Article Text

Download PDFPDF
Increasing physical activity by four legs rather than two: systematic review of dog-facilitated physical activity interventions
  1. Ryan E Rhodes1,
  2. Maria Baranova1,
  3. Hayley Christian2,
  4. Carri Westgarth3
  1. 1 School of Exercise Science, Physical and Health Education, University of Victoria, Victoria, British Columbia, Canada
  2. 2 Faculty of Health and Medical Sciences, Centre for Child Health Research, University of Western Australia, Crawley, Western Australia, Australia
  3. 3 University of Liverpool, Liverpool, Merseyside, UK
  1. Correspondence to Dr Ryan E Rhodes, University of Victoria, Victoria, BC V8W 3N4, Canada; rhodes{at}


Objectives Regular walking is a critical target of physical activity (PA) promotion, and dog walking is a feasible PA intervention for a large segment of the population. The purpose of this paper was to review PA interventions that have involved canine interactions and to evaluate their effectiveness. A secondary aim of this review was to highlight the populations, settings, designs and intervention components that have been applied so as to inform future research.

Design Systematic review.

Data sources We carried out literature searches to August 2019 using six common databases.

Eligibility criteria Studies included published papers in peer-reviewed journals and grey literature (theses and dissertations) in the English language that included any PA behaviour change design (ie, randomised controlled trial, quasi-experimental) that focused on canine-related intervention. We grouped findings by population, setting, medium, research design and quality, theory and behaviour change techniques applied.

Results The initial search yielded 25 010 publications which were reduced to 13 independent studies of medium and high risks of bias after screening for eligibility criteria. The approaches to intervene on PA were varied and included loaner dogs, new dog owners and the promotion of walking among established dog owners. Findings were consistent in showing that canine-assisted interventions do increase PA (82% of the studies had changes favouring the canine-facilitated intervention). Exploratory subanalyses showed that specific study characteristics and methods may have moderated the effects. Compared with studies with longer follow-up periods, studies with shorter follow-up favoured behaviour changes of the canine intervention over the control condition.

Conclusion Canine-based PA interventions appear effective, but future research should move beyond feasibility and proof of concept studies to increase rigour, quality and generalisability of findings.

  • aging/ageing
  • exercise
  • health promotion
  • intervention
  • walking

Statistics from

Request Permissions

If you wish to reuse any or all of this article please use the link below which will take you to the Copyright Clearance Center’s RightsLink service. You will be able to get a quick price and instant permission to reuse the content in many different ways.


Regular physical activity (PA) among adults has a significant number of health benefits, evidenced by considerably reduced risks of key chronic diseases, such as heart disease, type 2 diabetes, several cancers and even all-cause mortality.1 2 PA is also a critical health behaviour among children and adolescents as it protects against high blood pressure, high blood cholesterol, metabolic syndrome, low bone density, depression and obesity,3–5 and forms the behavioural patterns that track into adulthood.6 7 Despite the irrefutable benefits of regular PA, participation rates are low, particularly among industrialised and wealthy nations.8 Consequently, sustainable PA promotion initiatives are of high importance to public health.

Regular walking is one of the most common PAs and thus a key target for intervention.9 Walking is one of the safest and most affordable forms of PA.10 It is easily achievable by a large majority of people with little skill or equipment involved, and is an activity that can be performed individually or with others in groups. People walking their dogs is a common sight in most urban communities in many countries. Thus, perhaps unsurprisingly, dog ownership is an established correlate of regular walking.11–13 It is now well established that dog owners report more recreational walking than non-owners.14–17

While there is ample cross-sectional evidence of higher walking levels associated with dog ownership, such evidence has limited application to PA promotion. Dog ownership is a considerable responsibility with cost implications, and so recommending that people adopt dogs merely for the health benefits of walking is not practical nor ethical.18 Approximately 30% of households in high-income countries own dogs,12 yet it is estimated that up to half of these dog owners do not walk with their dogs regularly.14 Promoting more walking as a low-cost sustainable behaviour among existing dog owners seems like a logical way to engender both human and canine health benefits simultaneously in a large target population.12 19 20 Dogs are also often used in various rehabilitation contexts and social programmes for marginalised or clinical populations.11 13 Leveraging these programmes to increase PA may be an additional benefit.

Research focused on the correlates of dog walking has been building in order to inform interventions. A review of 31 studies on the correlates of walking among dog owners found that the relationship with a dog, often in the form of feelings of responsibility/obligation/support, is the foremost factor associated with dog walking that may distinguish it uniquely from other forms of PA.21 The review also highlighted environmental access to suitable walking areas with dog-supportive features (eg, off-leash exercise) as consistently associated with increased dog walking. Individual-level theories of behaviour change have also been applied to understand dog-walking behaviour, such as social–cognitive theory,22 self-determination theory,23 theory of planned behaviour24–26 and multi-process action control.27 These approaches have highlighted the importance of canine-related outcome expectations, social support, intention to walk, autonomous motivation, habit, and identity.

Taken together, there are a number of correlates of dog-walking behaviour that could be targeted as mediators to change behaviour across multiple levels of agency, from policy and environment to social support groups and individual motivation. At present, however, there is no systematic review of interventions where dogs are used as the means to increase human PA. Thus, the purpose of this paper was to review PA interventions that have involved canine interactions and to evaluate their effectiveness. As a secondary objective, we sought to explore the reasons behind successful interventions and the breadth of the intervention approaches, and highlighted the populations, settings, designs and intervention components that have been explored in order to inform future intervention research.


This systematic review was conducted and reported in accordance with the Preferred Reporting Items for Systematic Reviews and Meta-Analyses guidelines.28

Eligibility criteria

Four eligibility criteria were implemented for papers to be included: (1) published in an English, peer-reviewed journal or the grey literature; (2) used a design that can infer changes in PA (randomised controlled trial (RCT), quasi-experimental, predesign–postdesign) and not cross-sectional or qualitative designs; (3) measured dog-facilitated PA; and (4) included an intervention component to impart PA change (ie, not passive observational).

Information sources, search strategy and study selection

From August to September 2019, the second author searched six electronic databases (PsycINFO, Web of Science, ERIC, PsycARTICLES, HealthSource: Nursing/Academic Edition and PubMed) to identify eligible articles. Search terms were the following: “dog” OR “pet” OR “canine” AND “physical activity” OR “walking” AND “intervention” OR “trial” OR “experiment”. Boolean searching was used to ensure all articles included at least one term from each of the three groups as follows: the phrase ‘OR’ was used within groups and ‘AND’ was used between groups. To streamline the search, the author applied a filter to the search strategy: English language (see online supplementary appendix 1). Theses and dissertations were also included in this search strategy to assist in retrieving the grey literature. Finally, a manual search of all reference lists from the retrieved eligible papers was performed to cross-reference eligible studies that may have been missed in the formal search. Articles were screened for eligibility by title and abstract. At this point, exclusion was mainly due to study design or absence of human PA. See figure 1 for a detailed depiction of the literature search screening process.

Supplemental material

Figure 1

PRISMA flow diagram. Caption: source inclusion process. Adapted from PRISMA statement (Moher et al 28). 1In the 25 010 publications, 382 were identified from PsycINFO; 16 578 were from Web of Science; 5677 were from ERIC; 252 were from PsycARTICLES; and 2121 articles were from PubMed. Search fields: title and abstract. Search time: 3 September 3 2019. PRISMA, Preferred Reporting Items for Systematic Reviews and Meta-Analyses.

Data collection process and data items

Data were extracted independently among two authors (first and second authors) for intervention, measurement and participant characteristics, as well as statistical significance of the findings. Intervention characteristics included use of theory, intervention duration and quality, setting, medium of dissemination and behaviour change techniques (BCTs). Measurement characteristics included PA measures (objective, self-reported or both). In all cases where possible, we sought the most robust measure of PA in our analyses (eg, total walking, compared with dog walking specifically; objective assessments, compared with self-report). Participant characteristics included sample size, mean age, sex, baseline activity status, population type (general community or clinical) and geographical location.

These two coders also coded BCTs according to Michie et al’s29 93-item taxonomy. Independently, the two coders completed a binary checklist (0=no and 1=yes) for each included study. Any discrepancies were resolved by discussion until 100% consensus was reached. The total number of BCTs per study was also coded.

Risk of bias assessment

We conducted a risk of bias assessment using the ROB-2 for RCTs30 and the ROBINS-I for nonrandomised interventions.31 Three independent coders (first, third, and fourth authors) assessed study quality, and any disagreements were resolved via discussion to reach a final decision. The studies were then classified into overall low risk, some concerns (medium quality) and high-risk categories based on the scoring protocol of the instruments. A detailed quality assessment for each item by study is included in online supplementary appendices 2 and 3.

Supplemental material

Supplemental material


Meta-analysis was precluded for two main reasons. First, there was extensive heterogeneity in the study designs, statistical tests employed and populations, all of which impact the ability to accurately pool the studies for quantitative synthesis.32 Second, a descriptive synthesis is most appropriate when there are caveats or other idiosyncrasies specific to some studies that could change the outcome in a meta-analysis. Thus, we employed two methods of synthesis suggested by McKenzie and Brennan.33 Following initial read-throughs of the studies, our analysis collated the median effect size and the quartile range, as well as a count of the direction of the findings in support of the dog-related intervention compared with its comparison condition, independent of whether the finding was reported as statistically significant. Subanalyses of these results were also conducted to explore group differences according to the design employed, the follow-up length of the studies, the population of the sample, the quality of the studies, the setting and medium of intervention, and the theoretical frame and BCTs employed using the same analysis approach.


Study selection

As shown in figure 1, the electronic database search yielded 25 010 articles, out of which 116 were determined to be potentially relevant through title screening. Of those, 42 records were screened out by titles and abstracts, including duplicates that were removed. In the remaining 74 records, 62 studies were excluded because they (1) covered topics other than PA in a form of dog walking (n=27), (2) did not use the appropriate design (n=25) or (3) were review articles (n=10). Three additional articles were identified by manual cross-referencing and searching theses and dissertations. If there was any doubt regarding the relevance of a paper, the full text was discussed among authors. A total of 13 independent studies passed the inclusion criteria and were included for analysis (see online supplementary table 1).

Supplemental material

Study characteristics and measures

Table 1 provides study characteristics for the final 13 studies and 844 participants in the analysis. Sample sizes across studies ranged from n=16 to n=236. Participant samples also varied in age from 10 to 95 years. Three studies included children; eight studies included middle-aged adults; one study included an older adult sample (primarily over 65+ years); and one study did not specify the age of the participants. In terms of gender distribution, all studies included both male and female participants. The PA levels of participants also varied across studies. Six studies included only participants below the recommended PA guidelines (as in the inclusion criterion), and seven studies had no criterion of baseline PA for inclusion in the study.

Table 1

Characteristics of the studies in the review

Our risk of bias assessment revealed that five studies were classified as high risk/low quality, and eight were classified as medium quality/risk. Most of the studies (n=10) were conducted in USA, while two studies were conducted in Canada and one in the UK. Seven studies did not explicitly mention a theoretical framework, and six studies employed either one or more theories, such as multi-process action control (n=1), social–cognitive theory (n=4), or the health promotion model (n=1). BCT application ranged from 0 to 15 across studies, with six studies employing more than four. Eight studies were randomised designs; three were quasi-experimental; and two studies had a single pregroup–postgroup design. A number of different tools were employed to measure PA: nine studies used objective measures such as accelerometers (n=5), pedometers (n=3) and counts of attendance checks, and four studies used self-report. Three studies conducted interventions in a group setting; four studies used weekly or biweekly emails; four studies used face-to-face intervention; and two studies used alternative means (mail and internet). In terms of the intervention setting, 3 studies took place in a clinical setting (hospital, vet clinics and assisted-living facilities) with observation, and the remaining 10 studies were conducted within a community or home setting.

Changes in PA behaviour

Of the 13 studies included in the review, we used 934–42 for effect size-based analyses and 1134–44 for vote counting of the direction of effects in the interventions (see table 2). Overall, there was positive evidence that canine-related interventions increased PA across time with a median effect size d of 0.28 (quartile range=0.02), and 82% of the studies showed a direction in favour of the dog-walking intervention.

Table 2

Group differences by study characteristics and methods

Results of the effectiveness of these interventions were mixed, however, when exploratory subanalyses were conducted by study methods and characteristics (see table 2). For example, studies with passive control groups that did not involve a PA promotion component34 36–40 resulted in a larger and more consistent effect (d=30, 86% of studies in the direction favouring the dog intervention) than the small number of studies that included active PA comparison interventions35 41 43 (d=0.06, 67% of studies in the direction favouring the dog intervention). Studies were not heterogeneous enough to evaluate risk of bias in subgroup analyses (all but one study with extracted effect sizes was medium risk). Studies34 36–38 42 43 with a shorter (<5 months) follow-up appeared to have larger (d=0.50) and more consistent effects favouring the dog-walking intervention (100% of studies) than studies35 39–41 44 with longer (>5 months) follow-up periods (d=0.02, 60% of studies in the direction favouring dog intervention). Similarly, studies with supervised settings34 42 and some components of face-to-face intervention34–36 38 42 43 appeared to have greater consistency in findings (100% of studies favoured supervised dog intervention; 86% of face-to-face studies favoured dog intervention). These face-to-face studies also had larger effects (supervised d=0.68, face-to-face d=0.42) than community35–41 43 44 PA with e-health or distance-based39–41 44 forms of intervention (community d=0.10, 77% of studies in the direction favouring dog intervention; e-health d=0.05, 75% of studies in the direction favouring dog intervention).

To explore the potential effect of sample size on the findings, we used an n=70 cut-off based on a rudimentary power analysis using G-Power for a small effect size (f=0.17) typically observed in PA interventions,2 with alpha=0.05, power=0.80 and a pre–post, two-group experimental design.45 Studies with larger samples35 41 44 had a lower median effect size (d=0.06) and less consistent results (67% of studies in the direction favouring the dog intervention) than samples within the smaller size34 36–40 42 43 grouping (d=0.34, 88% of studies in the direction favouring dog intervention). Finally, studies with an explicitly stated theoretical basis for the intervention34 37 39–43 had some indication of a larger effect (d=0.31) and more consistency (86% of studies in the direction favouring dog intervention) than interventions35 36 38 44 based on no stated theory (d=0.10, 75% of studies in the direction favouring dog intervention), but the volume of BCTs was not clearly linked to behaviour change outcomes.


Regular walking is a critical target of PA promotion, given its ease of performance, relative affordability and high preference among other PA alternatives.9 Promoting dog walking is a viable PA intervention for a large segment of the population.12 While a considerable amount of observational evidence on the correlates of dog walking has been published,21 this review provides the first systematic assessment of interventions that aimed to increase walking. Overall, the findings of this review show that interventions targeting dog walking do result in behaviour change, with a quartile range of the median effect size d from 0.26 to 0.30, commensurate with traditional forms of PA intervention.2 Regular dog walking is a type of PA that is sustainable because of its routine nature and its resilience to changes in season/weather.46 Additionally, dog walking serves multiple collateral goals, such as canine well-being and human health improvements, and can be a catalyst for more socialising in the community.12

Overall, 13 mostly randomised or non-randomised studies met the inclusion criteria and were included in our review of a total of 844 participants from three countries. There was a mix of medium and high risk of bias, and studies were heterogeneous in terms of age, ranging from children to older adults, with a balanced mix of male and female participants. The approaches to intervene on dog walking were also varied and included loaner dogs, new dog owners and promotion of walking among established dog owners. The theoretical approaches to the dog-walking intervention were also varied, with six studies stating a theoretical basis for its proposed mediators, employing a total of 26 different types of BCTs that ranged from 0 to 15 different techniques used within the studies. By contrast, the measurement of the criterion variable of PA was consistently sound, with 10 of the studies using objective assessments. Finally, implementation of these interventions included a wide variety of dissemination, such as face-to-face consultations, email, telephone, physical mail, group presentations, internet delivery in settings from home or group living facilities, and walking groups at parks, classrooms and medical facilities. Intervention effectiveness follow-up also varied considerably and ranged from 4 weeks to 1 year. Thus, the available sample of studies represents a rich data data set to appraise the state of current evidence in an attempt to synthesise findings and to identify areas for future research.

From this review, we recommend that dog-walking interventions with extended follow-ups be considered to understand the sustainability of this intervention approach and advance this literature beyond the feasibility and proof of concept stage. Specifically, there was evidence that shorter-duration intervention follow-ups (eg, 12 weeks or less) may have been more effective than longer durations (eg, 6 months+), so this needs further exploration. Long-term behaviour fits with the longer-term nature of dog ownership, yet future research may also need to consider the life stage of humans and dogs within different households as interventions begin to scale up.

The results of our review also showed differences in the effectiveness of these interventions, depending on the comparison groups included within the designs. Specifically, walking in the dog intervention condition increased over the comparison condition within the studies where these groups were given no/minimal PA promotion content, but dog-based interventions seemed less effective when comparison conditions also included PA promotion content. There are a couple of possible reasons for this finding. First, it may be that generic PA/walking promotion interventions are used by dog owners as a means to increase walking with their dog anyway. Thus, the participants adapt the information themselves and apply it to increase their own dog walking.43 Alternatively, the tailored approach to targeting dog walking specifically is no more or less sensitive than standard PA interventions. This was supported in one study we reviewed that compared walking interventions for dog owners and non-dog owners39 ; both intervention conditions increased walking over time, although the small sample size prevented definitive conclusions. Future research using three-armed randomised trials (eg, no contact control, generic PA intervention and tailored dog-walking intervention) are needed to answer these possible scenarios more definitively. The results would help identify whether the added cost of dog-related intervention content and tailoring is necessary/beneficial in PA campaigns.

One of the most noteworthy potential moderators, through our exploratory analyses of the findings, was the tendency for larger sample studies to report lower effectiveness estimates. While this finding was also linked to the above-mentioned content differences in the comparison arms, it suggests the presence of publication biases. It is therefore crucial for the field’s development to move away from feasibility/pilot trials towards adequately powered larger trials examining the effects of dog-walking interventions on human PA behaviour change.2 Such studies will provide the evidence required to pragmatically assess the scalability of dog-walking interventions as a means for population health improvements.12

We found that interventions with supervision and a face-to-face component of delivery may have produced larger behaviour change effects than distance-based and community interventions. This finding is commensurate with past reviews of PA interventions47 48 but poses a challenge to scalability and reach. It may be worth exploring hybrids of these approaches as these had success in some of the studies reviewed.36 37

Finally, we noted that half of these reviewed studies were not based on an explicitly stated theoretical framework. Among better designed studies, there was a tendency for theory-based interventions to elicit more meaningful increases in dog walking compared with those that did not state a theoretical framework. While the role of explicit theory in improving behavioural interventions is debatable,49–51 it was clear that almost all of the theoretical approaches in this literature at present are couched in the social–cognitive tradition (see Rhodes et al 52 for a review of approaches). This was explicitly stated in 6 of the 13 interventions and obvious by the types of BCTs employed among the studies. For example, behavioural instructions on dog walking and social support were included in over half of the studies (eg, information on dog walking for canine health, places to walk dogs and walking groups), and problem solving (listing barriers and forming solutions), behavioural demonstration (practice walks with instructors) and information about health consequences (information about how PA affects health outcomes) were used in almost half of the interventions. These approaches attempt to address rational expectations about outcomes, perceptions of capability, social connections and self-regulation tactics as the cornerstones of behaviour change.53 There was evidence for the effectiveness of these approaches in these studies and within the larger PA intervention literature.2 54 Still, future research may improve on these strategies by also targeting automatic/reflexive factors (eg, habit, affective response and identity) or broader socioecological environment and policy factors (dog parks, walking areas for dog owners and off-leash sites). The former suggestion had some apparent success in one of the studies within this review,27 37 while the latter approach is backed by considerable observational research evidence.12 21

Strengths and limitations of reviewed studies

The majority of reviewed studies had considerable strengths, including randomised trial methodology with direct assessments of PA, and applications to diverse populations. Still, the current literature also has several weaknesses. The available literature at present is composed of primarily small sample, short-assessment, proof of concept/feasibility interventions.

Strengths and limitations of the current review

Our paper used systematic review reporting standards. Still, there were some limitations of the review methods. Because of the limited number of studies, we mixed a variety of populations and settings into the same analysis. It will be important to separate these interesting aspects with formal testing (once this literature matures) in order to provide evidence that is more conclusive. This literature review is limited by the search terms and search engines employed, as well as the English language restrictions. Our analysis methods were also limited to a median effect size estimate and vote counting of the direction of effects at present, given the heterogeneity of the methods employed. This means that the effect size estimates are not corrected for sampling bias and thus do not represent a weighted estimate. While this is an appropriate first assessment, a more refined analysis in the future involving formal meta-analysis with subsequent formal moderator analyses will give more concrete answers to the questions or review addressed.


To our knowledge, this is the first systematic review on the effectiveness of canine-based interventions to increase human PA. The 13 studies we synthesised showed that these interventions do increase regular walking, but it is not clear whether the dog-related tailoring of the intervention is more effective than more generic walking intervention content. We recommend that future research move beyond feasibility and proof of concept studies to increase rigour and quality, and longer follow-up. We also recommend that researchers expand on the range of theoretical approaches and BCTs used in these interventions.

What is already known

  • Dog ownership is estimated at over 30% of households in high-income countries, yet approximately half of owners do not walk their dogs regularly.

  • This is the first systematic review of physical activity (PA) interventions focused on dog walking.

What are the new findings

  • The 13 interventions identified in the literature review showed evidence of increases in dog walking over time.

  • Whether dog walking-specific content results in increased PA over generic content is inconclusive at present.

  • Most studies were proof of concept or feasibility in nature. Future research needs to explore dog-walking interventions using larger sample sizes in diverse community settings.


Supplementary materials


  • Contributors RER, CW and HC conceived of the paper and developed the search strategy. MB provided the main searches, wrote the Methods section and developed the tables. RER performed the analyses and wrote the paper. All authors edited the paper and gave the final approval of the manuscript.

  • Funding RER is supported by funds from the Canadian Institutes of Health Research, the Heart and Stroke Foundation of Canada, the Social Sciences and Humanities Research Council of Canada and the Canadian Cancer Society. HC is supported by an Australian National Heart Foundation Future Leader Fellowship (number 100794).

  • Competing interests None declared.

  • Patient consent for publication Not required.

  • Provenance and peer review Not commissioned; externally peer reviewed.