The results of this multilevel study provide important evidence that the relationship between park area and levels of walking is far from straightforward. Contrary to what we expected, respondents were less likely to walk on a weekly basis when they lived in areas with high amounts of parkland. These effects arose at both the 800 m and 1200 m buffers, but there were no significant findings at the 400 m buffer. Importantly, the size of parks appears to make no difference - results for ‘all parks’ and ‘larger parks’ were similar.
Of the studies looking specifically at the relationship between park access and walking, results have been mixed. Several studies have found positive associations between access and walking [29–31, 35], others have found no significant relationship [32–34], negative associations , and mixed results [26, 36]. Panter and Jones  reported results similar to ours when they found that there was a non-significant trend for those living closest to parks to be less likely to walk five or more times a week. Looking more broadly at the relationship between parks and physical activity or health generally (rather than walking), there are examples of similar counter intuitive [38, 51, 52], or non-significant negative relationships [21, 34, 53].
While the results are perplexing, there are a number of potential explanations. Firstly, it is highly possible that the types of neighbourhoods containing large areas of parkland differ from those offering less area of parkland. The results we observed may reflect area differences in urbanisation, street connectivity, population density, or land use mix. These are all aspects of the built environment that have previously been associated with walking levels [54–57]. If the areas in our study that contained high levels of parkland were also in outer suburbs with poor connectivity, few destinations and little diversity in land use mix, then we would expect results similar to those attained. Additionally, it is possible that those respondents with less park area in their neighbourhood may have had other destinations such as cafes, schools, shops and community facilities that they walked to. Such destinations, especially food shops and schools, are likely to be frequented by a higher proportion of the population, more frequently than recreational destinations. In support of this, Cerin and colleagues  found that residents in more commercial/industrial areas reported significantly more walking for transport than residents of areas with a recreational profile.
Secondly, we expected that the amount of park area would encourage walking by offering a destination for people to walk to and in, by improving the aesthetics of the neighbourhood, or alternatively by offering a short cut for people to walk through (compared to a street network journey). However, it is possible that parks are not regarded as a destination to walk to on a regular basis. Rather, it is possible that parks are seen as a place to kick the football in, play the sport prescribed by the space (i.e. lacrosse, if on a lacrosse field; football if on a football field), or have a picnic in. Furthermore, it is conceivable that path legibility has an over-riding impact on walking: people walking for transport may be more likely to seek direct routes through streets, and those walking recreationally/for exercise may seek clear/smooth pathways (i.e. not across ovals/grassland). In support of this, walking in the neighbourhood has been found to be influenced by footpaths, walking paths, local shops and perceived safety , as well as walking track length, having paths located closer to roads, and a greater number and variety of destinations . Parks may in fact, offer less direct routes than the connectivity of an urban grid – they may be fenced or bounded by waterways or train lines.
Finally, our park dataset contained no information on qualitative aspects of parks such as the facilities offered by the different parks, or the perceived safety or aesthetics of the parks. Our dataset did not distinguish between parks with unkempt grass, parks with dense tree coverage, parks with manicured garden beds, and parks with adventure playgrounds. There is evidence that park usage is influenced by facilities/amenities [24, 59], and aesthetics/attractiveness [30, 59]. Evidence of qualitative differences in parks influencing usage also comes from the United Kingdom, where respondents closest to a formal park (with a structured path network and organised layout) were more likely to be sufficiently active, but other types of parks had no significant effect on being sufficiently active . It is therefore possible that park quality may have varied across our sample, and influenced our results.
This study improves on previous studies in a number of ways. Firstly, most studies investigating park accessibility in terms of park area have either used neighbourhood level measures of park area (i.e. park area as a percentage of city acreage), or used a single buffer distance, (rather than multiple buffers). In this study, we used three different buffers specific to each respondent, and are therefore better able to understand at what distance park area may influence walking behaviour. Secondly, most previous studies have investigated the total amount of park area, which, depending on the source of geo-referenced parks, may include a vast number of small parks that may have little impact on activity levels. We use two measures of park area: total park area (all parks) as well as the area of larger parks in an attempt to understand whether the size of the park is important in encouraging walking. Thirdly, few studies have measured park area specific to each individual. Increasing sophistication of GIS technology is enabling increasingly complex analysis of neighbourhoods. By calculating buffer areas for each individual, we were able to investigate neighbourhood effects with much greater specificity than would otherwise be possible.
There are some limitations of this study. Firstly, as with all cross-sectional studies, any significant associations arising from the analysis cannot be interpreted as suggesting causation. Secondly, we did not control for the self-selection of people who perceive that there are benefits to physical activity, and select into an area that supports physical activity. If self-selection exists, but is not controlled for, there is the risk that the effect of the built environment on travel behaviour is estimated incorrectly. Importantly however, a recent paper examining the influence of self-selection on the relationship between park area and walking found that self-selection did not exclusively explain the relationship, and that those who placed greater importance on neighbourhood parks, were not more likely to live near more open space . Thirdly, walking has been shown to vary by walking purpose . It may be argued that the fact that we did not distinguish between walking purpose in our analysis may have led to imprecision or mis-estimation of the effect of park area on walking. In consideration of this, we tested for an interaction between park area and walking purpose, and found none. We also ran separate models for transport walking and recreational walking. However we found that our results held for both types of walking, and therefore chose to use total walking frequency as the outcome measure. Importantly too, and in defence of this general measure of walking, it is often difficult for both respondents (and analysts) to distinguish between walking trips on the basis of purpose. This may be particularly the case for parks, where it is conceivable that confusion may arise due to the ‘recreational’ nature of parks. If a person was to walk to the park with the intention of kicking the football, should they be classified as walking for recreation or transport?
A further limitation of the study is our measure of disability/injury, which did not distinguish between the type of injury or disability suffered by respondents, nor the extent to which it affected their daily life, in particular their mobility and capacity to walk. We ran separate exploratory analyses excluding respondents with a disability or injury. We found that the results did not change substantially other than an increase in the strength of the association between park area and walking for the 400 m radius, but that this remained non-significant (All Parks: p = 0.11 for tertile 1, p = 0.84 for tertile 2; Large Parks: p = 0.398 for tertile 1, p = 0.199 for tertile 2). As there was little effect on the results, we retained these respondents in the model, but included this variable as a confounder. Finally, our use of Euclidean buffers constitutes a potential limitation. While the use of Euclidean distance is considered more appropriate in determining open space access, it is possible that some pedestrians encounter network barriers en route to parks; Euclidean buffers may therefore overstate access.