This research showed that age (in both models) and gender (in the cross-sectional model) were the variables most strongly related to BMI. However, the importance people place on reasons for choosing their neighbourhoods were also related to BMI. In particular, choosing neighbourhoods that make walking easier was associated with lower BMI in the cross-sectional model. This variable had as strong a relationship with BMI as other variables that are known to contribute to weight such as physical activity . In the longitudinal model the interaction between moving status and the importance of choosing a neighbourhood for ease of walking was more strongly related to BMI than was change in physical activity. Specifically, participants who moved and rated ease of walking as not important had larger increases in BMI compared to those participants who moved and rated ease of walking as important. For those participants who did not move, there was not a significant change. These results reflect qualitative research in which developers reported believing that objectively measured walkable neighbourhoods attract walkers and if you are not interested in walking a walkable neighbourhood will not make a difference . These findings are also similar to the cross-sectional findings of Frank et al.  who reported people who preferred a highly walkable neighbourhood were the least likely to be obese whereas those who preferred low walkable neighbourhoods were more likely to be obese. Thus, research should further investigate the characteristics of people who move to walkable neighbourhoods. Is it a case of people choosing neighbourhoods that fit with already held values or that features within the built environment can influence the behaviour of those previously not interested in walking?
This question is important given that objectively measured walkability was not a significant influence on BMI in either of our models, contradicting the findings of some researchers [5–8] while supporting a growing body of longitudinal evidence that has not supported the relationship between neighbourhood walkability and obesity . However, neighbourhood walkability may influence physical activity [11, 12] and this may occur despite no change in weight. Therefore, more longitudinal research is needed that examines the relationship between objectively measured neighbourhood walkability and change in physical activity behaviour, while controlling for questions of choosing a neighbourhood because it is perceived to be walkable.
Being close to outdoor recreation facilities was the other neighbourhood choice variable significantly related to BMI change but the results were mixed. For participants who moved, the more participants indicated that choosing a neighbourhood to be close to outdoor recreation facilities was important, the greater their BMI increased over the six years. The opposite relationship was found for participants who did not move and this variable was not related to BMI in the cross-sectional model. The longitudinal findings for movers are similar to those of Boehmer et al.  who reported lower odds of obesity were related to the perception of having fewer recreational facilities nearby (including both indoor and outdoor facilities such as trails, parks, and recreation centres). Future research should examine this relationship as it may reflect the "urban paradox" outlined by Lopez and Hynes . That is, in lower SES neighbourhoods using outdoor recreation facilities such as parks may be considered risky or unappealing.
It is also important to consider that people who live in lower SES neighbourhoods likely have less ability to choose their ideal neighbourhood. It is also possible that our participants chose to live near outdoor recreation facilities for the benefit of other family members (e.g., children) rather than for themselves. It is also possible that people drive to outdoor recreation facilities and are not particularly active once they are there. Clearly, more research is needed to disentangle these findings. Indeed, the issue of neighbourhood SES is important to address as our findings support the conclusions of a systematic review that reported neighbourhood SES was related to obesity . Our findings further contribute to the growing body of longitudinal evidence [15, 19] showing low SES is associated with increased likelihood of being overweight. Although more research is needed to fully understand how neighbourhood SES contributes to obesity, it is without question that individuals in socially disadvantaged neighbourhoods face more barriers to health than their wealthier counterparts.
Another important variable that emerged was activity status. In both models, activity status was related to BMI. In the longitudinal model, those participants who increased their activity levels over six years showed a decrease in BMI. In the cross-sectional model, more activity was correlated with lower BMI. These findings are consistent with other research . Creating more opportunities to be active can help problems of obesity. Our measure of activity did not capture activity related to walking, and in particular, walking for transportation or recreation. This may help to explain why we found no influence of neighbourhood walkability on BMI. Another potentially important variable in the weight equation, fruit and vegetable intake, significantly predicted BMI in the cross-sectional but not the longitudinal model. The results from our cross-sectional study replicate similar studies . It may be that the findings were not significant in the longitudinal study, due to the short one week measure six years apart.
The inclusion of two separate studies, one of them longitudinal, is a strength of this research, providing a greater breadth of evidence. However, there are some limitations that should be noted. First, BMI was measured by self-report. Given that people tend to under report their actual weight , self-reported BMI usually provides an under representation of overweight and obesity at a population level. However, we have no reason to believe this bias differentially affected our findings. It should also be noted that the changes in BMI over 6 years were relatively small (i.e., less one BMI point) but the trends found in this research should be considered and examined in research that takes place over a longer period of time given the implications greater body weight have for health. Another limitation is the measures of physical activity and fruit and vegetable consumption, both of which ask participants to recall their behaviour over one week. This was particularly problematic in the longitudinal model where behaviour across one week was measured six years apart. An additional limitation is that we do not know when our participants moved in the longitudinal study. This is an important factor to consider in relation to BMI change. Further specificity of when participants moved would be of benefit and future research should take the recency of the move into account.