively, if the dogs were learning what to do (i.e., if the no-choice trials decreased over time). However, we found no significant effect of the passage of time; all three groups remained motivated at seemingly constant levels throughout the experiment.
This study has a number of limitations. First, this study took place during the COVID-19 pandemic, which means that all participants and the experimenter were required to wear masks for the duration on the study. Because the masks cover the lower half of the face this may have been confusing to the dogs who are not used to their owners interacting with them in this way. However, because this study took place in the United Kingdom, in May and June of 2021, it seems likely that the dogs will have had some interaction with people wearing masks, previously. However, it would be interesting to complete this study again with no masks used, when it is safe to do so, to see if the same effects are found. Second, we are grateful to an anonymous reviewer for pointing out that dogs of non-working breeds tend to be smaller than working dog breeds, and therefore our specific findings of poorer performance by non-working dog breeds is potentially confounded with body size, which we did not measure—the reviewer pointed us towards McGreevy et al. [
44] for further information about the effects of size on performance in dogs. Third, we thank an anonymous reviewer for pointing out that dogs’ individual preferences for the toys were not assessed—theoretically, this means that, at the group level, rather than failing to demonstrate that dogs displayed no apparent bias for or against the toys on the basis of their familiarity, it could be the case that relatively equal numbers of dogs displayed preferences for the familiar and unfamiliar toys, respectively, with results cancelled out at the group level. Finally, our sample size is relatively modest, which suggests that one possible explanation for some of our findings that diverge from other similar studies might be attributable to sampling error.