Estimation and validation of hybrid choice models to identify the role of perception in the choice to cycle
Sottile, Sanjust di Teulada, Meloni, Cherchi, 2019, in International Journal of Sustainable Transportation
doi:10.1080/15568318.2018.1490465
Location |
Cagliari, Italy |
Population |
Other (specify) |
Sample size |
2752 |
Factor analysis type |
confirmatory factor analysis, none rotation |
Stepwise regression |
no |
Removal of insignificant variables |
yes |
Reviewed by |
LCM |
Abstract
Cycling is one of the most sustainable and ecofriendly modes of travel and a good form of exercise. Many government and public health authorities recommend cycling to stay fit as well as to reduce air and noise pollution, CO2 emissions, traffic congestion, and other negative consequences of car use. In light of these benefits, a major challenge for researchers today is how to promote cycling. However, in countries where cycling is not common, apart from the need for proper cycling facilities, one major issue concerns people’s perception of cycling for sport or recreational activities rather than as a mode of transport. The aim of this paper is to explore the role of perception in the likelihood of the bike being used for utilitarian purposes. We focus on the perception of: the bicycle as a means of transport; bikeability (in terms of usefulness and safety) and of bike infrastructure. Hybrid Choice Models (HCMs) have been used to estimate the effect of people’s perception on the propensity to bike. The HCM also accounts for the serial correlation between error terms in the discrete and latent perceptions, to allow for agent-common unknown factors. Furthermore, we also validate the model results using a hold-out sample and discuss some policy measures aimed at changing travel behavior. The results suggest that, besides individual characteristics, latent aspects related to the perception of the context and of the bicycle as a means of transport strongly affect the propensity to cycle. © 2018, © 2018 Taylor & Francis Group, LLC.
Factors
Models
Dependent variable |
Decision to cycle |
Model type |
Hybrid choice model |
Sample size |
2202.0 |
R2 |
0.441 |
Adjusted R2 |
|
Pseudo R2
(nan)
|
nan |
AIC |
nan |
BIC |
nan |
Log-likelihood at zero |
-21282.88 |
Log-likelihood at constants |
nan |
Log-likelihood at convergence |
-11891.36 |
Perception of bicycle as a means of transport |
Variable |
Coefficient |
p-value |
Level of education (from 1 to 5) |
1.03
|
0.0 |
# of bikes in household |
0.965
|
0.0 |
Standard deviation |
0.646
|
0.0 |
delta1 |
1.27
|
0.0 |
delta2 |
1.44
|
0.0 |
lambdaPerc_6 |
0.776
|
0.0 |
delta1Perc_6 |
0.753
|
0.0 |
delta2Perc_6 |
1.05
|
0.0 |
lambdaPerc_7 |
1.4
|
0.0 |
delta1Perc_7 |
0.594
|
0.0 |
delta2Perc_7 |
1.75
|
0.0 |
lambdaPerc_9 |
1.02
|
0.0 |
delta1Perc_9 |
1.05
|
0.0 |
delta2Perc_9 |
1.51
|
0.0 |
lambdaPerc_11 |
1.36
|
0.0 |
delta1Perc_11 |
0.611
|
0.0 |
delta2Perc_11 |
0.642
|
0.0 |
Perception of context characteristics |
Variable |
Coefficient |
p-value |
Level of education (from 1 to 5) |
0.48
|
0.0 |
Children in household (0 is base) |
-0.165
|
0.103 |
Available car |
0.573
|
0.0 |
# of bikes in household |
0.567
|
0.0 |
Income per month 0-1000 € |
1.01
|
0.0 |
Income per month 1000-2000 € |
0.591
|
0.0 |
Standard deviation |
-9.48
|
0.0 |
Actual biker, formerly car as driver (past behavior) |
0.417
|
0.18 |
Serial correlation (error component) |
1.91
|
0.0 |
lambdaContext_2 |
2.29
|
0.0 |
delta1Context_2 |
2.05
|
0.0 |
delta2Context_2 |
2.42
|
0.0 |
lambdaContext_3 |
0.858
|
0.0 |
delta1Context_3 |
0.996
|
0.0 |
delta2Context_3 |
1.4
|
0.0 |
Variable |
Coefficient |
p-value |
Constant |
-2.48
|
0.0 |
Age |
-0.014
|
0.026 |
Male (female is base) |
0.893
|
0.0 |
Children in household (0 is base) |
-0.713
|
0.0 |
# of bikes in household |
1.58
|
0.0 |
Non-subscriber (not informed is base) |
0.528
|
0.0 |
Subscriber (not informed is base) |
2.94
|
0.0 |
Perception of bicycle as a means of transport |
1.09
|
0.0 |
Perception of context characteristics in propensity to cycle |
1.34
|
0.0 |
Perception of bicycle as a means of transport*# of memebrs in household |
-0.064
|
0.0 |
Perception of context characteristics in propensity to cycle*# of cars in household |
-0.058
|
0.018 |
Dependent variable |
Decision to cycle |
Model type |
Hybrid choice model |
Sample size |
2202.0 |
R2 |
0.485 |
Adjusted R2 |
|
Pseudo R2
(nan)
|
nan |
AIC |
nan |
BIC |
nan |
Log-likelihood at zero |
-23496.83 |
Log-likelihood at constants |
nan |
Log-likelihood at convergence |
-12090.43 |
Perception of bicycle as a means of transport |
Variable |
Coefficient |
p-value |
Level of education (from 1 to 5) |
0.868
|
0.0 |
# of bikes in household |
0.778
|
0.0 |
Standard deviation |
0.684
|
0.0 |
delta1 |
1.08
|
0.0 |
delta2 |
1.27
|
0.0 |
lambdaPerc_6 |
0.836
|
0.0 |
delta1Perc_6 |
0.682
|
0.0 |
delta2Perc_6 |
0.962
|
0.0 |
lambdaPerc_7 |
1.39
|
0.0 |
delta1Perc_7 |
0.48
|
0.0 |
delta2Perc_7 |
1.49
|
0.0 |
lambdaPerc_9 |
1.14
|
0.0 |
delta1Perc_9 |
1.03
|
0.0 |
delta2Perc_9 |
1.44
|
0.0 |
lambdaPerc_11 |
1.45
|
0.0 |
delta1Perc_11 |
0.571
|
0.0 |
delta2Perc_11 |
0.599
|
0.0 |
Perception of context characteristics |
Variable |
Coefficient |
p-value |
Level of education (from 1 to 5) |
0.519
|
0.0 |
Children in household (0 is base) |
-0.213
|
0.023 |
Available car |
0.193
|
0.121 |
# of bikes in household |
0.595
|
0.0 |
Income per month 0-1000 € |
0.781
|
0.0 |
Income per month 1000-2000 € |
0.328
|
0.001 |
Standard deviation |
0.509
|
0.0 |
Actual biker, formerly car as driver (past behavior) |
0.566
|
0.035 |
lambdaContext_2 |
1.53
|
0.0 |
delta1Context_2 |
1.12
|
0.0 |
delta2Context_2 |
1.42
|
0.0 |
lambdaContext_3 |
0.926
|
0.0 |
delta1Context_3 |
0.982
|
0.0 |
delta2Context_3 |
1.35
|
0.0 |
Variable |
Coefficient |
p-value |
Constant |
-2.38
|
0.0 |
Age |
-0.014
|
0.02 |
Male (female is base) |
0.866
|
0.0 |
Children in household (0 is base) |
-0.675
|
0.0 |
# of bikes in household |
1.56
|
0.0 |
Non-subscriber (not informed is base) |
0.516
|
0.0 |
Subscriber (not informed is base) |
2.83
|
0.0 |
Perception of bicycle as a means of transport |
0.98
|
0.0 |
Perception of context characteristics in propensity to cycle |
1.53
|
0.0 |
Perception of bicycle as a means of transport*# of memebrs in household |
-0.074
|
0.0 |
Perception of context characteristics in propensity to cycle*# of cars in household |
-0.068
|
0.015 |
Dependent variable |
Decision to cycle |
Model type |
Hybrid choice model |
Sample size |
550.0 |
R2 |
0.488 |
Adjusted R2 |
|
Pseudo R2
(nan)
|
nan |
AIC |
nan |
BIC |
nan |
Log-likelihood at zero |
-5865.876 |
Log-likelihood at constants |
nan |
Log-likelihood at convergence |
-3001.59 |
Perception of bicycle as a means of transport |
Variable |
Coefficient |
p-value |
Level of education (from 1 to 5) |
1.02
|
0.0 |
# of bikes in household |
0.6
|
0.0 |
Standard deviation |
0.709
|
0.0 |
delta1 |
1.03
|
0.0 |
delta2 |
1.42
|
0.0 |
lambdaPerc_6 |
0.829
|
0.0 |
delta1Perc_6 |
0.624
|
0.0 |
delta2Perc_6 |
1.03
|
0.0 |
lambdaPerc_7 |
1.3
|
0.0 |
delta1Perc_7 |
0.5563
|
0.01 |
delta2Perc_7 |
1.35
|
0.0 |
lambdaPerc_9 |
0.956
|
0.0 |
delta1Perc_9 |
0.95
|
0.0 |
delta2Perc_9 |
1.15
|
0.0 |
lambdaPerc_11 |
1.3
|
0.0 |
delta1Perc_11 |
0.655
|
0.004 |
delta2Perc_11 |
1.24
|
0.0 |
Perception of context characteristics |
Variable |
Coefficient |
p-value |
Level of education (from 1 to 5) |
0.509
|
0.0 |
Children in household (0 is base) |
-0.075
|
0.689 |
Available car |
0.489
|
0.061 |
# of bikes in household |
0.49
|
0.0 |
Income per month 0-1000 € |
1.3
|
0.002 |
Income per month 1000-2000 € |
0.443
|
0.022 |
Standard deviation |
0.554
|
0.0 |
Actual biker, formerly car as driver (past behavior) |
1.55
|
0.024 |
lambdaContext_2 |
1.53
|
0.0 |
delta1Context_2 |
1.14
|
0.0 |
delta2Context_2 |
1.65
|
0.0 |
lambdaContext_3 |
0.898
|
0.0 |
delta1Context_3 |
0.892
|
0.0 |
delta2Context_3 |
1.5
|
0.0 |
Variable |
Coefficient |
p-value |
Constant |
-1.88
|
0.012 |
Age |
-0.024
|
0.066 |
Male (female is base) |
0.803
|
0.0 |
Children in household (0 is base) |
-0.982
|
0.0 |
# of bikes in household |
1.32
|
0.0 |
Non-subscriber (not informed is base) |
0.456
|
0.04 |
Subscriber (not informed is base) |
1.53
|
0.023 |
Perception of bicycle as a means of transport |
1.25
|
0.0 |
Perception of context characteristics in propensity to cycle |
0.946
|
0.0 |