• No results found

Chapter 7 Analysing Car Users’ Perceptions

7.3 Multinomial Logistic Regression Analysis with Clustered Data

The next step of the analysis was to use multinomial logistic regression considering the 5 clusters obtained from MCA and 3 factors from PAF. This analysis explores whether any relationships exist between clusters and factors and if any patterns and trends occur over 4 successive years (2011 to 2014). Given that Cluster 1 was unique in being solely of older car users, for this analysis, therefore, Cluster 1 was chosen as a baseline or reference category to investigate how attitudes towards and perceptions of environmental issues of other car users differed from older car users. Clusters were

factors relative to the older male retired car users. The following text serves to identify key messages that emerged from this analysis presented in Table 7.11.

The attitudes to transport and environment (factor 1), consisting of car users’

perception towards reducing the amount of car use and intention towards buying a car with lower CO2 emission in the future, of clusters 2, 3, 4 and 5 were statistically significant at 90% - 99% confidence level in 2011 only, but in 2012 to 2014 were seen as less important. In 2013, car users were found to be aware of the sustainability associated with owning or using sustainable modes such as public transport, cycling and walking for a short journey of less than 2 miles for the environmental benefits. This statistically significant relationship was observed for car users in cluster 2 with factor 3 (modal shift potential), confirming middle–aged males in full–time employment have become more aware of the need for mode shift and the car users in this cluster are a favourable target for any campaign to encourage mode switch to sustainable modes.

There was a high level of statistical significance between clusters 2, 3 and 4 with factor 2 (traffic awareness) every year with statistical significance at 95% – 99% across all four years. The reason that all three factors are statistically significant for car users in cluster 3 in 2011 is possibly due to their maturity, being in a higher income bracket, in full–time employment and having a strong affinity towards car use which is exhibited by their attitudes to transport and environment, traffic awareness, and also travel mode shift. Unlike cluster 3, all factors are not significant for respondents of cluster 5 in 2012.

Within cluster 5 (Middle–aged (35-44), female, looking after the home), traffic congestion and exhaust fumes in towns, cities and motorway (factor 2) became more significant from 2013 to 2014. Although the p-value does not change greatly (0.00 to 0.01), it may represent a noteworthy change in attitudes among this particular group.

Initially there was a statistically significant positive relationship with attitudes to transport and the environment in 2011. However, due to active use of cars and rare use of sustainable modes, this factor seems to be less important for middle–aged females who were looking after the home in cluster 5. This may be due to more responsibility for children putting them in situation of needing to travel by car at peak times. However, this is a conjecture as the data is not available in the BSA to explore this further.

171

ctors EFA2011201220132014 Bp-value Bp-value Bp-value Bp-value ttitudes to transport and environment 1.770.04b0.850.471.040.890.870.58 raffic awareness0.370.00a0.470.00a0.560.03b0.510.01a odal shift potential tercept 0.74 -0.79 0.18 0.000.85 -0.65 0.41 0.000.61 -1.11 0.04b 0.000.98 -1.31 0.94 0.00 ttitudes to transport and environment 1.540.03b1.350.08c1.060.781.230.19 raffic awareness0.510.00a0.460.00a0.660.03b0.640.01a odal shift potential tercept 0.82 0.460.03b 0.000.80 0.020.16 0.280.83 0.110.24 0.430.82 -0.07 0.18 0.57 ttitudes to transport and environment 1.700.01a0.800.281.400.121.060.77 raffic awareness0.620.01b0.630.02b0.720.10C0.540.00a odal shift potential tercept 0.82 0.170.24 0.290.84 -0.33 0.33 0.041.00 -0.26 1.00 0.111.21 -0.56 0.28 0.00 ttitudes to transport and environment 1.710.07c0.960.870.120.710.540.88 raffic awareness0.720.240.750.240.390.00a0.240.01a odal shift potential tercept 0.78 -1.15 0.32 0.001.10 -1.10 0.69 0.001.24 -1.33 0.38 0.000.45 -1.74 0.26 0.00 McFadden Pseudo R2 No of samples0.03 3650.03 4000.03 3410.03 403 nificant at 99%, b significant at 95%, andc significant at 90% aged (35-44), male, full–time employee adults (45-64), male, full–time employee and middle–aged (35-44), female, full–time employee aged (35-44), female, looking after the home category is cluster 1 (older–aged (65+), male, retired) LR analysis examining the effects of respondents’ travel behaviour relative to senior citizen through relationships between the clusters and factors

Back in 2011, all clusters were significantly more positive towards attitudes to transport and environment (factor 1) compared to the older male retired group who were really reluctant to give up their cars. They were more likely to be thinking positively about transport and the environment in 2011, and then year on year that slowly waned. However, cluster 3 who were mature male adults in a small household with only one car, were also more positive than the retired group in 2012. Nevertheless, it was no longer statistically significant in the following years.

This brings two messages: either the older generation has become more aware of environment and the need for sustainable transport and environment, given that the majority will possess a free bus pass, or as the population has aged over years, these factors become less important to them. They own cars so they use them.

In 2011, all clusters were found to be more aware of traffic congestion (factor 2) than the older group. This could be because the older group travel less mileages by car compared to the other groups. Respondents in cluster 2 and cluster 3 who were males, 2 person households, likely childless couples owning one or two cars, were statistically significantly consistently aware of traffic congestion throughout the year 2011 – 2014.

This probably reflected the amount of travel (probably day-to-day commuting) and dependency on the cars of these two groups. Respondents in cluster 2 could be aware of traffic congestion of cycle users, as they were reported as frequent cycle users.

Interestingly, respondents in Cluster 2 relative to the older group in 2011 were not statistically significant with modal shift potential (factor 3), but by 2013 had begun to accept the need for action to switch travel from private to sustainable modes for a short journey of less than 2 miles, with evidence of occasional use of cycles. Meanwhile, respondents in Cluster 3 show a reverse trend, accepting the need for action to switch transport modes in 2011 only. Therefore, the car users in Cluster 2 are a favourable target for any campaign to encourage mode switch to sustainable modes. They were estimated to be the most prone to take action to help reduce the impact on environmental and climate change problems and were the most likely groups to be willing to change travel modes.

7.4 Conclusions

This chapter has presented an extensive analysis of the categorical variables using MCA and MLR, which aims to investigate the groups within the population with greater levels of concern and awareness of climate change and environmental problems. Changes in attitudes to and perceptions of climate change over time were also sought in order to decide which group of travellers would be willing to take action to help reduce their impact on climate change.

By using MLR, each cluster within the population with the higher levels of mode shift potential over time (significant value) was sought in order to decide which group of car users can be used to produce targeted travel behaviour campaign and acknowledge which cluster is more susceptible for sustainability. The results confirm that the population is not uniform in terms of their attitudes and motivations in relation to reducing CO2 emissions from personal travel, therefore policies and universal solutions to encourage more sustainable transport behaviours are deemed unlikely to be effective. The results also show that the cluster groups that exist are not defined or differentiated by demographic features alone. Motivations and barriers to change in travel behaviour and to use alternative modes differed widely between the groups. A degree of influence from environmental concerns was found for all groups.

When the datasets from 2011 to 2014 were combined together to gain further insights into groups with similar views, 5 clusters emerged from the analysis. Differences were found in each year, yet similar views were seen spread throughout the five clusters which differed in demographic characteristics and views. The evidence recommends that acknowledgement of the concept of climate change among the car users were high.

The results demonstrated that different reasons could influence the same behaviour;

however, different behaviours could lead to the same attitudes.

MCA additionally identified various significant correlations of socio-demographic attributes with travel behaviour. The hierarchical cluster analysis distinguished 5

that the respondents in Clusters 2 and 3 were more sensitive to factor 3 “modal shift potential”. On the other hand, even though almost all of the respondents from 2011 to 2014 strongly agreed that human actions are partly responsible for the impact of climate change, the results also revealed an opposite trend, because almost all groups paid less attention to environmental problems in the later years during this period.

The respondents’ demographic characteristics such as age, gender, employment status, household size and car ownership play an important role and are revealed to have a significant effect on travel behaviour patterns and the willingness to switch to other transportation modes. This is similar to Fatmi and Habib (2017) where they found that bigger household size and driver’s licence also influence travel modes switch decisions.

In the next chapter we will describe the use of log-linear and multivariate probit models, fitted using Bayesian inference.

The model so-far developed using multinomial logistic regression does not consider the correlations between responses to several questions. Therefore, in the next chapter, we will use a multivariate probit model (MPM) which will allow us to include the ordinal responses to several questions, which can be correlated, in a single model. This will allow us to look at the whole collection of responses from an individual and relate this collection to explanatory variables.

Furthermore, in the next chapter, we will adopt Bayesian inference. This will allow us to do such things as computing predictive probabilities of responses, given particular values for explanatory variables, in a way which allows for both the sampling variation between individuals and the remaining uncertainty in the values of model parameters.

The practicality of this approach will be demonstrated and show that the methodology could be applied to different datasets in the future. For example, the methodology could be applied to research conducted using datasets from developing countries to compare with the results from developed countries.

Chapter 8 Bayesian Inference Approach