We first of all present some descriptive analyses as a basis for discussing the regression models.
Descriptive analyses
The commute mode and time distributions are summarised in Table 4. The majority of people included in the sample drove to work (61%) with walking the second most common means of travelling to work (12%).
Table 4 Commute mode and time, sample distributions The mean (one-way) commute time across the pooled sample was 27 min, but there are large differences between modes. For example, rail commuters travelled for nearly an hour on average, compared to 25 min for drivers.
Changes to commuting behaviour
We examined the prevalence of commute mode changes occurring from one observation to the next and found that a change in commute mode occurred for 18% of observations. Consistent with a previous study (Clark et al. 2016), driving is found to be the most stable commuting mode compared to alternative options. For example, 91 per cent of car drive commutes were still being driven by the following observation, compared to just 12 per cent of walking commutes.
Substantial changes in commute time are most likely to occur for those who change the origin or destination of their journey or change their mode. Increases in delay owing to congestion are only likely to introduce modest changes in commute time from 1 year to the next. Table 5 shows that a change in origin or destination occurred from 1 year to the next for 15% of panel observations. The mean wave to wave change in commute time for those with no change in origin or destination was 6 min, compared to 16 min for those that changed the origin or destination of the commute.
Table 5 Prevalence of changes to the commute origin or destination Bivariate associations
Figure 2 shows the bivariate (between-individual) associations between commute time and each measure of SWB (for the wave 1–6 sample). This indicates that the satisfaction with job, leisure time and life scores are lower amongst those with longer commute times. The trend is strongest for leisure time satisfaction and only modest for life satisfaction. Strain is higher amongst those with longer commute times, and this is a contributing factor to slightly lower mental health (GHQ12) for those with longer commutes. The higher job and life satisfaction and mental health for those with commutes of over 90 min only applies to a small group, constituting just 1% of the sample. A limitation in the data is that we do not know how many times in the week such long commutes are performed. Self-reported health is higher for those with longer commute times.
Regression models
The results for the main regression models are summarised in Table 6, with the results for the working from home variant reported separately in Table 7. The additional tests of interaction performed to identify population sub-group differences are summarised in Table 8. We provide a detailed description of the results for each measure of SWB in this section. The key findings and their implications are then summarised in the “Discussion” section.
Table 6 Main models: correlated random effects model coefficients for commuting variables Table 7 Working from home models: correlated random effects model coefficients for commuting variables Table 8 Fixed effects model coefficients for commuting variables and interaction terms Job satisfaction
We found a negative association between job satisfaction and time spent travelling to work (every extra 10 min (each way) reduces job satisfaction by 0.011 points on the 7-point scale—Table 6). This is based on the within-individual commute time coefficient which provides the strongest indication of a causal relationship. The between-individual coefficient is consistent with the within-individual coefficient, providing confidence in the result. The coefficients for control variables (not shown in Table 6) show that job satisfaction increases with personal income and with certain job roles (self-employed and manager) and decreases when working long hours (over 40 h per week). We examined the magnitude of effect of commute time by comparing it against the effect of personal income on job satisfaction (see Appendix A for details). This suggested that an additional 10 min (each way) of commuting time is associated with the equivalent effect on job satisfaction as a 19% reduction in gross personal income, i.e. a £4080 reduction in gross annual personal income for a worker earning the sample median annual gross income of £21,600.
With respect to differences in the population (Table 8), the job satisfaction of females is more sensitive to longer commute times (although this is not significant at the 95% confidence level, p value of the interaction term is 0.12), as is the job satisfaction of workers with higher incomes (noting that job satisfaction increases with income). By contrast, the job satisfaction of young workers aged 16–29 (who tend to have lower job satisfaction than older workers) is not sensitive to commute time.
The commute mode is also shown to moderate the commute time relationship. Job satisfaction is higher for short bus journeys (compared to short journeys by other modes) but decreases by more for every additional minute of commute time by bus compared against other modes (Table 8). Beyond the commute time effect, walking to work is associated with higher job satisfaction on average by 0.10 points (on the 7-point scale, see Table 6, between-individual variation) compared to driving (after accounting for journey time differences and other personal differences such as income and occupation type), but for the same individuals we do not see walking to work associated with higher job satisfaction on occasions when they walk to work compared to driving.
Working from home is associated with 0.08 points higher job satisfaction compared to driving (on the 7-point scale-Table 7, within-individual variation).
Leisure time satisfaction
The regression models indicated that lifestyle factors that intensify personal time constraints are associated with lower leisure time satisfaction. This included longer commute times, as well as working long hours and having children (these all have negative and statistically significant coefficients). Every extra 10 min of commute time (each way) is associated with a reduction in satisfaction with leisure time availability by 0.030 points (on the 7-point scale-Table 6, within-individual variation). The between-individual coefficient is consistent with the within-individual coefficient, providing confidence in the result.
With respect to differences in the population (Table 8), we find that people in the middle income quintile are less sensitive to longer commute times with respect to their leisure time satisfaction than people with lower and higher incomes (although the confidence level is 90%). The association with commute time is the same for females and males, although it should be noted that females have substantially lower leisure time satisfaction scores than males. There are no differences in commute time associations by age group.
Walking to work is associated with higher levels of satisfaction with leisure time availability (by 0.08 points) compared to driving (after accounting for journey time differences-Table 6, within-individual variation). Commuting by rail is associated with higher leisure time satisfaction on average by 0.13 points (on the 7-point scale-Table 6, between-individual variation) compared to driving (after accounting for journey time differences and other personal differences such as income and occupation type), but for the same individuals we do not see commuting by rail associated with higher leisure time satisfaction on occasions when they use rail compared to driving.
Working from home is associated with 0.10 points higher leisure time satisfaction (on the 7-point scale-Table 7, within-individual variation).
Self-reported health
Longer commute times were not found to be associated with a difference in self-reported health, although Fig. 2 indicates that people that undertake longer commutes report good health. This suggests that those people in good health are more likely to take on a longer duration commute.
We found that cycling to work is associated with higher self-reported health compared to driving (after accounting for differences between individuals, including commute time-Table 6, between-individual variation), but for the same individuals we do not see cycling to work associated with higher self-reported health on occasions when they cycle compared to driving. In a similar way, commuting by bus is associated with lower self-reported health compared to driving.
Strain
Longer commute times are associated with increased strain (by 0.004 points on the 4-point scale for every extra 10 min each way-Table 6, within-individual variation) with higher strain also associated with working long hours (over 40 h per week), a management job, a higher personal income, having children and having a longstanding health condition (based on coefficients for control variables which are not shown in Table 6).
With respect to differences in the population (Table 8), females are associated with higher levels of strain than males (by 0.12 points on the 4-point scale, not shown in Table 6), but the positive association of strain with commute time applies more strongly to males (increasing strain by 0.007 points for every extra 10 min (each way)). There are no differences in associations of strain with commute time by age or income quintile.
Walking to work is associated with lower strain compared to driving (by 0.047 points on the 4-point scale-Table 6, within-individual variation) and this is equivalent to the increased strain (of 0.041 points) associated with having a management role. The positive association between strain and walking to work found for between-individual variation contradicts this result and suggests the need to examine the within-individual relationship further in future when more waves of data are available. Commuting by bus is associated with lower strain compared to driving based on between-individual variation (Table 6), but no statistically significant association is found for within-individual variation.
Mental health
Not many personal factors were found to be associated with the GHQ12 measure of mental health, but longer commute times were found to be associated with lower mental health (by 0.028 points on the 36-point scale for every extra 10 min each way-Table 6, within-individual variation). Females have substantially lower mental health than men (by 1.02 points) but are no different in sensitivity to longer commute times (Table 8). This contrasts with the result of Roberts et al. (2011). Their analysis of the same GHQ12 measure (for an earlier time period) indicated that longer commutes were more detrimental to the mental health of women than men.
The commute mode was shown to moderate the commute time relationship. Every additional minute of commute time is associated with a greater decrease in mental health when people use the bus compared to driving. On the other hand, commuting by bus is associated with higher mental health scores compared to driving (Table 6—between-individual variation) for which reasons are unclear.
Life satisfaction
Our models confirmed that living with a partner, being in good health, financial security, and not being in middle age are associated with higher life satisfaction (as has been shown by previous studies). The fixed effects coefficient is not significant for commute time, i.e. for the same individuals we did not find that longer duration commutes are associated with lower life satisfaction (and no differences were identified by population sub-group-Table 8). The between-individual coefficient is negative and statistically significant, indicating that longer duration commutes are associated with lower life satisfaction after accounting for other differences between individuals (by 0.014 points on the 7-point scale for every extra 10 min each way—Table 6).
The finding of a statistically significant between-individual association but not within-individual association merited further investigation. It implies individuals with varying commute times over the six observations did not tend to have lower life satisfaction when their commutes were longer. They might have been compensated by better jobs or housing when this was the case. A separate analysis we conducted comparing those who persisted with long commutes (over 45 min) for the six observations with those who persisted with short commutes (45 min or less) showed that the former had consistently lower life satisfaction. This indicates there is a group of commuters who accept unfavourable commutes and are unable or unwilling to change them despite adverse impacts on their life satisfaction (for further details see Chatterjee et al. (2017)).
Commuting by rail is associated with higher life satisfaction compared to driving (Table 6) in terms of between-individual variation but the opposite association is shown for within-individual variation. This implies that rail users tend to have higher life satisfaction (for reasons not captured by our control variables) but switching to rail by individuals is not associated with an improvement in life satisfaction.
Working from home is not associated with any difference in life satisfaction (Table 7) in terms of within-individual variation, although the between-individual association indicates working from home is associated with lower life satisfaction, potentially due to characteristics of this group that were not captured by our control variables.
There are no notable differences in associations of life satisfaction with commute time by commute mode, gender, age or income quintile.
Life satisfaction mediation analysis
Overall, the results indicate the absence of a direct within-individual association between commute time and life satisfaction. However, life satisfaction would be expected to be associated with the SWB sub-domains which have been found to be negatively associated with longer commute times; namely job satisfaction, levels of strain, and leisure time availability.
Table 9 shows the results of model variants estimated on life satisfaction, in which the SWB sub-domains that are negatively associated with commute time have been included as additional controls. In models (1)–(3) the three SWB sub-domains are included individually, while all three sub-domains are included together in model (4).Footnote 7
Table 9 Life satisfaction mediation models: correlated random effects model coefficients The results indicate that the relationship between commute time and life satisfaction is mediated through leisure time satisfaction. When leisure time satisfaction is included, the commute time coefficient switches from negative (and non-significant) to positive (and significant), i.e. longer commute times are associated with reduced leisure time satisfaction which in turn is associated with reduced life satisfaction (since leisure time satisfaction is positively correlated with life satisfaction). After accounting for this mediation pathway, the residual effect of longer commute times is found to be positive. This implies that there are unobserved compensatory factors associated with longer commutes—such as improved residential and/or employment situations which are not captured in the control variables we had available.