An Econometric Analysis of the Impact of Reliability on Passenger Rail Demand
R Batley, J Dargay, J N Ibanez, M Wardman, J Shires, ITS, University of Leeds, UK
This paper reports a study commissioned by the UK Department for Transport to investigate the impact of reliability on passenger rail demand, synthesising econometric analysis at the market level with discrete choice analysis at the individual level.
This paper reports a study commissioned by the UK Department for Transport to investigate the impact of reliability on passenger rail demand. The study develops two parallel research themes, motivated by an interest in revealing distinct but complementary insights, as follows: ? aggregate: dynamic econometric methods are employed to yield evidence on the market elasticity properties of changes in reliability, and the characteristics of any inherent dynamics ? disaggregate: discrete choice methods are employed to yield evidence on behavioural responses to reliability at the individual level, and associated monetary valuations
Further to our presentation of the discrete choice analysis at ETC2007, the project has progressed to completion, and we are now equipped to report the econometric analysis, as well as the synthesis of both analyses.
Whilst econometric analysis of rail ticket sales data has made a major contribution to the understanding of UK rail demand, applications to rail reliability have been limited. Whereas the two most significant previous studies, by SDG (2003) and OXERA (2005), considered TOC-level reliability metrics (the so-called Public Performance Measure (PPM) in the former, and delay/train miles in the latter), we sought a deeper level of disaggregation. This we achieved through exploiting reliability data held at service group level within Network Rails PEARS database.
Informed by the SDG and OXERA works, our assumption at the outset was that the impact of reliability on demand would be relatively small. We felt that it would be potentially most revealing to contrast flows with and without significant changes in reliability. To this end, we assembled a list of 248 specific O-D pairs, and set about acquiring the relevant reliability data. The latter involved data at service group level (specifically the metrics of Average Lateness Minutes (ALM) and Average Performance Minutes (APM), where the latter is defined as deemed plus actual lateness), as well as TOC-level PPM data. We were able to source reliability data for the period 2002 until 2007, which were then combined with demand and revenue from LENNON ticket sales data, generalised journey times from MOIRA, as well as further socio-economic-demographic data from ONS.
Following previous work on rail demand, a constant elasticity demand model was specified on our data, and estimated by Generalised Least Squares. We applied this in both fixed- and random-effects formulations, and on the basis of statistical tests found the former to be preferred. We then proceeded to develop static and dynamic specifications. In our static model the estimated reliability coefficients are significant and of the correct sign. Reliability, as characterised by the metrics used, is found to have only a marginal effect on rail demand. Although season tickets are slightly more sensitive to changes in reliability, the differences in the elasticities for ALM and APM (ranging from 0.03 to -0.06) across the different ticket types are not statistically significant. The elasticity for PPM, however, is significantly greater for reduced (0.25) and season tickets (0.27) than it is for fullfare tickets (0.09). Proceeding to dynamic model specification, we find that the ALM and APM elasticities are of the same order of magnitude as in the static model, even in the long run. In contrast to the static model, the PPM elasticity is only significantly different from zero for full-fare patrons, estimated at 0.05 in the short run and 0.19 in the long run.
Previous research on travel time reliability, both in the UK and internationally, has usually adopted either market level or individual level analysis of the problem. A novel contribution of our paper is its synthesis of both levels of analysis. We achieved this by extending our representation of Generalized Journey Time (GJT) within the econometric models to include ALM. To convert ALM to equivalent journey time, we used the factor of 1.6 emanating from our discrete choice analysis. The resulting elasticities for static and dynamic models are very similar to those obtained from the comparable model including ALM. If we work on the basis of the average proportion of lateness in GJT, the ALM elasticities derived from the GJT elasticities indicate that reliability has a minimal effect on demand, thus confirming our earlier results. However, in practice, the share of lateness in GJT varies by O-D, from near zero to about 40%. Accounting for this, the ALM elasticity will range from zero (at perfect reliability) to about 0.3 (for fullfare tickets).
Association for European Transport