4.3. Logistic regression models for non-captive stations
It can be seen from Table 7 that the correlation between distance
and travel time, travel mode and time, purpose and InboundOut and
InBoundOut and travelFeeD are 0.5, 0.45, 0.5 and
0.35 respectively. Therefore, travel time and InboundOut were
removed for model selection.
Three variables were identified to be significant from the best
fitting logistic regression model for non-captive stations (see
Table 8). There are 559 records for the non-captive stations
(Table 2), but the sample size for this regression model is 486 with
73 missing records being removed for the purpose of the analysis.
The most influential variable is travel cost (from a chosen station to
a destination). The less the cost of travelling from a chosen station
to a destination, the more likely a chosen station will be a nonnearest
station, which is consistent with the results from the
overall model. However, different from the model for all chosen
station, cost (origin to station) was found to be significant. The less
the cost from origin to the chosen station, the less likely it is that
chosen stations will be the nearest station.
5. Sensitivity test for policy implication
A sensitivity test was conducted using the overall model shown
in Table 4 to identify the influence of travel distance (from an origin
to a chosen station) and travel cost (from a chosen station to a
destination) on the nearest station choice by holding other
independent variables at their mean. The resulting sensitivity
plot for all stations (Fig. 4a) indicates that the predicted
probabilities of choosing nearest station decrease as travel
distance increases for all
five different travel fees, which are
travelling over one zone ($2.70), two zones ($4.00), three zones
($4.90), four zones ($5.80) and
five zones ($7.10). Generally, the
closer the chosen station to the destination, the lower probability
of a chosen station is the nearest train station to the origin, except
4.3. Logistic regression models for non-captive stationsIt can be seen from Table 7 that the correlation between distanceand travel time, travel mode and time, purpose and InboundOut andInBoundOut and travelFeeD are 0.5, 0.45, 0.5 and0.35 respectively. Therefore, travel time and InboundOut wereremoved for model selection.Three variables were identified to be significant from the bestfitting logistic regression model for non-captive stations (seeTable 8). There are 559 records for the non-captive stations(Table 2), but the sample size for this regression model is 486 with73 missing records being removed for the purpose of the analysis.The most influential variable is travel cost (from a chosen station toa destination). The less the cost of travelling from a chosen stationto a destination, the more likely a chosen station will be a nonneareststation, which is consistent with the results from theoverall model. However, different from the model for all chosenstation, cost (origin to station) was found to be significant. The lessthe cost from origin to the chosen station, the less likely it is thatchosen stations will be the nearest station.5. Sensitivity test for policy implicationA sensitivity test was conducted using the overall model shownin Table 4 to identify the influence of travel distance (from an originto a chosen station) and travel cost (from a chosen station to adestination) on the nearest station choice by holding otherindependent variables at their mean. The resulting sensitivityplot for all stations (Fig. 4a) indicates that the predictedprobabilities of choosing nearest station decrease as traveldistance increases for allfive different travel fees, which aretravelling over one zone ($2.70), two zones ($4.00), three zones($4.90), four zones ($5.80) andfive zones ($7.10). Generally, thecloser the chosen station to the destination, the lower probabilityof a chosen station is the nearest train station to the origin, except
การแปล กรุณารอสักครู่..
