Effect of latent variables on travel behavior
7.1 Mode choice with latent variables
Data used in this work are taken from the Agartala Household survey which was discussed in detail in the Chapter on data collection. Responses to the qualitative questions provided important insights about the individuals’ opinions on travel mode. The respondent had to rate his/her level of agreement on a five point Likert scale ranging from a total disagreement (response of 1) to a total agreement (response of 5). Indicator variables were grouped suitably based on prior assumptions to construct four latent variables, comfort,
safety, flexibility, and reliability. Three explanatory variables education, age, gender were used to model the latent variables. Figure 7.1 shows the grouping of indicator variables to construct the latent variables.
In this analysis, two different models were estimated (a) Multinomial logit model (MNL) as base model; and (b) an integrated choice and latent variable model. In the model outlined in Figure 7.2, it is hypothesized that there is an underlying latent attitude among individuals towards different travel modes which vary across the respondents. The structure of the hybrid choice model is provided in the following section. Description of variables used in the models is presented in Table 7.2.
Table 7.1: Observed mean perception rating for different modes
Perception attribute Car Bus MThW MTW Cycle Cycle
rickshaw
Comfortable in journey 4.11 2.99 3.23 3.73 3.00 3.57
Always availability of comfortable seats 3.78 3.13 3.39 3.69 3.09 3.63
Very easy accessibility 3.69 3.08 3.50 3.90 3.08 3.31
Ability to reach destination in time 3.91 3.22 3.53 3.90 3.18 3.21 Can exactly calculate travel time prior to trip 3.90 3.16 3.37 3.91 3.10 3.21
Safety from accident 3.83 3.55 3.43 3.24 2.95 3.15
Safety from theft 3.92 3.18 3.35 3.69 3.17 3.17
Safety from weather 3.89 3.55 3.49 2.20 2.20 3.01
Ability to make more trips 3.85 3.03 3.51 3.68 2.72 3.02
Can travel without changing vehicles 3.94 3.09 3.40 3.93 3.93 3.16
Figure 7.1: Construction of latent variable as a function of indicator variables
Table 7.2: Description of variables used in modeling latent variable choice models
SN Variables Description
1 Age Age in years
2 Private Vehicle Ownership Number of private vehicles in a household
3 Education Years of education.
4 Gender Dummy variable
1 for male respondent 0 for female respondent
5 License 1 for respondent having license
0 for respondent not having license 6 Family Size Number of members in the family
7 Income Gross monthly
household income
Code
0-2000
1 2001-5000 2
5001-10000 3
10001-15000 4
15001-20000 5
20001-30000 6 30001-40000 7 40001-50000 8 Comfort
Reliability
Flexibility
Safety
Comfortable in journey Comfortable seats are always available
Very easy accessibility Ability to reach destination in time Can exactly calculate travel time prior to trip
Safety from accident Safety from theft Safety from weather Ability to make more trips Can travel without changing vehicles
7.1.1 Specification for structural equation model
Various specifications of structural equation model have been tried and the final model used in the present analysis is shown below. Reliability and safety could not be modeled using any of the socioeconomic characteristics. In the final model comfort related to car, flexibility associated to MTW and MThW are found to be significant in explaining the mode choice behavior. Socioeconomic variables that are significantly entering the SEM were also found to be significant in modeling the systematic component of the preference heterogeneity (from the previously estimated models). In the final SEM model, age, gender, and education variables are used in modeling the comfort of car and flexibility of the MTW and MThW.
50001-70000 9 70001-90000 10
90001-150000 11
>150001 12 8 Bicycle Ownership Dummy Variable
1 for households having bicycle 0 for households having no bicycle
9 MTW Ownership Dummy Variable
1 for households having Motorized Two wheeler 0 for households having no Motorized Two wheeler.
10 Time In vehicle travel time.
11 Cost Fare or fuel cost
12 Area Index Land use parameters obtained from GIS analysis.
13 LVcomfortcar Comfort of car measured by indicator variable 14 LVflexMTW Flexibility of MTW measured by indicator variable 15 LVflex MThW Flexibility of MThW measured by indicator variable 16 LV Education Education in number of years used in structural equation 17 LV Age Age in number of years used in structural equation 18 LV Gender Gender as dummy variable used in structural equation
19
ASC Car
ASC Bus
ASC MThW
ASC MTW
ASC Bicycle
ASC Rickshaw
ASC Walk
Alternate specific constants for different modes.
LVcomfortcar = λ1car * education + λ2car * age + λ3car * gender + ζcar (7.1)
LVflexMTW = λ1 MTW * education+λ2 MTW * age+λ3 MTW * gender + ζ MTW (7.2)
LVflex MThW = λ1 MThW * education + λ2 MThW * age + λ3 MThW * gender + ζ MThW (7.3) where, λ is the associated parameter to be estimated and ζ is the error term normally distributed with zero mean and standard deviation σ.
7.1.2 Specification for the measurement equations
Measurement equations explain the relationship between the latent variables and the indicator variables. Following are the possible significant relationships between various indicator variables and the latent variables associated to Car, MTW, and MThW. All the relationships are found to be logically convincing.
Comfortable journey car(Indicator) = ϒ2Car * LVcomfort car + δ1car (7.4) Comfortable seat car(Indicator) = ϒ2Car * LVcomfort car + δ2car (7.5) Accessibility MTW (Indicator) = ϒ1 MTW * LVflex MTW + δ1 MTW (7.6) Ability to make more trips MTW (Indicator) = ϒ2 MTW * LVflex MTW + δ2 MTW (7.7) Travel without changing vehicle MTW (Indicator) = ϒ3 MTW * LVflex MTW + δ3MTW (7.8) Accessibility MThW (Indicator) = ϒ1 MThW * LVflex MThW + δ1 MThW (7.9) Ability to make more trips MThW (Indicator) = ϒ2 MThW * LVflex MThW + δ2 MThW (7.10)
Travel with changing vehicle MThW (Indicator) = ϒ3 MThW * LVflex MThW + δ3 MThW (7.11)
where, ϒ is the associated parameter to be estimated and δ is the error term normally distributed with zero mean and standard deviation σ.
A
Figure 7.2: Schematic diagram of simultaneous latent variable and mode choice model
Comfort (Car)
Income Education Age Gender License Motorcycle
ownership
Time Cycle ownership
Area Index NMT Family Size
Cost
Flexibility (MTW)
Flexibility (MThW)
UCar
UBus UMThW UMTH
UCycle UCycle rickshaw
UWalk
Choice
Very easy accessibility Ability to make more trips
Can travel without changing vehicles Very easy accessibility Ability to make more trips
Can travel without changing vehicles Comfortable journey Always availability of
comfortable seat
δ1 δ2
δ3
δ4
δ5
δ8
δ6
δ7