Study design & participant recruitment

The present analyses use data collected from the iConnect study. Full details of the study methods, including the survey instrument and its development, are provided elsewhere [20, 21]. In brief, the iConnect study is a longitudinal study investigating the impact of newly constructed infrastructure for walking and cycling on travel, physical activity and carbon emissions. In April 2010, adults aged over 18 years (n=22500) residing in the three core study areas (Cardiff, Kenilworth and Southampton) were randomly selected from the edited electoral register and sent a survey which asked about their travel and physical activity behaviour and included standard sociodemographic questions. One year later (April 2011), the respondents were sent a second copy of the survey. Ethical approval was granted by the University of Southampton Ethics Committee.

Exposure measures

At both time points, respondents were asked to recall all journeys made over the previous seven days for journeys to and from work, to and from a place of study (categorised as commuting travel), for shopping and personal business and for visiting friends or relatives or other social activities (categorised as non-commuting travel). For each of these purposes, respondents recalled the number of journeys made as well as the total time (min/week) spent and distance (miles/week) travelled using each of six modes of transport (walking, cycling, bus, train, car and ‘other’). Participants were instructed to assign each return journey to a single ‘main purpose’ and report all modes used on each journey. Where a respondent had reported the distance but not the time travelled, time was imputed using the sample mean of the observed speed for each mode/purpose combination. Time spent walking and cycling for commuting and non-commuting purposes, and for all purposes, was summed to provide aggregate measures of active travel.

Change in active travel was calculated by subtracting weekly time spent in active travel at baseline from weekly time spent in active travel at follow-up. To account for the likely imprecision in the measure respondents with a change value of ±15 min/week were categorised as having maintained their active travel, those with a value of >15 min/week as having increased their active travel, and those with a value of <−15 min/week as having decreased their active travel. To explore possible dose–response relationships, respondents were also categorised according to the magnitude of this change using cut points of ±15, ±45, ±90 and ±135 min/week. Change values were also computed for four separate sub-exposures: change in (a) commuting active travel, (b) non-commuting active travel, (c) walking for all transport purposes and (d) cycling for all transport purposes.

Outcome measures

Recreational physical activity was assessed by asking participants to recall the number of sessions and total time spent over the past seven days in four types of recreational activities: walking, cycling, other moderate-intensity physical activity and other vigorous-intensity physical activity. The questions were adapted from the short form of the International Physical Activity Questionnaire (IPAQ) so that they specifically asked about recreational walking and cycling separately [22]. Respondents were asked to recall the number of sessions and total time spent over the past seven days in the four types of recreational activities. Data cleaning procedures similar to those applied to the short IPAQ were used whereby for each activity category, data were truncated at 1260 min (21 h/week) and respondents who reported greater than 6720 min (16 h/day) were excluded [22]. Total recreational physical activity (min/week) was computed by summing time spent in these four activities. Total physical activity (min/week) was computed by summing time spent in all active travel and in recreational physical activity.

Sociodemographic and other characteristics

Respondents were categorised according to their sex and baseline age, body mass index (BMI, computed from self-reported height and weight), highest educational qualification, employment status, ethnicity, housing tenure, household car access and the presence of children in the household.

Analyses

The number of cases with missing data for sociodemographic covariates at baseline ranged from 16 (for sex) to 260 (for household income). Where possible, these missing baseline data were replaced with follow-up measures of the same variables for between 1 and 106 cases (depending on the variable), which reduced the number of cases with missing data for a given variable to between 0 and 114. Multiple imputation using chained equations (with five imputations) under the assumption of missing at random was used to impute the remaining missing data. All covariates and outcome variables entered into the regression models were included in the imputation model. Substantive findings were unchanged in sensitivity analyses using complete case analysis and/or excluding income, the variable with the highest level of missing data.

Descriptive statistics

A completed baseline survey was returned by 3516 (16%) respondents, of whom 1885 (54%) returned a completed follow-up survey one year later. Of these, 232 had missing physical activity or travel data and 25 had extreme differences (>600 min/week) in active travel between baseline and follow-up. Excluding these cases left 1628 respondents. The likelihood of being included in follow-up analysis was associated with being older (56.1% of baseline respondents aged over 65 were included, compared with 30.1% of 18-34-year-olds), owning a home (52.7%, versus 26.3% of those who rented privately) and having access to a car (48.4%, versus 31.5% of those with no car; see Additional file 1). Comparisons with local and national data suggested that on average, respondents were older and more likely to have completed higher education, to own their home, to have access to a car, to be of a normal weight and to be economically active, and less likely to have children living at home, than the general population (see Additional file 2).

Respondents had a mean age of 54 years and just over half were female (Table 1). Total time spent in active travel changed little over one year, with a mean change of −4.4 min/week (95% CI −11.7 to 2.8). Recreational physical activity declined by a mean of 28.2 min/week (95% CI −43.4 to −12.9) resulting in a mean decrease in total physical activity of −32.6 min/week (95% CI −49.8 to −15.4). At the individual level, weekly active travel increased in 32% (n=529), was maintained in 33% (n=534) and decreased in 35% (n=565) of respondents over the year.

Respondents whose active travel decreased reported an average of 230.8 min/week (95% CI 214.5 to 247.2) of active travel at baseline which decreased to 93.6 min/week (95% CI 82.2 to 105.0) at follow-up. Conversely, respondents categorised as increasing their active travel reported an average of 85.3 min/week (95% CI 75.7 to 95.0) at baseline, increasing to 218.6 min/week (95% CI 204.0 to 233.2) at follow-up. Those who maintained their active travel reported an average of 31.5 min/week (95% CI 24.3 to 30.7) of active travel at baseline. Of the ‘maintainers’ the majority (n=382) reported no active travel, while the remaining (n=152) reported an average of 110.7 min/week (95% CI 90.3 to 131.1).

Association between change in active travel and change in recreational and total physical activity

Table 2 summarises descriptive statistics and linear regression models showing the association between change in active travel and change in recreational and total physical activity. The comparison of unadjusted mean changes is further described in Figure 1, which shows the relationship between the magnitude of change in active travel and the magnitude of change in total physical activity.

Small decreases in recreational activity were seen across all groups (Figure 1), and there was no evidence that this varied by category of change in active travel (p = 0.47 for heterogeneity; Table 2). Furthermore, there was no correlation (r=0.05) between change in active travel and change in recreational physical activity. In fact, the direction of the non-significant effect was the opposite to that which would be expected if activity compensation were operating, with a larger decrease observed in those whose active travel decreased (mean decrease 39.7 min/week) than in those whose active travel increased (mean decrease 21.0 min/week).

By contrast, there was strong evidence that a change in active travel was associated with a change in total physical activity. Those whose active travel increased had a significant increase in their total physical activity compared with maintainers (adjusted regression coefficient 135.1, 95% CI 94.3 to 175.9), while those whose active travel decreased had a corresponding decrease in total physical activity compared with maintainers (adjusted regression coefficient −154.9, 95% CI −195.3 to −114.5; Table 2). This association showed an approximately dose–response relationship across the full range of change in active travel categories (Figure 1). Similar results were also seen when active travel was entered as a continuous variable.

By way of sensitivity analyses we explored whether the associations differed according to respondents’ absolute levels of active travel at baseline and follow-up, categorising respondents as: (a) maintaining zero active travel, (b) maintaining some active travel, (c) reducing their active travel but continuing to report some, (d) reducing their active travel to zero, (e) taking up active travel, or (f) increasing their active travel from non-zero at baseline. In linear regression models, the associations remained consistent with the main analyses. That is, a change in active travel was associated with a change in total physical activity regardless of whether the change was from ‘zero’ or from ‘some’ active travel at baseline (see Additional file 3).

Strengths and limitations

To the best of our knowledge this is the first study to examine the association between a change in active travel and change in total physical activity in adults. It builds further on previous work by examining the association in a relatively large sample and by including detailed disaggregated measures of both travel and recreational physical activity, although these were self-reported and subject to possible recall bias. To capture the necessary level of detail two existing instruments were adapted: travel behaviour was assessed using a seven-day recall instrument and recreational physical activity was assessed using a purposively adapted version of the IPAQ. Although the IPAQ has been extensively used, like many physical activity questionnaires its criterion validity with respect to objectively measured physical activity is modest and it tends to overestimate physical activity [26]. It is likely that the measurement error associated with the IPAQ and the seven-day travel instrument attenuated effect sizes. It is hard to know the relative magnitude of this potential bias, although it seems unlikely that it would be sufficient to disguise a genuine, substantial decrease in recreational physical activity among those whose active travel increased. Moreover, the criterion validity of the two measures when used together remains to be determined. Consequently, the extent to which respondents may have reported their walking and cycling as both recreation and transport remains unknown. Notably, however, change in active travel was not correlated with change in recreational physical activity, suggesting that ‘double counting’ is unlikely to have had much effect in this study. In addition, the measure of total physical activity derived for these analyses did not include time spent in occupational or household physical activity. It seems less plausible that activity compensation would occur in these domains than with respect to recreational physical activity given that in general, people have more control over the latter than the former. Finally, the items used to measure travel behaviour did not ask about the perceived intensity of the walking and cycling. This may be important in light of evidence suggesting greater health benefits from cycling at vigorous perceived intensity than from cycling at light or moderate perceived intensity [27].

The study achieved an acceptable follow-up rate of 54%. Nonetheless, the original response to the survey was low, albeit comparable to that achieved in other surveys implemented for the purpose of evaluating natural experiments [28, 29]. Although the travel patterns of respondents were comparable to those of adults participating in the recent UK National Travel Survey (NTS) [17], the sample was composed of respondents who were older and more likely to have completed higher education than the local populations from which they were drawn. Baseline data also suggest that respondents were more likely to meet the current physical activity recommendations than respondents in the 2008 Health Survey for England (HSE) [30], although these differences may reflect the use of a much more detailed, disaggregated measure of travel and recreational physical activity than those used in the HSE. Given the differences between the sample and the general population, the observed associations may not be generalizable to the wider population.

Finally, while the findings indicate that a change in active travel is associated with a change in overall physical activity, the direction of causality is unknown. It is not possible to conclude from these results whether an increase in respondents’ overall physical activity was reflected in an increase in walking or cycling, or whether an increase in walking or cycling was followed by an increase in overall physical activity.