Estimating the impact of new rail station openings on through-passenger demand: a difference-in-differences approach

The opening of a new rail station on an existing line imposes an additional in-vehicle time costs for all passengers on existing flows whose journeys now include a stop at the new station. We henceforth refer to these passengers as ‘through-passengers’. This paper is concerned with the following question: what is the impact of this small time penalty on the travel demand for through-passengers? Consumer theory would tell us that, even with a small travel time increase, we should expect a proportionate demand reduction (Ben-Akiva and Lerman 1985). Consequently, the application of standard demand forecasting models in transport would predict that a new station would lead to a reduction on through-passenger trips. In some cases, depending on the demand elasticities on journey time, these can predict arguably large patronage reductions. This is a particular case of the more general issue of small travel time changes in appraisal, relevant to any transport mode.

At first sight, this may seem an innocuous issue. However, the evidence from appraisal practice tells us the contrary, providing the motivation for this paper. The estimated through-passenger demand reductions can undermine some proposals for a new station, as observed in a recent review of rail appraisals in the UK (Ojeda-Cabral et al. 2023). The review illustrated the severity of the issue with a real example. This was the case of a new train station being planned, which would cause a 2-minute penalty to through-passengers. Its appraisal resulted in a Benefit-Cost Ratio (BCR)¹ of 1.1. However, analysts realised that the BCR was extremely sensitive to one single aspect of the appraisal – namely, the modelled reductions in through-passenger demand – and ran sensitivity tests. If the modelled through-passenger demand reductions were halved, the BCR jumped to 18. If analysts disregarded the through-passenger demand reductions, the BCR became negative due to revenues exceeding the costs of the scheme.² This is a remarkable level of sensitivity. But while intuition from analysts and decision-makers suggested that the original model outcomes (i.e. BCR = 1.1) were not realistic – e.g. due to how alternative modes are in that corridor, in particular due to levels of road congestion – and might not be believed, there was no evidence to justify adjustments.³ Unfortunately, the issue of through-passenger responses to small journey time increases has received no attention in the literature. To the best of our knowledge, this would be the first academic article to expose the relevance of through-passenger demand for new station appraisals. While there is body of literature on the topic of small time savings, the focus has been on valuation using Stated Preferences (SP) (Dubernet 2019; Daly et al. 2014) and its implications for appraisal (Welch and Williams 1997; Bates and Whelan 2001), but not on demand forecasting. The broader issue of non-linear relationships between travel time and demand has also been studied (e.g. Balcombe et al. 2004; Heggie 1976), but existing work is limited and mostly dated. From a demand analysis perspective, the relevant question is whether journey time elasticities are lower for small time changes. Only Wheat and Wardman (2017) and a recent unpublished industry report (The Railway Consultancy et al. 2024) explore such question, with inconclusive results.

Against this background, this paper is the first to explore this question by means of a case study approach. A recent UK station opening provides a controlled quasi-experimental setup. We undertake an ex-post analysis of ticket sales data, using Difference-in-Differences (DiD) techniques, and set to answer the following questions:This paper aims to provide initial critical evidence for the planning and appraisal of new stations and to position the practical problem of "through-passengers" in the academic literature. In doing so, the paper also makes a novel contribution to the literature concerned with the wider issue of small journey time changes and heterogeneity in demand elasticities. The problem of small time changes has been around since at least the 1970s and poses significant, unresolved, questions for the ranking and selection of different project/policy types by different modes. However, there are gaps in this literature in addition to the aforementioned lack of focus on patronage analysis. Daly et al. (2014) argue for the use of Revealed Preference (RP) data given the perceived challenges with SP data on this topic. Jennings and Sharp (1976) highlight how the arguments for and against a special treatment of small changes may or may not apply depending on the context or scheme type. Yet, the academic literature has steered away from analysing particular interventions. This paper addresses these gaps.

The paper is structured as follows. Section 2 reviews the relevant literature strands. Sections 3 and 4 introduce respectively the data and the case study prior to describing the methodology in Section 5. Section 6 presents the results, with a discussion in Section 7. Section 8 concludes.

For any given new station opening, only specific flows are directly affected. Analysis of the impact of station openings on through-passenger demand therefore requires the use of flow-level demand data. A Difference-in-Differences approach requires such flow-level data for two kinds of flows:We use calendar weekly, flow-level, data on the number of passenger journeys, and on GJT and its components. These are taken from RIDEA (Rail Industry Dataset for Econometric Analysis), a new dataset that provides higher-frequency data on demand, GJT, and external factors than the annual RUDD (Rail Usage and Demand Drivers) dataset used by the majority of existing econometric studies of rail demand in Great Britain. Both datasets ultimately derive their demand data from the official national rail ticket sales database (known as LENNON⁹), and their GJT measures from MOIRA (a UK rail industry modelling tool for rail demand and revenues at OD flow level).

The GJT includes the following components: in-vehicle time (IVT), frequency/headway and, where applicable, an interchange penalty. Thus, GJT data is used to identify the impacts of station openings on treatment flow IVT and to detect cases where nothing else (other than IVT) changes. The GJT data can also be used to calculate the resulting demand impacts as implied by the industry standard demand forecasting methods (PDFH), providing a useful benchmark prior to our econometric estimations. Data on passenger journeys are also broken down by ticket type, and are used as our measure of demand in our Difference-in-Differences models.

The use of the higher-frequency data in RIDEA offers a number of advantages in this context. First, since station openings have taken place at various times of year, their impact on demand would be difficult to discern using only annual data. The use of weekly data allows us to map the periods before and after an opening to the weekly demand data with a high degree of accuracy.

Second, using high-frequency data allows us to more easily disentangle any immediate impacts of station openings from the confounding influence of annual fare changes, semi-annual timetable changes, and the effects of external factors (which over the course of a few weeks should remain relatively stable). This permits visual inspection of any obvious immediate demand impacts in the weeks following a station opening.

Third, since RIDEA includes timetabled GJT data relating to both the summer and winter timetables – in contrast to RUDD which uses only the summer timetable data – we should be able to identify the changes in IVT attributable to a station opening.

Careful case study selection is essential to try to study in isolation the effects of just a small time penalty. For a given new station, the initial aim is to identify a set of flows that were affected by its introduction in the form of a time penalty. These are the so-called ‘through-passenger’ flows. Flows involving close adjacent stations, where demand abstraction was likely, were avoided, since demand reductions would be expected primarily due to station substitution.

As our search for case studies confirmed, new station openings often coincide with other changes, e.g. changes to routes, service frequencies, and rolling stock. Sometimes, direct services are retained (effectively avoiding the time penalty for many through services).

Due to data availability, and the need to avoid the exceptional impacts of Covid-19, the search was restricted to the period between 2015 and 2018, when 24 new stations were opened in the UK. Of these, we initially identified a subset of 7 new station openings potentially suitable for our study. This subset filtered out new stations from new branch lines (no through-passengers affected) and the most ‘noisy’ or complex cases involving multiple simultaneous changes. However, even this pre-selected subset did not fully avoid the presence of simultaneous changes, retention of direct services and stations which are part of complex network/service patterns.

After further exploration, it became clearer that only 3 out of these 7 stations provided an adequate context to study the impact of a small time penalty in isolation: Maghull North, Coventry Arena and Bermuda Park. Since the last two are part of the same line and opened at the same time, this reduced the number of case studies to two. In-depth analysis of these two cases, with the assistance of rail industry partners, revealed that 1 of the cases (Maghull North) provided the exceptionally clean context that we were looking for. That is, the changes in IVT were clear-cut and perceivable, no other changes took place, and there was a sufficient number of treatment and control flows that shared a similar trend prior to the opening (including some controls on the same line). Thus, this paper focuses on the analysis of the Maghull North case study.

We are concerned with estimating the impact of station openings on ‘through-passenger’ demand on affected flows. Clearly, however, it is inadequate to simply compare demand on these flows before and after the station openings in question. This is because there are many factors influencing rail demand also at work – variation over time in fares, services, seasonal effects, and external factors such as incomes, populations, employment, and so on.

We estimate the effect of the station opening using a Difference-in-Differences (DiD) approach. This method compares the evolution of demand for flows affected by the station (the treatment group) to that for unaffected flows (the control group), before and after the intervention. Provided key assumptions hold, this strategy can isolate the causal effect of the station opening from general time trends.

The DiD estimator is widely used in policy evaluation and applied microeconometrics (e.g., Card and Krueger 1994; Bertrand et al. 2004), and has become increasingly common in transport settings. Applications include congestion charging and traffic safety (Li et al. 2012), infrastructure and house prices (Dubé et al. 2014; Murray and Bardaka 2022), walkability and active ageing (Marquet et al. 2017), and tourism impacts of high-speed rail (Liu 2024).

Identification in a DiD framework rests primarily on the parallel trends assumption, i.e. the assumption that, in the absence of the treatment, outcomes in the treatment and control groups would have continued to follow a common trend. This cannot be tested directly, but can be supported by comparison of pre-treatment trends. Violation of this condition can lead to biased estimates (see, e.g., Lechner 2011).

Tables 4 and 5 report regression outputs for our simpler specifications – including only one post-treatment dummy – and the more detailed model with two separate post-treatment dummies, respectively. In both cases, both OLS and TWFE estimates are shown side-by-side, with the lack of a constant term or coefficients on AFTER or the post-treatment dummies in the TWFE cases owing to their being dropped, as discussed in Section 5.2. The models are estimated first using all control flows, and subsequently with only control flows on the same branch line of the treatment flows (group 1 as in Table 2).

Coefficient estimates are reported, with standard errors in parentheses. As discussed, the coefficients of interest are \(\beta _3\) in Table 4 and \(\beta _4\) and \(\beta _5\) in Table 5. These coefficients correspond to the interaction terms between the dummies indicating the treatment group and the post-treatment period(s), and therefore estimate any differential change in the (the logarithm of) demand on treatment flows relative to control flows after the station opening, isolating the treatment effect by controlling for both common time trends and fixed group differences.

As discussed in Section 5, the raw regression results are not themselves the end product, since we must transform the estimated coefficients in order to express them as percentage impacts on demand. Transformed coefficients, interpretable as percentage impacts, are reported in Tables 6 and 7, which show the estimates from the OLS and TWFE models, respectively. Also shown are their corresponding standard errors, and symbols indicating the results of two hypothesis tests.

The first test is of the null hypothesis \(H_0: \hat{\delta }_j=0\), i.e. that the station opening had no impact on demand. The second is a one-sided test, where the null hypothesis is \(H_0: \hat{\delta }_j\le -0.1\), i.e. that the station opening caused a reduction in demand at least as great as 10%, as implied by the relevant PDFH elasticity. Significance levels for the former test are denoted using stars, while for the latter they are denoted using daggers.

Note that failure to reject no demand impact does not imply that we can reject the null hypothesis of a demand impact in line with the benchmark (PDFH) elasticity – 10% in this case. If the standard errors on these estimates are large enough, it is possible that we fail to reject both null hypotheses – the estimates would be too imprecise to be informative.

In this case, for the simpler model specifications (both OLS and TWFE) – first rows of Tables 6 and 7 – we can see that there are many daggers and no stars next to our estimates. This indicates that we strongly reject the null hypothesis of demand reductions in line with – or greater than – those implied by the relevant PDFH elasticity, while we fail to reject the null hypothesis of no demand impact. In other words, the demand impacts were less than the 10 per cent PDFH would have predicted, and with the possibility that they were actually zero (i.e. no demand impact). When looking at our more detailed models with separate estimates for the first and second years following the station opening, the same results hold for the second-year impacts, but we see that the estimates for the first-year impact are more imprecise: we still cannot reject the null hypothesis of no demand impact, but we can only reject the null hypothesis of demand impacts in line with the PDFH at the 10% confidence level using control group 1 – i.e. control flows on the same branch line as the treatment flows only.¹³ In the second year, although not significantly different from zero, the estimated demand impact is positive – the opposite of what we would expect.

Summarising, these results suggest that the opening of Maghull North did not reduce through-passengers’ demand as much as industry models (PDFH) would have predicted, and possibly did not reduce demand at all.

Let us start by recalling that a GJT increase of 4 min multiplied by PDFH GJT elasticities suggested a 10% reduction in demand for our treatment flows passing through Maghull North. Against this background, our estimates fail to support such a large reduction in peak demand, at least within the first two years following the opening of Maghull North. Indeed, we were also unable to reject the null hypothesis of no demand impact at all.

These findings are aligned with the widespread evidence that people attach low or even negligible value to small time changes (Daly et al. 2014), which can be due to multiple reasons such as perception or usability of small amounts of time (Welch and Williams 1997). Other reasons for these findings may include prevalent passengers’ constraints and habits, the uncompetitiveness of other travel modes or even the possibility of positive utility of adding a few minutes to a “too-short” rail journey (Redmond and Mohktarian 2001). All these, especially if combined, could explain the observed behavioural response from existing through-passengers in our case study. For instance, between Ormskirk and Liverpool, car journey times are at best 35 min, but roads are highly congested and journeys can actually take 1 h or longer. It would seem plausible that pre-existing rail users were not convinced to switch after their rail journey went up by 4 min – even if they did perceive it.

Before we discuss implications and usability of results, we must acknowledge the limitations. First, we cannot rule out that what we are observing is only a short-term effect. The analysis is restricted to a short-term demand response (2 years after opening). It is possible that the long-run impact takes longer to realise or differs, since short-term mode choices involve different trade-offs than long-term residential choices (e.g. Beck et al. 2017). However, PDFH guidance suggests that for a GJT deterioration approximately 90 per cent of the impact should be realised after a 2-year period. Second, the data on all ticket types was not available, e.g. season tickets. A full dataset would enable a more complete analysis, although the data on full fare tickets does provide significant initial insights with respect to a large market segment. Third, there are other demand drivers which could not be included in the data and might be affected by a new station, such as service reliability (e.g. via changes to dwell times and additional boarding/alighting). For the case where control flows are within the same line, however, there is no reason to suspect that any potential reliability impacts would differ between treatment and controls, which offers some reassurance. Last but not least, the case study approach limits the transferability of results.

Despite these limitations, the quasi-experimental setup is robust and provides us with reliable findings. These findings provide very valuable initial information for the railway industry and community on an important, yet unexplored, issue. Second, our results are also a first step towards addressing key gaps in the literature on small time savings, largely based on hypothetical rather than observed behaviour (Daly et al. 2014).

Our evidence has implications for the appraisal of new rail stations. Recall the motivation for this study: modelled demand reductions of through-passengers can have a large and critical impact on the BCRs for new stations. This evidence provides grounds to begin to challenge the expectation of large demand reductions, while flagging the need for more research. The next step should be to gather more evidence to corroborate and generalise the findings to enable transferability of results, covering other case studies and longer time-frames.

While acknowledging the early nature of these findings, it is worth discussing how appraisal may deal with them. We saw how appraisal practice has largely avoided the use of lower VTT for small time changes, despite the overwhelming evidence (e.g. Dubernet 2019), for reasons such as the ‘adding up’ argument (Jennings and Sharp 1976). But the debate has focused on ‘valuation’, not on actual demand responses and ‘forecasting’. Demand responses and the associated impacts (e.g. in revenues) are observable facts. If it is observed that demand, and thus revenues, did not drop (or not as much as conventional elasticities would imply), it would simply be incorrect to assume anything different in the appraisal – it is an empirical matter. Unlike the question of which VTT to assign to small changes, which also involves judgement. Suppose that a body of evidence emerged that support lower demand elasticities for small time losses for through-passengers, it would presumably be difficult to argue against their use in appraisal practice.

Some thought may also be given to whether the ‘adding-up’ argument holds in the case of time losses and new stations. The argument is rooted on the typical case of subsequent road improvements. But, for a given railway line, how likely are continuous subsequent deteriorations in GJT to occur? In the last 25 years in the UK, approximately 100 new stations have opened.¹⁴ It may be possible that a given, say 3-minute, increase is a one-off scenario with no expectation of subsequent increases. And, often, projects and policies move in the opposite direction (i.e. providing GJT reductions). Furthermore, if we consider the case of short-distance routes, the scope for subsequent additions of stations is going to be even more limited. As Jennings and Sharp (1976) pointed out, the validity of the adding-up argument will be context specific.

On the other hand, special treatment of small changes in one forecasting context (new station openings) may be controversial without generalisation to other contexts of small changes (e.g. other rail schemes and other modes), and would require further consideration.

For the time being, while more general evidence is needed, this study certainly warrants more scrutiny around through-passenger impacts. Our study supports the use of sensitivity tests in cases where potential through-passengers demand reductions are likely to act as a barrier to a new station being built. For instance, sensitivity tests halving the modelled reductions could be recommended and given sufficient weight in the decision-making process.

This paper has explored the presumed demand reductions caused by new rail stations on through-passenger flows where a few minutes are added to the timetable. An ex-post analysis of a station opening in the UK was conducted to observe what had happened with through-passenger demand, using DiD econometric techniques. The case study provided a unique opportunity whereby the only effect of the new station on affected flows was a small increase in in-vehicle time (between 1 and 4 min). Nothing else, including frequency, quality or route patterns, changed. No other recent UK stations provided such controlled environment.

The question of rail demand responses to small changes in GJT has received very little attention and the few previous studies were inconclusive. Previous attempts have been only a sub-task of larger demand studies (The Railway Consultancy et al. 2024; Wheat and Wardman 2017) aiming to estimate rail demand elasticities on national datasets – naturally a difficult setting to identify the effect of small time changes due to the many confounding factors. To the best of our knowledge, our study is the first to apply a case study approach and DiD techniques to this question, offering a more controlled environment than previous studies. This work also contributes to the literature on the broader issue of small time savings, which has nevertheless focused primarily on whether these are valued at a lower unit rate using stated preference techniques, as opposed to exploring observed demand impacts (Daly et al. 2014).

The key finding of our study is that we could not reject the null hypothesis that no demand impact occurred – i.e. demand response was statistically no different to other similar flows in the same line and same area that were unaffected. This held across the various model specifications tested. We also compared the model estimates with the expected predictions of standard rail forecasting methods in the UK (PDFH), rejecting also the hypothesis that demand impacts were as large as what standard models predict. In other words, our results suggest that the opening of Maghull North did not reduce through-passengers’ demand as much as industry standard models would have predicted, and possibly did not reduce demand at all.

These observations will be of general interest to the railway and transport communities, constitute a novel contribution to the study of "small time changes" and have important implications for the planning and appraisal of new stations – although these need to be threaded carefully as discussed in the paper. In the UK, the forecasts of demand reductions for ‘through-passengers’ can play a game-changing role in a station Benefit-Cost Ratio (Ojeda-Cabral et al. 2023) and overestimating these could be a barrier to building the station. Given the similarities of appraisal methods in many developed countries (Mackie and Worsley 2013), we hypothesise that this issue extends beyond the UK borders. In Europe, for instance, local communities and authorities are seeking to achieve mode transfer to rail. If our conclusions were to hold, there could be further commercial opportunities for the opening of local rail stations. At the very least, our evidence warrants additional scrutiny around any modelled through-passenger demand reductions from new stations. Perhaps more importantly, it warrants further research to corroborate if the findings hold in other case studies, ideally within a longer-term frame and covering other countries, and in the more general context of small time changes – both in rail and other modes.