TY - JOUR
T1 - How effective are fixed-effects models in fixing the transit supply–demand bidirectional interaction?
AU - Diaz-Gutierrez, Jorge
AU - Ranjbari, Andisheh
N1 - Publisher Copyright:
© 2024 The Author(s). Published by Informa UK Limited, trading as Taylor & Francis Group.
PY - 2024
Y1 - 2024
N2 - Transit agencies use direct demand models (DDM) to allocate services. Since the service supply–a crucial predictor in DDMs–is endogenous to demand, including it in the model might yield biased estimations. A widely used methodology that is believed to handle this issue is Fixed Effects (FE). However, the underlying assumptions of FE are valid only if service adjustments take a considerable amount of time. This study investigates the performance of FE for estimating transit ridership. We collected 2013–2019 data and constructed 16 DDMs, employing four methodologies with a shared set of variables. We found that FE has significant limitations in handling endogeneity and will result in parameter estimates that significantly differ from those produced by methodologies that are specifically designed to control for endogeneity (such as FE-IV). Moreover, the use of FE leads to the omission of certain predictors and inaccurate ridership predictions, misguiding agencies as to what changes to implement and potentially impacting revenue projections.
AB - Transit agencies use direct demand models (DDM) to allocate services. Since the service supply–a crucial predictor in DDMs–is endogenous to demand, including it in the model might yield biased estimations. A widely used methodology that is believed to handle this issue is Fixed Effects (FE). However, the underlying assumptions of FE are valid only if service adjustments take a considerable amount of time. This study investigates the performance of FE for estimating transit ridership. We collected 2013–2019 data and constructed 16 DDMs, employing four methodologies with a shared set of variables. We found that FE has significant limitations in handling endogeneity and will result in parameter estimates that significantly differ from those produced by methodologies that are specifically designed to control for endogeneity (such as FE-IV). Moreover, the use of FE leads to the omission of certain predictors and inaccurate ridership predictions, misguiding agencies as to what changes to implement and potentially impacting revenue projections.
UR - http://www.scopus.com/inward/record.url?scp=85210014414&partnerID=8YFLogxK
UR - http://www.scopus.com/inward/citedby.url?scp=85210014414&partnerID=8YFLogxK
U2 - 10.1080/19427867.2024.2422713
DO - 10.1080/19427867.2024.2422713
M3 - Article
AN - SCOPUS:85210014414
SN - 1942-7867
JO - Transportation Letters
JF - Transportation Letters
ER -