accessibility score) may result in measurement error: participants living closer to public transport may be more likely to use and walk to
public transport, but those users living further away who do walk, are more likely achieve recommended physical activity levels through
the longer walk distance to the stop or station.
Although models were adjusted for individual level and environmental covariates known to be associated with the exposure and
outcomes, confounders may still have been omitted or mismeasured. In addition, the lack of data on public transport use and other travel
behaviours also significantly limits the scope of possible conclusions about if and how public transport accessibility is associated with
walking. Are the increased odds of walking observed with higher public transport accessibility exclusively an effect of public transport use
and related walking? Or do users and non-users living in areas with higher public transport access also walk more for other reasons, such
as the journey to and from work?
Data underpinning the walking outcomes were self-reported and therefore subject to errors of recollection and report. The Active
Australia walking question is a global measure of weekly walking, conflating transport and recreational walking, so increases in the odds of
walking with increased public transport accessibility were across both these domains. If the amount of time spent walking was classified
separately for recreation or transport it is likely that stronger associations between PTAI and walking would have been observed. The
question also does not differentiate between walking in the home neighbourhood and other environments, reducing its specificity in
measuring walking associated with local environmental characteristics. This may have also led to an underestimation of observed effects;
it is possible that the odds may have been higher if transport related walking in the home environment was the specific outcome
measured. In addition, because of the large number of null and small values, the walking outcomes were classified as binary, and as such
they may have been underpowered to detect small dose-response changes in the association between PTAI and walking.
Although the original AusDiab sample was large and randomly selected, the 5241 participants that formed the sample for this study
were only one quarter (26.6%) of those eligible to participate in AusDiab wave 1 in 1999. Participants who underwent the biomedical in
1999 were more likely to have completed school, technical and tertiary education suggesting that they may have tended to have higher
socioeconomic status than those who didn’t (Dunstan et al., 2002). The results could be biased if the associations between the PTAI
exposure and the five outcomes differed for participants and non-participants. The extent to which this is a problem is unknown.