交通系统规划与控制教学（清华大学）Transportation Systems Analysis.pdf

Transportation Systems Analysis
Problem Set 1
LIU Zhe, 2009010845
Mar 11, 2012
Part I
Case Study: Developing a Binary Choice
Model
1 herlands Mode Choice Case
Model 1: Straightforward specification
Best model specification and model estimation
Vcar = ar + β1 purpose + β2gender + β3carivtt + β4carcost
Vrail = β5railtime + β6railcost + β7railtrans f ers + β8railivtt
where railtime = time + railegrtime.
ar β1 β2 β3 β4 β5 β6 β7 β8
Car 1 purpose gender carivtt carcost 0 0 0 0
Rail 0 0 0 0 0 railtime railcost railtrans f ers railivtt
Value - - - - - - -
Table 1: Best model specification
Num of Para. ρ2 ρ¯2 L(βˆ) L(0) −2(L(0) − L(βˆ))
9 - -
Table 2: Model estimation
Preliminary statistical analysis
Here we use GLM procedure in SAS to perform variance analysis so that the factors which
have more significant impact on people’s choice could be found.
1
Figure 1: Quantiles for continous variables
To perform ANOVA we should first convert continous variables to discrete variables that
have several levels. And since all the original discrete variables are taking values 0 or 1, we
leave them unchanged. While for continous variables we denote the value below median as
level 0 and value above median as level 1. Because it is an unbalanced experiment design,
we could use GLM procedure to perform ANOVA.
Figure 2: Analysis of factors’ effect
From the Pr > F column of the above result, we know age, rail_ivtt, rail_transfers and
car_walk_time have relatively low effect on the choice of people’s selection of transportation
mode. Removing those four factors, we re-perform the GLM procedure and get
2
Figure 3: Analysis of factors’ effect after modification
From the above result, we see than under a 95% confidence level all factors are significant
in their effect. So we could start our model specification process with these factors.
Process of finding th