2 Literature reviews

2.1 Forced experimentation and COVID-19 pandemic

Short disruption to public transport system can cause lasting impact in people’s behaviour. A growing body of literature has explored the impact of transport disruption on people’s travel behaviour. In Larcom et al. (2017), such disruption is referred to as “forced experimentation”, which defines the situation where agents are forced to modify their behaviours due to circumstances beyond theirs, and researchers’ control.

The concept of forced experimentation shares certain similarities with natural experiments. While the term forced experimentation is not commonly used, natural experiment has been widely discussed in the field of social science, despite the ongoing controversy of its precise definition. In this study, adapted from DiNardo (2008) and Dunning (2012), a natural experiment is defined as an empirical situation in which external circumstances or events that are outside the control of the researcher or participants, produce variation in exposure or treatment that is plausibly exogenous. This allows for comparison between affected and unaffected groups, approximating the conditions of a controlled experiment without deliberate intervention.

Larcom et al (2017) treated a strike in London Underground in 2014 as a forced experimentation/natural experiment, By using the entry/exit data of public transport, they analysed commuters’ behaviour change before and after the strike with quantitative methods. The results suggested that even a short, two-day strike can have long term influence on travel choices.

COVID-19, the pandemic happened in 2020, is widely recognised as a natural event reshaped people’s daily life globally. Compared with past incidents such as SARS outbreak in 2003 and various terrorist attacks, COVID-19 had an unprecedented impact on public transport. In the early stage of the lockdowns, passenger demand decreased by between 80% and 95% around the world (Vickerman, 2021). In London, the underground system was forced to close during the pandemic. At the peak of the lockdown in 2020, Underground ridership plunged to just 5% of its pre-pandemic levels, while bus use dropped to about 20% (Transport for London, 2020). It is also believed that the pandemic might have long term effect worldwide (Vickerman, 2021; Swinney, 2023). Although lockdown in London has long been lifted, its disruption to public transport may still linger. Therefore, this study treats COVID-19 as a natural experiment and explores the recovery pattern of underground usage in London over the years since the outbreak, setting the context for the following review and discussion of previous studies.

2.2 Post-pandemic mobility pattern in London

2.2.1 Commuting and work-related travel

The pandemic reshaped commuting flows across London, with studies showing sharp differences between central and peripheral areas. Murcio et al. (2023) analysed Origin–Destination patterns of public transport commuting trips collected during June, 2022, finding central boroughs remained suppressed long after lockdowns were lifted, while outer areas exhibited faster recovery. Their OD-based approach provided a detailed picture of spatial disparities in work-related travel, demonstrating that central London’s reliance on office-based employment left it particularly vulnerable to remote-working policies. These findings suggest that commuting recovery cannot be viewed as a uniform process, but as one shaped by the geography of employment and land use.

Batty et al. (2020) examined commuting flows in London during the pandemic using mobile location data collected in 2020, distinguishing between essential and non-essential travel. Spatially, their analysis revealed central commuting flows collapsing while some peripheral areas linked to healthcare and logistics retained activity. The finding was similar to the result of Murcio et al. (2023), as discussed above. At the same time, Batty et al. discovered that essential and non-essential workers showed more similar mobility patterns than expected. Together, these findings show that while central London was especially exposed to the pandemic’s disruption, differences between worker groups alone cannot fully explain the uneven recovery.

2.2.2 Broader mobility adaptation

Beyond commuting, other research has examined wider shifts in mobility and activity patterns. Trasberg and Cheshire (2021) used mobile phone data collected from January to July 2020, to map social activities across London during lockdown. Their results showed that central London lost much of its activity, while some outer areas remained more stable. Although the central area remained the dominant activity hub in absolute terms, it experienced the most sharp decline compared with the suburbs, where activity levels proved more resilient. This difference between overall concentration and relative change is important, as it highlights the uneven vulnerability of central London during the pandemic.

Gao et al. (2023) studied London’s bicycle hire scheme data from 2019 to 2021, in order to examine post-pandemic mobility recovery. They found that bicycle usage spiked at the same time as Underground ridership fell, showing a substitution effect shaped by risk perception and flexibility. Trips also shifted towards areas near parks and hospitals. Although their study did not focus on the Underground directly, it showed how different transport modes responded to the pandemic, adding another perspective to understanding public transport recovery in London.

Bansal et al. (2021, 2022) surveyed London commuters using different modes of public transport in March - May 2021 and found that the share of Underground travel fell from almost 70% before COVID-19 to around 40% by 2021. At the same time, the proportion of workers not commuting rose sharply to nearly half of the sample, reflecting the impact of remote work. They also reported moderate increases in car use and active modes such as cycling and walking. Although their main focus was on behavioural factors, these results provide useful evidence of how Underground demand fell and commuting purposes shifted during the pandemic.

Beyond spatial and travel mode shifts, researchers have also highlighted the temporal dimension of post-pandemic mobility. Reports show that traditional weekday patterns have been reshaped, with Fridays no longer the busiest commuting day and Thursdays emerging as the new peak (Transport for London, 2023). At a broader level, Batty (2023) raises the question of whether such changes are merely transitionary, part of a slow recovery trajectory, or represent a more permanent new urban normal. This temporal perspective is important for understanding whether observed patterns reflect temporary shocks or enduring transformations in travel demand.

2.3 Recovery Rate as a Measure of Transport Trajectory

In this study, the commuting pattern recovery is measured by a “recovery rate”, an index comparing post-pandemic underground usage against a 2019 baseline. Similar approaches and research designs have been used in various existing studies on public transport response towards COVID-19.

2.3.1 Recovery rate from temporal scope

From the temporal aspect, Ziedan et al. (2023) measured transport ridership during COVID-19 by designing a recovery rate. In their study, recovery rate was used on identifying change points within the pandemic time period, which are distinct sub-phases in the disruption and recovery process. To do so, they employed change point detection methods (Killick and Eckley, 2014), an algorithmic approach implemented in R, which enabled them to locate points where the underlying recovery trend shifted significantly.

Recovery rates are also designed for assessing overall trends, apart from change points. Zhang et al. (2023) analysed U.S. metro ridership data during COVID-19 by constructing recovery rates relative to pre-pandemic baselines. Their work was framed within resilience theory, which broadly describes a system’s capacity to absorb disruption and return to stability (Allen et al., 2016). To operate this, they used multiple resilience indicators (vulnerability, robustness, degree of return, and return time) to evaluate system performance. Compared with this study’s straightforward recovery rate calculation, their approach was more comprehensive, and was driven by spatial resilience theory.

Sharma et al. (2023) also designed recovery rate of public transport to study post-pandemic ridership within the approach of resilience theory. In contrast, they combined the temporal and spatial dimensions within their recovery rate design. Their approach calculated recovery curves at the station level and used the slope of these curves as a measure of recovery, framing this as an indicator of their spatial resilience test. In this way, the spatial structure of recovery were linked directly with the temporal recovery index.

2.3.2 Recovery rate from spatial scope

After defining recovery rate from both temporal and spatial angle, Sharma et al. (2023) aggregated these measures to census units of London (MSOA) for spatial modelling. The purpose of this step was to test how socio-economic characteristics of MSOAs shaped resilience. Then Geographically Weighted Regression (GWR) was used to examine how these characteristics such as population density, income, and car ownership explained differences in public transport recovery across London.

Forouhar et al. (2025) processed mobile phone footfall data of Toronto in a similar way. Also using 2019 as the baseline, they calculated transit recovery rates by comparing the mobile device count per neighbourhood before and after the pandemic. Neighbourhoods were then classified by proximity to nearby underground stations, creating “treated” and “controlled” groups. They applied Difference-in-Differences (DID) regression model in order to examine the relationship between proximity and recovery rates. Unlike traditional DID which isolates the causal effect of a single treatment, Forouhar included multiple socio-economic factors as explanatory variables, therefore the combined effects of proximity and community characteristics on underground transport recovery can be studied. Both Sharma et al. (2023) and Forouhar et al. (2025) approached the analysis of post pandemic transport recovery by designing a recovery rate from mobility data. Although with different regression models, they both examined the transport pattern with socio-economic variables.

2.4 Spatial analysis of ridership recovery

Moran’s I (Moran, 1948) as one of the most widely used statistics for measuring spatial autocorrelation, is used to analyse recovery pattern in this study. Global Moran’s I provides a single value that indicates whether similar values cluster together or are dispersed across an entire study area. As a result, it only offers a system-wide summary and does not reveal where clustering occurs locally. To address this, Anselin (1995) introduced local indicators of spatial association (LISA), which capture spatial structure at finer scales. Among these indicators, Local Moran’s I extends Moran’s framework by identifying the location and type of local clusters, distinguishing high value and low value clusters as well as spatial outliers. An alternative indicator is the Getis–Ord Gi* statistic (Getis and Ord, 1992; 1996), which focuses on detecting “hot spots” and “cold spots” of unusually high or low values, emphasising the intensity of clustering rather than outlier detection.

Global Moran’s I and LISA are applied in the spatial analysis of some post COVID public transport pattern studies. For example, after assigning recovery rates to London MSOAs, as discussed in section 2.3.2, Sharma et al. (2023) applied Local Moran’s I to this index in order to identify which MSOAs had statistically significant clustering. This allowed them to show where resilience to the pandemic was spatially concentrated, for example, distinguishing areas of London where recovery was consistently stronger or weaker than their surroundings. In this case, LISA served as a way to validate and visualise their resilience index, complementing their regression-based analysis of socio-economic drivers.

In a related application, Fischer et al. (2022) applied LISA to bike share ridership network in Canada, which was aggregated from the activity counts of each street segment. Unlike the areal identification of Local Moran’s I, Getis–Ord Gi* method could emphasis continuous spatial units like street segments with consistent high or low ridership change. This method gave the use of LISA tools in different spatial structures more flexibility, made it well-suited to line-based dataset of cycling routes in their study.

This study draws inspiration from Larcom et al. (2017) and Dunning’s (2012) forced experimentation definition. In the context of post pandemic mobility in London, recent papers have analysed social activity (Trasberg and Cheshire, 2023) and public transport demand (Gao et al., 2023; Murcio et al., 2023). The study adapted the design of recovery rate as ridership indicator from Zhang et al. (2023) and Sharma et al. (2023). Similar to Sharma et al., recovery is analysed from both temporal and spatial perspectives. In contrast, the temporal analysis here focuses on station-level recovery itself, rather than being framed within resilience theory. For spatial analysis, both LISA approaches are employed: Local Moran’s I to identify point-based clusters, with Getis–Ord Gi* providing a complementary view, in contrast to the line-based application of Fischer et al. (2022). Overall, this study provides a quantitative analysis of post-COVID recovery on the London Underground, identifying temporal dynamics and spatial patterns of ridership change at the station level.