Travel price and time elasticities are increasingly being derived from disc
rete choice models of the multinomial or nested legit form. These elasticit
ies are then applied to obtain predictions of changes in travel demand cons
equent on a policy change in prices and travel times. The majority of the c
hoice elasticities are estimated within the behavioural setting of modal ch
oice, holding total travel fixed. A few mode choice models have recently re
laxed the multinomial legit model assumption of equal variance in all the r
andom components of the indirect utility function to permit unconstrained v
ariances' across all-alternatives (subject to identification for one altern
ative). This enables the derivation of behaviourally meaningful and unique
cross choice elasticities for each pair of alternatives. Under constant var
iance, only the direct choice elasticities have behavioural meaning. While
this is an important advance in discrete choice modelling, the derivation o
f share elasticities is conditional on a fixed total demand, and the proced
ure cannot be relied on to carry through two important properties of the-mo
del into the demand elasticity matrix-namely symmetry and zero share weight
ed column sums. This paper takes a set of empirically derived choice elasti
cities and presents a second stage procedure to adjust these elasticities t
o arrive at an internally consistent matrix of demand elasticities. We draw
on a recent data set collected in Sydney which utilises revealed preferenc
e and stated choice data to estimate a joint model of ticket choice conditi
onal on mode and choice of mode for commuter travel. (C) 1999 Elsevier Scie
nce Ltd. All rights reserved.