Premium
NETWORK LOADING VERSUS EQUILIBRIUM ESTIMATION OF THE STOCHASTIC ROUTE CHOICE MODEL MAXIMUM LIKELIHOOD AND LEAST SQUARES REVISITED *
Author(s) -
Anas Alex,
Kim Ikki
Publication year - 1990
Publication title -
journal of regional science
Language(s) - English
Resource type - Journals
SCImago Journal Rank - 1.171
H-Index - 79
eISSN - 1467-9787
pISSN - 0022-4146
DOI - 10.1111/j.1467-9787.1990.tb00082.x
Subject(s) - inefficiency , flow network , estimation , flow (mathematics) , least squares function approximation , mathematical optimization , econometrics , computer science , mathematics , statistics , economics , estimator , geometry , management , microeconomics
ABSTRACT Daganzo (1977, 1979), Daganzo and Sheffi (1977), Sheffi (1985), and Sheffi and Daganzo (1980) have used one assumption about traveler behavior in developing estimation techniques for the stochastic route‐choice problem and another assumption in predicting flows on networks by using the same model. In estimation, they calculate the congested travel costs of the network links from observed flows on the network, and the network is loaded based on these costs. In prediction, they follow their stochastic user‐equilibrium assumption by which travelers evaluate costs using the mean of the observed flows (or equilibrium flows). The travel‐cost coefficient obtained from the loading method systematically overestimates the true travel‐cost coefficient from which the observed flow data (which must be used in loading) is generated. The estimates of the same coefficient, obtained in this paper, by constraining the estimation results to conform to the equilibrium conditions are unbiased, and only marginally less efficient (have larger standard deviations). The average percentage error and inefficiency of the link flow predictions based on the loading method increases as the level of congestion on the network rises. In contrast, the average percentage error of link flow predictions based on the equilibrium estimation method declines and their efficiency remains very high as the level of congestion on the network rises.