Do cognitive traits associate with everyday behaviour in the domestic dog, Canis familiaris ?

Even though a large amount of research has focused on testing cognitive traits of dogs, we still know almost nothing about whether these traits associate with dogs' daily behaviour outside the test situation. For these tests to have applied value it would be bene ﬁ cial that they re ﬂ ect dogs' behaviour in a broader sense rather than simply during the test situation. Our aim was to deepen our understanding of what cognitive traits can tell us about dogs' behaviour in their daily lives with their owners. We used two validated owner-completed questionnaires, the Canine Behavioural Assessment and Research Questionnaire (C-BARQ) and the Dog Impulsivity Assessment Scale (DIAS), to determine daily behavioural traits of dogs. A cognitive test battery was used to measure cognitive traits. We used linear mixed models for analysis ( N ¼ 987) and accounted for control variables such as sex, breed, age and training level of dogs. We found that high inhibitory control in the Cylinder Test was positively associated with train-ability and negatively associated with impulsivity (DIAS). A shorter latency to solve the V-detour task was positively associated with management problems, and dogs that were unsuccessful at the Logical Reasoning task were rated as slower at learning new tasks. Human-directed behaviour during the Un-solvable Task was associated positively with trainability and stranger-directed fear, and negatively with impulsivity (DIAS) and management problems. Our ﬁ ndings suggest that cognitive tests can help us to predict and explain dogs' behaviour, personality and behaviour problems. These results provide important

Dogs are often valued primarily based on their ability to adapt to life with humans.Some cognitive traits can facilitate or hinder the suitability of dogs for various working roles (Bray et al., 2017;Lazarowski et al., 2019Lazarowski et al., , 2020;;MacLean & Hare, 2018;Tiira et al., 2020), and it is possible that the same is true for pet dogs.For pet dogs, cognitive abilities such as inhibitory control and social cognition may be especially important, since they are required to continuously inhibit prepotent responses (e.g.chasing animals, jumping on tables, stealing food, etc.) in favour of cooperating with their human family members (Fagnani et al., 2016;Imponen & Pelliccioni, 2021;Marshall-Pescini et al., 2008).Cognitive tests could therefore help us understand traits that can potentially have implications for a wide range of real-life situations, ultimately impacting the wellbeing of both dogs and their owners.
Even though there has been extensive scientific research on canine cognition, research on the relationship between test results and dogs' everyday behaviour is still very limited, and there is often no agreement between researchers on what specific cognitive tests are measuring (Kabadayi et al., 2018;Kaminski & Nitzschner, 2013;Mendes et al., 2021).For tests to have applied value it would be useful if they reflected dogs' behaviour in their daily life as well as during the test situation.
Although several studies suggest a link between certain cognitive and personality traits, the results are largely patchy and inconsistent.In addition, most animal studies have included only small sample sizes (Dougherty & Guillette, 2018), using only a limited number of behavioural tests.Most animal research has focused on the association between exploration and learning, whereas evidence for associations between a wider variety of personality traits and cognitive paradigms is still lacking (Carere & Locurto, 2011).As discussed by Griffin et al. (2015), in order to develop a more comprehensive understanding of how personality and cognition correlate, we need to look at a larger number of cognitive and personality tests, with larger sample sizes.
The aim of the current study was to deepen our understanding of cognitive tests and what they can tell us about dogs' behaviour in their daily lives with their owners, outside the experimental setting.For this reason, validated owner-completed questionnaires were used for collecting information on everyday behaviour from dogs that had earlier taken part in a smartDOG (Tiira, 2016) cognitive test battery.These questionnaires have been found to correlate well with behavioural (Clay et al., 2020;Dalla Villa et al., 2017;Hsu & Serpell, 2003;Krichbaum & Lazarowski, 2022;Mongillo et al., 2019) and physiological measures (Wright et al., 2012), and to show good testeretest reliability over time (Brady et al., 2018;Duffy & Serpell, 2008;Riemer et al., 2014;Wright et al., 2011), providing a valuable measure of consistent behavioural traits.In addition, to further increase our understanding of the correlates of cognitive test performance, we also investigated possible associations between various cognitive and behavioural (shynesseboldness) test measures.Further, since consistency of traits is an important prerequisite for investigating correlations between traits, and few studies have investigated the consistency of cognitive traits (Griffin et al., 2015), we also measured the temporal repeatability of the cognitive tests included in our study.

Subjects
A total of 4383 dogs took part in the smartDOG cognitive test battery between 2016 and 2022.The test battery was developed by author Katriina Tiira, based on previous research.In order to participate, dogs were required to be motivated by food rewards and to have no overt aggression towards unfamiliar people.Most dogs were pet dogs living inside the house with their owners.Out of the total number of tested dogs, we received valid questionnaire replies for 987 adult dogs.We also investigated possible associations between different tests within the cognitive test battery from a larger data set of dogs, which included all tested dogs whether they had questionnaire answers available or not (N ¼ 2874 adults).We limited our analysis to dogs over the age of 7 months, as cognitive traits take time to develop in puppies (Lazarowski et al., 2020;Passalacqua et al., 2011;Bray et al., 2021).In addition, we analysed interrater reliability and testeretest reliability of the cognitive tests using separate data sets.
Our sample for the associations between cognition and questionnaire answers included a total of 157 breeds, and the analysis for associations between test battery results included 223 breeds.The most common breeds were the border collie, Labrador retriever and mixed breed.The total number of each breed is outlined in Table A1.See Table 1 for ages, sexes and other demographic details of the dogs for each analysis.

Cognitive Test Battery
Cognitive tests were performed by nine trained female testers at different locations across Finland, in a testing room of minimum size 30 m 2 .The test battery lasted approximately 1.5 h and included eight cognitive tests and three behavioural/personality measures.To ensure consistency across subjects, the test order was the same for all dogs.In the current study, we included five cognitive tests and three behavioural measures.These are described in Table 2.More detailed descriptions about each test can be found in our previous publication (Junttila et al., 2022).
All cognitive tests involved solving problems with either food or toy rewards, depending on which the dog valued more.The rewards were chosen by the owners, who were asked to bring in their dogs' favourite treats (and toys, if necessary).Most often the reward was food, and food rewards were always used in the Gesture Test.The owners were asked not to feed their dogs for a few hours before the beginning of the test battery to ensure food motivation.Fresh drinking water was available and the dog was off lead throughout the test battery.The owner was always present during testing.The test battery included a short break during which the owner could take the dog outside for about 2 min.Some participant dogs had missing values for cognitive tests included in the test battery, because they either did not pass the criteria for the training phases, were too afraid of the test situation (usually the apparatus) or managed to break the apparatus.In addition, 12.7% of dogs took part in a shortened version of the test battery.The number of dogs that took part in each test are outlined in Table A2.
A small number of dogs with questionnaire answers available (24) had previously taken part in a test battery as a puppy, and these dogs were removed from the analysis.Within the larger data set of dogs, 61 had previously taken part in a puppy test and were therefore excluded from the analysis.In both data sets, if dogs had been tested two times as an adult, the results from the first test occasion were included in the analysis.

Questionnaire
A questionnaire was sent to all owners who had given written permission to participate in research.The questionnaire included two validated questionnaires: the Canine Behavioural Assessment and Research Questionnaire, C-BARQ (Hsu & Serpell, 2003) and the Dog Impulsivity Assessment Scale, DIAS (Wright et al., 2011).In addition, we included other relevant questions relating to the dogs' lifestyle, background, training history and so on.The additional questions included in our analysis are described below, and demographic details can be found in Table 1.

Illness
We included a question on whether the dog had been diagnosed (either by a vet or by the owner) with an ongoing or recurrent illness or health problem, and this was analysed as a binary variable (yes/no).

Food motivation
The owners were asked whether they agree with the statement: 'My dog is very motivated to work for food/treats'.This was answered on a scale of 0e4 (0 ¼ Disagree completely, 1 ¼ Partly disagree, 2 ¼ Partly agree and partly disagree, 3 ¼ Partly agree, 4 ¼ Agree completely).Since very few owners gave a response of 0 or 1, these were combined with score 2, so that the scale was 1e3.

Previously owned dogs
To determine the experience of participating owners, the questionnaire included the question 'How many dogs have you owned altogether (including this dog)?'.The options were 1, 2, 3e5 and 6 or more.

Background of dog
We asked owners where they had acquired their dog from.Since a large majority of the dogs were acquired from a breeder, we used a binary variable to distinguish between rescues/rehomes and other sources (breeder, friend or similar).

Training level
Owners were asked whether their dog takes part in dog sports/ hobbies (and what kind), how frequently they participate, whether they have entered competitions and whether their dog is in a working role (and what type of role).Dogs received a score of 0e4, mostly based on their answers to their frequency of participating in sports/hobbies (0 ¼ never, 1 ¼ less than once a month, 2 ¼ 1e3 times a month, 3 ¼ 1e3 times a week, 4 ¼ daily or almost daily).If owners had answered 0, but they had included training or a hobby in their other answers, their score was changed to 1.If owners had answered that their dog is in a working role, training for a working role, or that their dog had participated in competitions, their score was changed to 4. Dog sports/working roles that were not categorized as training included participating in dog shows, racing, skijouring, hiking and swimming, as these were considered to have less of a training focus.

Questionnaire development
For both the C-BARQ and DIAS questionnaires, the same process of translation was followed.We translated the questionnaire into Finnish, after which it was back translated by a person fluent in Finnish and English.The questionnaire was changed accordingly if any parts did not retain the same meaning.A pilot version of the questionnaire was then answered by a small number of dog owners, and the wording of the questions was changed if necessary.
The final version of the questionnaire included 100 questions for C-BARQ, 18 questions for DIAS and 130 additional questions related to the dog's background, training history and so on.Study data were collected and managed using REDCap (Harris et al., 2009(Harris et al., , 2019) ) electronic data capture tools hosted at the University of Helsinki.The link to the online questionnaire was sent via email, starting at the end of January 2022, and the questionnaire was open until November 2022.In total we received responses for 1420 dogs, out of which 1407 had taken part in the cognitive test battery, resulting in a response rate of 32.1%.
Out of all dogs, 43 (4.3%) were not alive at the time of the questionnaire.In these cases, the owners were instructed to answer the questionnaire based on their dogs' behaviour prior to the dog becoming ill (if this was the reason for the dog passing away), and the dogs' age at the time of their death was used in the analysis.

C-BARQ
A Finnish version of C-BARQ was developed based on the 100item English-language version (Duffy & Serpell, 2012).The questionnaire consists of items that ask respondents to use a series of 5point ordinal rating scales (0e4).After removing dogs 7 months old or under, the sample consisted of 1094 dogs.We only included answers with a minimum of 80% of the C-BARQ questions answered, resulting in a sample size of 1011 dogs, on which a factor analysis was performed.A more detailed description of the factor analysis can be found in the Supplementary Material.
For the current study we decided to include the factors 'energy level', 'stranger-directed fear', 'excitability' and 'trainability', as these were predicted, based on our best knowledge, to be most likely to show associations with cognitive measures.We used the extracted factor scores from the factor analysis for analysing associations between cognition and questionnaire answers.
In addition, we included in our analysis the individual question 'Dog is slow to learn new tricks or tasks' and two additional variables that did not form a factor: 'stereotypies' and 'management problems'.Since dogs often display only one type of stereotypic behaviour, this variable was given a score based on the highest value from all stereotypy-related questions.For management problems, the sum of scores from the included questions were calculated for this variable (only for dogs that had a response rate of at least 80% for these four questions).The questions included in each questionnaire variable are outlined in Table 3.

DIAS
Following the original DIAS study (Wright et al., 2011), we included dogs of all ages that had a minimum response rate of 80% for the DIAS questionnaire.This resulted in a sample size of 1279 dogs, on which a principal component analysis was performed.A detailed description can be found in the Supplementary Material.Our analysis resulted in two factors that accounted for 40.2% of the variation.The components are described in Table 4.
We calculated scores for these two factors and the overall questionnaire score (with all 18 questions included) as per the original DIAS study (Wright et al., 2011).Only dogs with a minimum response rate of 80 % for each factor were included in the calculations.For 'impulsive aggression', very few dogs scored values higher than 0.7 and therefore the scores were divided into five groups: 0 ¼ 0.2; 1 ¼ 0.3; 2 ¼ 0.4; 3 ¼ 0.5; 4 ¼ 0.6; 5 ¼ 0.7e1.

Data Handling and Statistical Analyses
All analyses were performed using IBM SPSS version 29 (IBM, Armonk, NY, U.S.A.).The dog was first taught that one of two bowls always contained a treat.In the test trials, the dog was shown that one of the bowls was empty.The dog was then allowed to make a choice, being required to infer that the treat was hidden under the bowl that had not been lifted Binary variable: understanding of the task rated as unsuccessful or successful (first 4 trials correct or a minimum of 5 of 6 trials correct)

Testeretest repeatability
We measured testeretest reliability from the total data set of dogs that had taken part in the cognitive test battery (whether they had questionnaire data available or not).Of these, a total of 56 dogs had been tested two times as adults.Dogs that were over 9 years old at the second testing occasion were removed, which resulted in a remaining 49 dogs.The sample included 23 breeds (see Table A1).For 32 dogs (65.3%) the tester was a different person on the two test occasions, and for 26 dogs (53.1%) the testing location was different.
We used intraclass correlation coefficients using a two-way mixed-effects model for continuous, normally distributed variables, Cohen's weighted kappa for non-normally distributed variables and Cohen's kappa for binary variables.

Interrater reliability
For each cognitive test, eight smartDOG licence testers each watched 15 videos of the test situation, with author Katriina Tiira as the tester.The videos included 49 dogs from 25 breeds, ranging in age from 5 months to 8 years.We used intraclass correlation coefficients with a two-way mixed-effects model for continuous, normally distributed variables and Fleiss' kappa for ordinal and binary variables.

Associations between cognition and questionnaire answers
To analyse possible associations between cognitive test results and questionnaire variables, mixed model analyses were used.After removing dogs aged 7 months old or younger, dogs tested previously as puppies and dogs without the required 80% response rate for the questionnaire, the sample consisted of 987 dogs.
Questionnaire variables were used as the dependent variables.For continuous variables, linear mixed models were used with maximum likelihood estimation (overall DIAS, DIAS behavioural regulation, energy level, management problems, stranger-directed fear, excitability, trainability).Standardized residuals were assessed visually via histograms and PeP plots to check for normality, and homoscedasticity was assessed using a plot of studentized residuals versus unstandardized predicted values.Linearity was assessed by partial regression plots and a plot of studentized residuals against the predicted values.Outliers were checked using Cook's distance and leverage values.All assumptions were met apart from stranger-directed fear, which was log transformed due to non-normality.For ordinal variables (DIAS impulsive aggression, slowness to learn, stereotypies) we used generalized linear mixed models with multinomial logistic regression.For slowness to learn and stereotypies, scores of 3 and 4 were combined into one group, due to very few scores of 4.
Breed and tester identity were included as random variables (random intercept) in the models, but this was not possible in all models because one or both variables were redundant and resulted in uncertain model fit.Breed was therefore excluded from models with the dependent variables stereotypies and slowness to learn.Tester identity had to be excluded from all models except those with the dependent variables slowness to learn and energy level, in addition to dependent variables stranger-directed fear (only with independent variables Cylinder Test and V-detour), overall DIAS (only with Cylinder Test) and trainability (only with Gesture Test and Logical Reasoning).Attempts were also made to include test location as a random variable, but it was redundant in all models and therefore excluded.
Independent variables included the cognitive test variables and control variables which, according to our best knowledge, could potentially influence the variables of interest in the study.These were age (at time of filling in questionnaire), sex, illness, food motivation, previously owned dogs, background and training level.Correlations were assessed between all independent variables using correlation coefficients and VIF values, and there was no multicollinearity between variables.Continuous independent variables were transformed into z scores to aid with interpretation.For each cognitive test variable, a separate analysis was performed, since we In sum, each model included one of the questionnaire variables as the dependent variable, and independent variables included age, sex, training level, food motivation, background of dog, experience of owner, illness and one cognitive test variable.Breed and tester identity were included as random variables whenever possible.

Associations between test measures
We analysed associations between test battery measures using Spearman rank correlations in the case of ordinal and continuous variables, and ManneWhitney U tests in the case of Logical Reasoning.We used a sample of 2874 adult dogs (aged 8 months to 8 years) which had taken part in the test battery between 2016 and 2022.

Ethical Note
We confirm that the procedures complied with national and EU legislation.Research was performed in accordance with the Declaration of Helsinki.The study was approved by the University of Helsinki Viikki Campus Research Ethics Committee (Statement 12/2021, accepted on 18/05/2021).Before participating in the cognitive test battery and answering the online questionnaire, each dog owner gave informed written consent for using their dogs' test results in research.Reporting of results follows the recommendations of the ARRIVE guidelines.Dogs took part in cognitive tests voluntarily and all tests involved working for food or toy rewards.The welfare of dogs was a priority during testing, and testing was performed only if the dog voluntarily took part.If any dog showed signs of fear or distress towards test equipment, this particular test section was not performed.If the dog was fearful of the tester and did not get used to her, the test battery was not performed, although this was extremely rare, as most dogs found the test battery a positive experience.

Descriptives of Cognitive Test Scores
For Logical Reasoning, 164 dogs (19%) were rated as having successfully understood the task, whereas 699 dogs (81%) did not understand the task.Descriptive statistics for the other cognitive test scores can be found in Table 5.

Testeretest Reliability
Testeretest reliability ranged from 0.14 to 0.53.All cognitive tests therefore showed fair to moderate agreement, apart from Logical Reasoning, which showed only slight agreement (Koo & Li, 2016;Landis & Koch, 1977).See Table 6 for detailed results.

Associations Between Questionnaire Answers and Cognitive Tests
See Table 8 for detailed results of all significant and marginally nonsignificant associations between questionnaire and test variables.

Trainability (C-BARQ factor)
Dogs were evaluated by their owners as more trainable if they had higher inhibitory control in the Cylinder Test or spent more time on human-directed behaviour during the Unsolvable Task.

Slowness to learn (C-BARQ question)
Dogs that did not understand the Logical Reasoning task were rated as slower at learning by their owners.There was a marginally nonsignificant tendency for dogs to be rated as slower at learning if they had a longer latency to solve the V-detour or if they had lower inhibitory control in the Cylinder Test.

Stereotypies (C-BARQ variable)
Stereotypies were not significantly predicted by any test battery variable.

Stranger-directed fear (C-BARQ factor)
The more time dogs spent on human-directed behaviour during the Unsolvable Task the higher their fear of strangers was rated as.Management problems (C-BARQ variable) Dogs that were faster at solving the V-detour or spent less time on human-directed behaviour during the Unsolvable Task had more severe management problems (pulling on the lead, escaping, stealing food and chewing forbidden objects) according to owners.

Excitability (C-BARQ factor)
There was a marginally nonsignificant tendency for dogs with lower inhibitory control in the Cylinder Test to be rated as more excitable by their owners.

Energy level (C-BARQ factor)
There was a marginally nonsignificant tendency for dogs with lower inhibitory control in the Cylinder Test to be rated as having higher energy levels.

Overall DIAS score
Dogs with higher overall DIAS scores (higher impulsivity) had lower inhibitory control in the Cylinder Test and spent less time on human-directed behaviour during the Unsolvable Task.There was also a marginally nonsignificant tendency for more impulsive dogs to solve the V-detour faster.

Behavioural regulation (DIAS factor)
Similarly to the overall DIAS scores, dogs with higher scores for Behavioural Regulation (higher impulsivity) also had lower inhibitory control in the Cylinder Test and spent less time on humandirected behaviour in the Unsolvable Task.

Impulsive aggression (DIAS factor)
Impulsive Aggression (DIAS) was not significantly predicted by any test battery variable.

Associations Between Test Battery Variables
We found significant correlations between several test battery variables.The strongest correlations (with correlation coefficients between 0.24 and 0.42) were found between the three noncognitive measures 'greeting', 'exploration' and 'activity level', which were all positively correlated with each other (which would suggest that especially greeting and exploration may be good representations of a shynesseboldness personality dimension).See Table 9 for details of all correlations.

DISCUSSION
We found significant associations between cognitive test results and owner-derived questionnaire answers describing dogs' everyday life and behaviour problems.Almost all cognitive tests included in our study were associated with at least one behavioural variable from the questionnaire.This is the only study so far that has investigated associations between cognition and dogs' everyday behaviour on such a large scale.We can conclude that the Unsolvable Task, Cylinder Test, V-detour and Logical Reasoning task can tell us something about dogs' behaviour in their daily lives with their owners.Therefore, the tests seem to measure traits that have a wider impact on dogs' behaviour than just measuring their responses in the specific test situation.Similarly to humans, dogs' cognition seems to be linked to their everyday life and behaviour (See Fig. 1 for an overview of the associations we found for each cognitive test).
In addition, we found fair to moderate testeretest reliability for most cognitive measures included in our test battery, which seems to be consistent with repeatability of cognitive traits in other animals (Cauchoix et al., 2018).This is one of the very few studies that have investigated testeretest reliability of cognitive traits in adult animals (Griffin et al., 2015).
The Cylinder Test has been widely used in animals as a measure of inhibitory control, although some studies have failed to find evidence for associations with other inhibitory control measures (Barela et al., 2023;Olsen, 2022;van Horik et al., 2018; but see Krichbaum & Lazarowski, 2022).Using the largest data set (from dogs) so far, we found that dogs with low inhibitory control in the Cylinder Test had higher impulsivity scores in the DIAS questionnaire, which strongly suggests that the Cylinder Test can tell us something about dogs' inhibitory control and impulsivity.Moreover, dogs that had lower inhibitory control in the Cylinder Test were rated by their owners as less trainable, and they had higher activity and exploration levels during the test battery.There was also a marginally nonsignificant association between low inhibitory control in the Cylinder Test and owner-rated slowness to learn, excitability and energy level.We also found in an earlier study that Cylinder Test results differ between breeds (Junttila et al., 2022), suggesting a heritable component.
The associations we found seem to resemble associations observed between human ADHD traits.ADHD is a neurodevelopmental disorder in humans, which is characterized by hyperactivity, inattention and impulsivity, and it is often associated with impaired academic success, learning ability and cognitive functioning (e.g.Wender et al., 2006).People with ADHD have been shown to perform more poorly in tests of inhibitory control, especially motor inhibitory control (e.g.Castellanos et al., 2006;Schachar et al., 1995).ADHD-like traits have also been found in dogs (Lit et al., 2010;Sulkama et al., 2021;Vas et al., 2007), and these seem to correlate negatively with training focus (Salonen et al., 2022) and behavioural inhibition in a Go/No-Go paradigm (Bunford et al., 2019).
Previous canine studies have found that high impulsivity is positively associated with behaviour problems and negatively associated with trainability (Piotti et al., 2018;Wright et al., 2012) and that ADHD-like traits (such as inattention and impulsivity) correlate negatively with training focus (Salonen et al., 2022).It was also found that dogs with higher inhibitory control in the Cylinder Test were seen as less of a burden by their owners (Imponen & Pelliccioni, 2021).In addition, an earlier study found that police dogs that had lower inhibitory control in the Cylinder Test performed worse in an explosive detection task (Tiira et al., 2020).Our study seems to be in line with these results, suggesting that impulsive dogs may be more difficult to train than dogs with high inhibitory control.
Dogs with lower inhibitory control in the Cylinder Test also spent less time directing their behaviour towards humans during the Unsolvable Task, which was also associated with higher impulsivity (DIAS) scores as rated by owners.One reason for our findings could be that impulsive dogs may have difficulties switching from an unsuccessful strategy (manipulating the apparatus) to an alternative one (turning their attention towards a Associations between cognitive test variables, questionnaire variables and behavioural measures from the test battery.Direction of association: þ positive association, À negative association.Questionnaire variables refer to increasing scores for each variable, with DIAS variables referring to higher impulsivity.A higher score for 'slowness to learn' means slower learning, whereas a lower score means faster learning.Test battery variables refer to success, speed or higher score for each test.Cylinder Test: increasing inhibitory control.Gesture Test: increasing success at reading human gestures.V-detour: faster solving of the task (Note that this has been reversed compared to the Results section for easier interpretation).Human-directed behaviour: more time spent on human-directed behaviour during the Unsolvable Task.Logical Reasoning: dog successfully understood the task.
human), especially when in a high state of arousal.Humans with ADHD are often reported to have tendencies for hyperfocus (Ashinoff & Abu-Akel, 2021), difficulties switching between tasks (e.g.Halleland et al., 2012) and lower frustration tolerance (Horesh et al., 1997;Scime & Norvilitis, 2006;Seymour et al., 2019).Dogs that spent less time directing their behaviour towards humans in our study were also reported to be less trainable and to have more management problems, and they had a higher exploration score and mean activity level than more human-dependent dogs, all of which seem to be consistent with ADHD-like traits in humans.
Several studies suggest that dogs' human-directed behaviour in the Unsolvable Task is a form of communicative behaviour and may also indicate attraction to the human face (e.g.Hirschi et al., 2022;Mikl osi et al., 2003;Passalacqua et al., 2013).One study found that dogs that were successful in a social learning task were also rated as more trainable by their owners (Marshall-Pescini et al., 2008).Therefore, it is possible that dogs with higher sociocognitive abilities may be more attentive to humans and more willing to cooperate and communicate with them, which may also improve their trainability and reduce management problems.
Interestingly, human-directed behaviour (towards either the owner or the experimenter) during the Unsolvable Task was also associated with fear of unfamiliar people as reported by owners.Different forms of fear and anxiety seem to correlate with each other (Svartberg, 2005;Temesi et al., 2014;Tiira et al., 2016).Therefore, many stranger-fearful dogs might be generally fearful and lacking in confidence (Gazit et al., 2022), which may result in dogs giving up earlier on a difficult task and instead relying more on humans.In line with this, we found an association between humandirected behaviour and lower exploration (staying close to the owner when allowed to explore freely) in a novel environment, which often seems to be indicative of fearfulness (K.Tiira, personal observation).The fear/anxiety and resulting stress experienced in daily life could also be impairing dogs' cognitive and problemsolving ability, as has been found in humans and other animals (e.g.Calvo & Gutierrez-Garcia, 2016;Eysenck et al., 2007;Shackman et al., 2006).However, this seems unlikely since stranger-directed fear was not associated with any other cognitive measures.
An earlier study (Passalacqua et al., 2013) found that dogs diagnosed with anxiety-related disorders exhibited less gazing towards the experimenter than control dogs and instead spent more time away from the apparatus.A similar result was found in horses (Henriksson et al., 2019) and goats (Yoshida & Koda, 2020).Similarly to Passalacqua et al. (2013), we found that dogs that were more fearful of a stranger while greeting an unknown person and that stayed closer to the owner instead of exploring (both of which can be signs of fearfulness) spent slightly more time away from the apparatus (abandoning the task).However, our findings for humandirected behaviour were not consistent with Passalacqua et al. (2013), which could be a result of the type and severity of the fear/anxiety of participating dogs.A large majority of dogs in Passalacqua and colleagues' study exhibited generalized anxiety, whereas we included only one type of fear/anxiety.The dogs in our study were also reported to exhibit only mild forms of human fear, in contrast to the more severe cases in Passalacqua and colleagues' study, which could result in different expressions of behaviour between these dogs when faced with an Unsolvable Task.
Similarly to the Cylinder Test, the V-detour has previously been assumed to be a measure of inhibitory control (Kabadayi et al., 2018;van Horik et al., 2018).Our results, however, seem to suggest the opposite.We found no significant associations between impulsivity (DIAS) and V-detour results, but there was a tendency for dogs that were faster at solving the V-detour to be rated as more impulsive, and they were reported to have significantly more management problems than dogs that were slower at solving the task.Fast solvers were also significantly more active and explorative during the test battery, and it is possible they could have accidentally figured out the solution to the task since they were moving around more.Similarly to previous studies (Marshall-Pescini et al., 2015;van Horik et al., 2018), we also found no association between the V-detour and the Cylinder Test.These results would suggest that success in the V-detour task may be associated with low inhibitory control, rather than the other way round.However, our use of latency as the test measure may have provided different results than would have been found if using an alternative measure, and therefore more research on this topic may be warranted.
One explanation for our results could also be due to the strategy dogs were using during the V-detour task; many dogs which were slow to solve the task behaved similarly to dogs that turned towards the human during the Unsolvable Task.In an additional ManneWhitney U test, we found that turning towards the human (binary yes/no variable) during the V-detour task was positively associated with human-directed behaviour during the Unsolvable Task (U ¼ 439 285, z ¼ e3.82; P < 0.001), confirming that there are similarities between human-directed behaviour during the Unsolvable Task and the V-detour.This suggests that the reason some dogs are less successful in the V-detour task might not necessarily be related to their problem-solving ability or inhibitory control, but instead to their main problem-solving strategy and their preference for cooperating with humans.This may, in turn, be associated with impulsivity.
Note that there may be various reasons behind dogs' behaviour during these cognitive tests, and dogs with very different personalities may still behave similarly in a specific test.Thus, different motivators may produce similar behaviour/results in a cognitive test.As an example, stranger-directed fear did not correlate with impulsivity (DIAS) or trainability, even though they were all associated with human-directed behaviour during the Unsolvable Task.Therefore, it seems that some dogs might spend a large amount of time directing their behaviour towards a human because of underlying fearfulness, whereas other dogs may instead be motivated to solve the task by cooperating with humans; in both cases, the dog is very human directed during the Unsolvable Task.This phenomenon is similar to aggressive behaviour in dogs, where the behaviour itself looks very similar, but may be motivated by various reasons.The possibility of varying reasons behind dogs' performance in cognitive tests is something that has rarely been considered in previous research.
Note that our sample was not fully representative of the general population of pet dogs.Most owners who participated in the test battery were very active in dog sports or training, and some dogs were in working roles.Only a very small percentage of dogs had received no specific training.Inclusion of dogs may also have been biased due to requiring participant dogs to be motivated by food rewards and to not be aggressive to strangers (although there were several participants that took part despite not fulfilling these criteria).When analysing the questionnaire data, it was clear that behaviour problems (human-directed fear, stereotypies, impulsive aggression and management problems) were rare, and when they occurred, they were mostly very mild.Severe behaviour problems were almost nonexistent.Therefore, it is important to keep in mind that our results apply to mild forms of problem behaviour, and further research may be warranted to determine if similar associations exist for more extreme forms.However, it is noteworthy that significant associations were still found despite the mildness of the reported behaviour problems.
Another limitation of our study was that due to the nature of the data, almost all dog owners filled in the questionnaires after their dogs had taken part in the cognitive tests (median 1.83 years later).Based on their dogs' behaviour during the test, dog owners were provided with a summary of their dogs' personality traits and cognitive skills both verbally during the test and in written form after the test.Feedback given to owners was based on what was currently known about these cognitive tests, as well as on how the dog responded to the test situation, its body language and other behavioural and personality traits which appeared during the test situation.In order to confirm our findings, they may need to be replicated in the future with tests and questionnaires provided in a different order.
In conclusion, we found that dogs' daily behaviour and personality traits can indeed be predicted using experimental, dogfriendly cognitive tests; they tell us something about dogs' behaviour outside of the test situation.We found significant associations between cognitive test results and dogs' everyday behaviour as reported by their owners, despite the more extreme forms of behaviour problems being missing from the population.Our results mean that these cognitive tests have great potential for explaining, predicting and understanding the wider behavioural and personality traits of dogs.The associations we found were similar to those found in humans.In conclusion, our results provide important information which helps us to not only understand the possible underlying causes behind dogs' behaviour in cognitive tests, but also to understand the traits which might be in the background of pet dogs' behaviour in their everyday lives.We also provide additional findings for the growing research area of associations between cognition and personality in animals.
Figure1.Associations between cognitive test variables, questionnaire variables and behavioural measures from the test battery.Direction of association: þ positive association, À negative association.Questionnaire variables refer to increasing scores for each variable, with DIAS variables referring to higher impulsivity.A higher score for 'slowness to learn' means slower learning, whereas a lower score means faster learning.Test battery variables refer to success, speed or higher score for each test.Cylinder Test: increasing inhibitory control.Gesture Test: increasing success at reading human gestures.V-detour: faster solving of the task (Note that this has been reversed compared to the Results section for easier interpretation).Human-directed behaviour: more time spent on human-directed behaviour during the Unsolvable Task.Logical Reasoning: dog successfully understood the task.

Table 1
Demographic details of participant dogs for each analysis yThe most commonly reported illnesses and health problems are listed.

Table 2
Cognitive and behavioural tests included in the test battery in the order in which they were presented to the dog

Table 3
Factors and additional variables derived from the C-BARQ questionnaire, including factor loadings

Table 4
Extracted factors and factor loadings of each question from the principal component analysis on the DIAS questionnaire not wish to control for the possible influence of other test variables on each other.All P values were corrected for false discovery rate (FDR) using the Benjamini e Hochberg procedure.The FDR correction was done based on the number of independent variables of interest (cognitive test results), resulting in 50 P values included in the calculations (five independent variables and 10 dependent variables). did

Table 5
Descriptive statistics for the cognitive tests in the test battery (excluding Logical Reasoning, which is described separately)

Table 8
Results from all significant and marginally nonsignificant associations between dependent and independent variables FDR-corrected P values that were significant at the P < 0.05 level are in bold.

Table A1
Number of dogs tested per each breed for each analysis used in the study