Post-COVID-19 travel behaviour patterns: impact on the willingness to pay of users of public transport and shared mobility services in Spain

The survey was structured into three main sections: (1) socio-demographic information, (2) mobility behaviour in pre-COVID-19 pandemic times, and (3) the possible user adoption of post-COVID-19 adaptations of public transport and transport sharing services.

The survey included questions regarding both Revealed Preferences (RP) and Stated Preferences (SP). Firstly, RP questions characterise individuals’ socio-demographic and mobility behaviour, including travel frequency according to the purpose of the trip, and the transport mode was chosen for each particular purpose. These questions refer to the situation before and after the lockdown. Secondly, SP questions foresee the changes in the habits and collect respondents’ opinions towards some potentially implemented measures. These measures are related to possible post-COVID-19 adaptations of public transport and shared mobility services (car-sharing, moped scooter-sharing, bike-sharing, and kick scooter-sharing) and hailing services (including ride-hailing and taxi). Therefore, these questions capture individuals’ acceptance of each measure and their willingness to pay if they involve an extra cost that the transport operator transfers to the user. Finally, some control questions are also included to ensure that responses are coherent throughout the questionnaire. The information collected is exploited to conduct a modelling approach to design policies directed at maintaining the sustainability of the transport system. Table 1 shows the set of variables collected in the survey.

The survey was open to receive answers for two weeks, and 984 respondents participated with valid responses. We considered valid responses if the respondent would completely answer that subsection of the survey. All valid responses are used in section 3.

We adopt a choice modelling framework based on Heckman specification to explore individuals’ willingness to use and willingness to pay regarding specific transport modes in post-COVID-19 time. Five different transport options were included in the questionnaire: i) public transport, ii) car-sharing, iii) taxi/ride-hailing, iv) bike-sharing/kick scooter-sharing, and v) moped scooter-sharing. They all imply transport options in which cleanliness and sanitising would be managed by the operators, either public or private.

In this respect, willingness to pay values were only reported by the subsample of individuals who were willing to use a specific transport mode. According to Heckman [19], if a dependent variable is estimated just from a set of nonrandom observed values, thus modelling estimates may be biased. This fact is because there is a correlation between the errors that determine whether a case is unobserved/missing (in our case, individuals not willing to use a specific transport option) and the errors determining the outcome variable (in our case, willingness to pay). Heckman [19] proposed a two-step method in order to correct the problem of sample selection bias. Below, its most essential aspects are outlined.

Heckman approach is a two-equation model, eq. 1: selection, and eq. 2: outcome. First, willingness-to-use a specific mode of transport (Y_n) is a binary variable modelled in the selection equation. It indicates whether each case in the sample is observed or unobserved. Particularly, Y_n = 1 if the individual n would be willing to use a specific transport mode in post-COVID-19 times; 0 otherwise. Second, the dependent outcome variable to be estimated is y_n (willingness to pay for using a specific transport mode in post-COVID-19 times), which is an ordinal variable in our model. Willingness to pay was collected in the questionnaire, and is expressed in the model in relative terms (that is, percentage increase over current prices) to adapt to the different levels of prices and to purchase power across metropolitan areas and regions in Spain. Data on y_n are only available if Y_n > 0. Both variables y_n and Y_n are jointly modelled in the Heckman procedure to handle the sample-selection problem.

In the first step (eq. 1: selection), a binary probit regression is performed with the whole sample to determine the likelihood of Y_n being observed (Y_n > 0):

where X_np are vectors of regressors for the selection equation, γ_p are vectors of parameters to be estimated, 1(·) is the indicator function, and E_n is a random-error term.

In the second step (eq. 2: outcome), the modelling coefficients are calculated using the sample’s portion with observed values for the dependent variable solely (individuals willing to use a specific transport mode). An ordered probit model is adopted in terms of the probability of accepting a specific willingness-to-pay level. Then, we assume that the probability of a specific choice (the ordinal outcome y_n being equal to a specific value j_k), is given by the probability that the utility of individual n (U_n = β_p x_np) falls within τ_k and τ_k-1 thresholds:

where τ_1,…,m represent the thresholds defined, x_np are vectors of regressors for the outcome equation, β_p are vectors of parameters to be estimated and e_n is a random-error term. The model can be also expressed as:

Furthermore, we should note that the error terms E_n and e_n have a bivariate normal distribution with mean zero and the following variance matrix:

The Heckman procedure has been widely utilised for econometric analysis, mainly to calculate wage equations based on observed and unobserved wages (see e.g. [5]). In the transport field, the Heckman procedure has been used within the context of ride-sourcing platforms to analyse the two-step decisions made by drivers: to work or not, and if so, how long to work [34]. The reader is referred to as Xu et al. [37] to know more about the advantages of Heckman procedures concerning unobserved heterogeneity and selectivity bias/endogeneity problems simultaneously. Additionally, further details on the Heckman procedure for ordered categorical outcomes are provided in de Lucca & Perotti [13].

Although the questionnaire had been opened for two weeks in Spring 2020, most of the answers were received in the first days. At that time, the lockdown was still in progress. Thus, the responses received are conditioned by the widespread paralysis of economic sectors, the high uncertainty about the working future of many workers, and the break-in classes at schools and universities. Figure 1 shows the distribution of respondents by sex and age.

The evolution in the activity before and after the lockdown (Fig. 2) highlights that many workers expected either not to return to their jobs or telework at least a few days a week. The number of unemployed respondents is expected to double, from around 5% before lockdown to 10% after lockdown. These results have important implications for the number of trips.

Regarding modality for those who expected to continue working after COVID-19 lockdown, 38% reported that they would telework, 38% that they would work in person, and 24% do not know which modality they would do so. These responses reflect the general drop in travel frequency for commuting reasons. For instance, travelling daily for commuting reasons, including going to work/education centre, decreases by 30.3%, while travelling for this reason sometimes a week increased by 16.2%. However, as can be seen in Fig. 3, travel frequency is reduced for all trip purposes. Trips related to shopping/grocery and leisure activities, which in most cases took place once a week, reduced their percentage by 15.5% and 13.1%, respectively. Respondents expected to stop making a large number of trips. The trips that they expected to be avoided the most (7.5%) are those included in the “others” category. The following reasons are “work/study” and “leisure”. In the first case, it was expected that 5.6% of the trips would not be made, while in the second case, it was thought that 3.9% would be avoided. Only travelling for “care” reasons was expected to increase by 1.2%.

According to the responses obtained, the mode of transport mostly expected to be reduced is public transport. Its use was reduced for all travel purposes, especially for “work/study” (− 11.6%), “leisure” (− 8.8%) and “shopping/grocery” (− 5.4%). By contrast, the transport modes most expected to grow were walking and private bicycles and kick scooters. In the case of walking, it mainly increases regarding “leisure” trips (+ 4.5%). In the case of private bicycles and kick scooters, they mainly grow for “work/study” (+ 4.9%) and “leisure” (+ 3.1%) purposes. Shared transport services remain fairly constant, with slight differences between + 0.20% (shared bicycle/moped scooter for commuting or leisure) and − 1.1% (shared car/motorbike for commuting or leisure).

Ride-hailing services were also expected to remain constant for all trip purposes. Private car and motorcycle would also show few changes, with the most significant changes for commuting trips (+ 1.0%) and taking care of children (− 1.2%). A preference for the use of individual means of transport is then detected, with private car and walking/bicycle/kick scooter options standing out. The main reasons reported were fear of contagion and less congestion (20% and 14% of those who change modes of transport, respectively). The reasons for changing mode and starting to walk or ride a bike or scooter were the same.

These results are reasonable, especially given the high uncertainty regarding the future employment situation of many respondents and the limited information available about the virus when the survey was launched.

This subsection presents the main findings regarding individuals’ opinions towards the use of specific transport modes in post-COVID-19 times, namely: i) public transport; ii) car-sharing; iii) taxi/ride-hailing; iv) bike-sharing/kick scooter-sharing, and v) moped scooted-sharing. Respondents were asked whether they would use (willingness to use) a specific mode of transport if the operator would implement a specific measure. For individuals with an affirmative response were asked about their willingness to pay for these measures if they would imply an extra cost to be assumed by the user.

Preliminary findings on survey valid responses are displayed about the five modes, regarding both the whole sample (Table 2) and the subsample of potential users, i.e., individuals willing to use a specific transport mode (Table 3). As can be observed, the willingness to use transport modes in post-COVID-19 times greatly varies throughout the sample.

Public transport is the option with the highest willingness to be used. 89.7% of individuals reported that they would use these services in post-COVID-19 times (Table 2), which seems somewhat high, given that the survey was conducted in the critical period of the lockdown. Interestingly, around 64.3% of total respondents declared that they would pay more (compared to pre-COVID-19 times) for using public transport services if operators implemented sanitising measures.

As Table 3 shows, the main measures demanded among public transport users to use this mode are increasing supply to avoid crowding (70.6%) and increasing cleanliness and sanitising (52.1%). These findings may indicate that these measures seem to be enough to keep pre-COVID-19 levels in public transport demand and that citizens reasonably trust sanitising processes conducted under public transport authorities. It is worth noticing that 52.9% and 40.0% of these individuals would pay more, respectively, if operators implemented additional supply and sanitising actions. These results may indicate that individuals would not perceive such a dangerous option in public transport (in terms of sanitary conditions), or that they are captive of this transport mode and would use it in any case.

The willingness to use bike-sharing or kick scooter-sharing is reasonably high, around 67.7% among the sample (Table 2), which seems somewhat surprising given that demand for these transport alternatives was marginal in Spain in pre-COVID-19 times. This result may be explained by the fact that individuals find more comfortable or safer transport options that provide an open environment. Nevertheless, the willingness to pay particular measures against COVID-19 is relatively low: only 36.4%. The measures more demanded by people (Table 3) are, by far, the provision of covers for handlebars and steering wheels (51.3%), and the provision of masks, gloves, and sanitiser gel (38.1%). However, only 36.4% and 27.5% potential user would pay more.

The willingness to use taxi/ride-hailing services is in the same order of magnitude (66.4% of total respondents) than bike-sharing or kick scooter-sharing. We should note that 76.3% of people would use these services would demand increasing cleanliness and sanitisation, and 48.9% of them would pay more for it. Lower but noticeable percentages are observed for additional measures such as providing masks, gloves, and sanitiser gel before each use.

Similarly to taxi/ride-hailing, data for car-sharing reinforces the importance given by individuals to the increase of cleanliness and sanitising and the provision of masks, gloves, and sanitiser gel. These measures seem essential for transport options involving closed spaces and operated by private companies.

Tables 2 and 3 also include the distribution of survey responses concerning other transport modes addressed in the questionnaire, such as moped scooter-sharing. Comparatively lower positive responses (albeit noticeable) regarding this transport can be interpreted by the fact that it was a marginal transport option in pre-COVID-19 times in Spain. Moped scooter-sharing follow the same trend observed for bike-sharing/kick scooter-sharing, regarding the importance of providing covers for handlebars and steering wheels, masks, gloves sanitiser gel. Additionally, it is also relevant for this transport option to provide helmets with no contact with mouth, nose, and eyes (38.5% of potential users would demand this measure).

These trends should be observed in parallel with the potential influence that socio-demographic and mobility attributes may have on individuals’ responses (Table 2). Due to length limitations, only the most noticeable trends are commented. For instance, a noticeable higher proportion of females (75.4%) were willing to use taxi/ride-hailing services in post-COVID-19 times, compared to males (66.4%), which would reflect females’ preference towards ‘private’ modes. Additionally, as seems reasonable willingness to use bike-sharing, kick scooter-sharing, moped scooter-sharing, and car-sharing decreases with age, which is consistent with previous literature on shared mobility (see, e.g. [3]).

Similarly, the effect of age seems to be behind the fact that students present a higher willingness to use ride-hailing, car-sharing, and moped scooter-sharing options compared to other occupations. Finally, individuals who lost their job with COVID-19 lockdown reported lower willingness to pay for using specific transport modes, such as public transport or car-sharing, compared to the general sample. For instance, 40.0% of these individuals declared that they would not be willing to pay more for using public transport if additional sanitising actions would be implemented, compared to the percentage observed for the global sample (25.4%). This result is presumably explained by the loss of purchasing power generated by the pandemic for this segment of individuals.

It is also remarkable that demanding operators policies to only accept recently COVID-19-negative tested users while using their services are the least demanded option for every transport mode. However, potential users of car-based transport modes are more demanding: their willingness to use is 26.3% for car-sharing and 19.9% for taxi/ride-hailing, but only 17.6% and 13.6% would pay more for these kinds of measures.

Finally, there is a general reluctance to pay more, i.e., the low willingness to pay, for all the means of transport when respondents were also asked how much more they would pay at the service’s current price because of implementing those measures (Fig. 4). Interestingly, the most significant willingness to pay is perceived in public transport, even to a large extent accepting surcharges above 50%. This result is striking since it is the mode that would expect a higher decrease in the number of passengers. This high WTP occurs mostly among those who would be willing to use public transport if its frequency were increased and those who would do so if vehicles were cleaned and disinfected daily. This result shows that those who will continue to use it know this type of measure’s high economic cost.

A Heckman choice framework has been adopted to analyse individuals’ responses more rigorously. Explanatory variables used in the models were mostly categorical, so a base reference has been chosen in each case as referred to in Tables 4 and 5 when necessary. Multiple tests conducted for checking the presence of a strong correlation among the explanatory variables showed no multicollinearity problems in our data.

Table 4 shows the results for the ordered probit equation (willingness to pay, eq. 2: outcome), while results for individuals’ willingness to use a specific transport mode in post-COVID-19 times (eq. 1: selection) are included in Table 5 and commented below. Most of the explanatory variables that resulted in non-statistically significant were finally removed from the last version of the model with no impact in the overall fitting, as confirmed by multiple likelihood-ratio (LR) tests conducted.

The results for the selection equation confirm what was observed in the preliminary findings. Regarding socio-demographic attributes, we can observe that as age increases, individuals show a statistically significantly lower willingness to use car-sharing, bike-sharing/kick scooter-sharing, taxi/ride-hailing and moped scooter-sharing. Since these modes typically require the use of a smartphone, this result seems reasonable given the lower tech-savviness among older segments of the population (see, e.g., [24]. Additionally, modes such as shared bikes or mopeds require being in good physical condition (see, e.g., [2]). The positive and statistically significant results obtained for students regarding taxi/ride-hailing and moped scooter sharing can be interpreted in the same line.

Regarding gender, the higher likelihoods observed for females to use public transport and taxi/ride-hailing options have been widely referred to in the literature. Additionally, according to the modelling results, individuals with higher income levels typically present a statistically significant higher willingness to use taxi/ride-hailing services and a lower willingness to use public transport, bike-sharing, or kick-scooter sharing. This result can be explained by the higher car prone attitude of wealthy individuals and their tendency towards separating or differentiating from others as a signal of exclusivity as noted by Chevalier & Gutsatz [11].

Regarding mobility habits, the most noticeable results concern the lower likelihood to use public transport services in post-COVID-19 times among individuals who commuted intensively by using any other modes: private vehicle, personal bike, walking, shared modes, among others. Nevertheless, we should remind that overall willingness to use public transport services in post-COVID-19 times was high in the sample, as noted above. Detailed results concerning the frequency of use regarding any trip purpose and transport mode can be observed in Table 5.

Table 4 includes the modelling results regarding the Heckman regression equation (willingness to pay, eq. 2: outcome). As can be observed, socio-demographic variables play a minor role in explaining individuals’ willingness to pay. Only statistically significant results were found for individuals losing their job after the beginning of the COVID-19 lockdown regarding car-sharing (p-value = 0.000), scooter-sharing (p-value = 0.033), and public transport (p-value = 0.058), which is reasonable given the loss in the purchasing power of this segment of the population. Other socioeconomic variables such as gender or age did not result statistically significant for transport services analysed.

Regarding explanatory variables controlling for mobility habits, many coefficients appeared as statistically significant. Mainly, individuals who commute (intensively or not) by private bike present a lower willingness to pay for using bike-sharing systems, which seems evident given that using their own bike is cheaper and safer against contagion. Nevertheless, the opposite effect is observed for those who use their private bike non intensively for leisure trips (p-value = 0.008). Additionally, occasional users of public transport for commuting purposes showed a statistically significant lower willingness to pay for additional costs for sanitary measures on public transport (p-value = 0.036), car-sharing (p-value = 0.018) and hailing (p-value = 0.022) services, in case an additional cost was imposed to the user for implementing sanitation measures. Detailed results concerning the influence of mobility habits on individuals’ willingness-to-pay can be observed in Table 4.

More interestingly, Table 4 includes modelling results controlling for willingness to pay in specific modes of transport and potential measures to be implemented during post-COVID-19 periods. Reasonably, increasing service supply to avoid crowding (p-value = 0.018) and, to a lower extent, providing masks, gloves, and sanitiser gel (p-value = 0.099) would increase individuals’ willingness to pay for using public transport. Together with the preliminary findings above, these results are significant given the additional costs these measures imply to operators and current financing problems on public transport services. By contrast, no measures are statistically significant for car-sharing options, although one may expect that, e.g., improving cleanliness and sanitising or providing covers for handlebars and steering wheels, would increase willingness-to-pay for these services. Thus we can conclude that car-sharing users do not value sanitising measures in such a way to lead to a higher willingness to pay.

Similar to the findings on public transport, for bike-sharing/kick scooter-sharing we can observe (Table 4) that several measures could lead to a higher willingness-to-pay, namely: providing masks, gloves, and sanitiser gel (p-value = 0.035); certifying that no user within the same day was infected by COVID (p-value = 0.047); providing helmets with no contact with mouth, nose, and eyes (p-value = 0.076) or providing covers for handlebars and steering wheels (p-value = 0.000). Surprisingly, cleanliness and sanitising actions negatively influences the willingness to pay for these services. One may conclude that this type of measure is viewed by individuals as prerequisites for using a transport services in post-COVID-19 times, but not necessarily an element to lead to a higher willingness to pay, despite the greater costs it imposes on operators. Additionally, it seems that bike-sharing users demanding the improvement of bike infrastructure (e.g., extending current bike lanes) would be willing to pay more in a statistically significant way since this result is close to being statistically significant (p-value = 0.096).

Finally, modelling results for moped scooter-sharing present many similarities with car-sharing since almost no measure increases willingness-to-pay for these services. The only exception would be certifying that only individuals non infected by COVID-19 could use this service, but this result is not statistically significant (p-value = 0.110). Thus we can again conclude that, despite the higher costs that sanitising measures impose on operators, they are not valued by users of moped scooter-sharing in such a way to result in a higher willingness to pay. On the contrary, are seen as prerequisites for using the service.