The effects of free-fare public transportation on the total active travel in children : A cross-sectional comparison between two Finnish towns

Background: Free-fare public transport policies potentially increase walking to transport stops, but may replace other active travel modes with unknown effects on total active travel. We compared active travel behaviour of children living in a Finnish city with a free-fare public transport to a reference city. Methods: Children were recruited from primary school grades 4 – 6 in 21 participating schools, located in 11 neighborhood pairs from Mikkeli (free-fare bus) and Kouvola (no free-fare bus). Children marked all places they visited the previous week, visit frequency, and travel modes into a participatory mapping survey during a lesson. Active travel was assessed as a sum of all round-trips from home to destination by walking (4 km/h), walking-to-bus-stop, skating/scooting (7 km/h) and cycling (10 km/h). Active travel was compared between cities with linear mixed effects models. Results: A total of 427 children respondents (age mean and SD 11.0 ±


Introduction
Walking and cycling to school is common in Finland, with 70% of primary and lower secondary school students commuting actively (Tammelin et al., 2016).Active commuting to school can form more than half of daily moderate-to-vigorous activity time in urban areas, and one third in rural areas where active travel behaviour is less common (Rainham et al., 2012).Walking is most common at short trips, and both walking and cycling effectively decrease at distances exceeding 5 km, typically being replaced by car or public transport commuting (Fyhri and Hjorthol, 2009;Kallio et al., 2016;Børrestad et al., 2011).Active travel therefore decreases with increasing distance to destinations, like in suburban and rural areas (Rainham et al., 2012).
Active travel has an important role in public health.In 2016 Finland's report card on physical activity, active school commuting was graded on level B (61-80% commuting actively), although the total physical activity was graded only on level D (21-40% meeting the physical activity recommendations).Despite the still positive situation, active travel to school has decreased in Finland during the past decades.This decreasing trend may be due to a couple of reasons.The number of schools has decreased by 27% between the years 2000-2011 in Finland, which has resulted in a more centralized school network (Mehtäläinen et al., 2013).Potentially due to longer school-travel distance the proportion of primary school children driven to school by their parents has increased from 16 to 20% in Finland, and the trend has been similar in other Western countries (Turpeinen et al., 2013;Fyhri et al., 2011;McDonald, 2007;Salmon et al., 2005;van der P et al., 2008).
A total of 63% of the Finnish population lives out of inner urban areas, with an 80% higher car use in these areas as compared to inner urban areas (Helminen et al., 2020;Liikennevirasto, 2018).Finnish Basic Education Act guarantees free transportation for more than 5-km school commutes, enabling children to reach school with ease.During non-school time, children living far away from their places of interest are dependent on family resources through being driven or using public transportation (Liikennevirasto, 2018).In particular, public transport accessibility is positively associated with after-school activity participation rate in youth (Palm and Farber, 2020), while a total of 23% of those children and youth not participating in organized sports reported "not having a ride" as their main reason for non-participation (Blomqvist et al., 2019).The dependency on public transport can be particularly high in low income families having compromised possibilities for car ownership or use (Brockman et al., 2009;Sallis et al., 2000), whose children on average are less active than those from higher income families.Non-school active travel, and public transport use in particular, are less studied areas despite their potential contribution to total active travel duration and importance in children's daily life (Tammelin et al., 2016;Voss et al., 2015).
Public transport use requires accessing transport stops and the final destination often by foot or bike and therefore can effectively increase daily physical activity in those frequently using public transport (Voss et al., 2015).Elementary school children and secondary school youth using school bus walk for a similar distance and duration per trip and accumulate the same moderate-to-vigorous activity as compared to those directly walking to school (Voss et al., 2015;Owen et al., 2012), with both walkers and bus commuters being significantly more active per trip and during the whole day than those driven to school by car (Owen et al., 2012).Therefore, promoting public transport instead of car commuting can be an effective means to increase physical activity especially in areas where independent active travel by bike or foot is too laborious due to long distances, or otherwise inaccessible by children.
Fare compensation policies are widely used to increase public transport use.In the Finnish context, free-fare transport is provided for long school commute, but policies and service availability vary across cities for non-school travel.From public health and physical activity perspective, these services should be evaluated in relation to their effects on other travel mode use, and on total active travel duration.Many of free-fare transport experiments have resulted in increased public transport use, but against expectations, at the expense of active travel modes, but having minimal effect on car use (Cats et al., 2017;Storchmann, 2003).On the other hand, the effects can be age-group specific, with particularly positive physical activity benefits seen in older adults (Webb et al., 1978;Laverty et al., 2018).The few investigations in children have provided mixed results, such as replacement of short (<1 km) walking trips with bus, but decreased car use in urban London, UK (Edwards et al., 1978;Green et al., 2014), and decreased proportion of walking trips in Tallinn, Estonia (Cats et al., 2017).Because places of interest are generally more accessible in compact urban than rural areas, and public transport may be particularly important for children having compromised access to their places of interest, the experiments focusing solely to urban environments do not necessarily capture the optimal benefits of free public transport policies.In addition, the previous free-fare policy investigations have not comprehensively evaluated the contribution of walking to transport embarkation/disembarkation station on children's total active duration, and compared travel behaviour across public transport accessibility measures (urban vs. rural areas), and between those who do and do not use the free-fare transport service.
The city of Mikkeli has provided free-fare public transport to all primary and secondary school children since 2017.The purpose of this study was to compare the active travel behaviour of 10-12-year-old children living in Mikkeli to those living in Kouvola, a reference city without free-fare public transport.We hypothesize that children eligible for free-fare transport have a higher active travel duration.

Methods
The detailed study protocol has been published (Pesola et al., 2020).Due to the COVID-19 related local recommendations to avoid unnecessary visits to schools, we were forced to truncate our protocol and refrain from monitor-based activity measurements.Permission to recruit from schools was asked from the responsible authorities and school principals, and the recommendation compliant sampling procedures were discussed with the principals and teachers case by case.Children were asked for oral consent, and their legal guardians were asked to provide informed written consent.The study protocol has been approved by Aalto University Research Ethics Committee on 10th October 2019.The truncated methods used in the present study are described below.

Protocol
Mikkeli (54,000 residents) and Kouvola (83,000 residents) are small towns located in South-Eastern Finland and have a similar climate, geographical structure and possibilities for active transit, including an active bus network for local traffic.Children were recruited through primary schools located in neighborhoods that were paired between the towns based on objectively analyzed public transit accessibility.The detailed data used for neighborhood pairing is presented in the protocol paper (Pesola et al., 2020).The total number of primary schools is 17 in Mikkeli and 24 in Kouvola.As per the study design, we contacted a total of 22 schools, 11 from each city (Pesola et al., 2020).One school in Kouvola refrained from data collection due to local COVID-19 restrictions, and the final sample was recruited from 11 schools in Mikkeli and 10 schools in Kouvola.The contacted classes had a total of 331 pupils in Mikkeli and 369 pupils in Kouvola, totaling 700 pupils who had a possibility to participate in the study.Data collection was done between 8 th of September to 16 th of November 2020.To minimize weather and other condition confounding, the recruitment and data collection was done simultaneously between the paired Mikkeli and Kouvola schools.
During the time of data collection, the regional COVID-19 situation was either at the baseline (defined based on low incidence) or at the acceleration phase (defined as less than 25 confirmed cases per 100,000 population during the past 14 days).All pupils were visiting school and hobbies normally, without overall regional restrictions for these activities.Data collection was done either at the school (n = 17 schools) or remotely (n = 4 schools) to comply with the local COVID-19 restrictions regarding external visitors.Children were asked to return the informed consent forms signed by their legal guardians to the teacher, and the researcher collected the signed forms before the data collection.Data were collected during one lesson totaling approximately 45 min.

Measurements
The child PPGIS questionnaire was completed on a computer browser (Maptionnaire® tool) either during computer class or at home simultaneously with their class peers.In both cases, the researcher first demonstrated in person or via video conference software how to complete the questions and navigate the map on the browser.The researcher was available for questions during the data collection period.Children were asked to mark on a map their home, school, hobbies, and other destinations that they had visited during the previous seven days.After each place marking, they were asked questions about each place, including on which days (Monday-Sunday), how (walking, cycling, scooting, by car, by bus), and with whom (alone, with peers, with adults) they visited the place.If a destination was visited more than once per day, children were asked to mark the same place multiple times.After the map-based questions, subjective accessibility of public transport [41], and children's use of public transport were asked.

Analyses
Primary outcome: active travel duration.A geospatially derived active travel duration was measured based on the mapped places and their travel mode.Distance between respondents' home and each marked point (destination) was analyzed using a Network Analysis extension tool by Esri ArcGIS Pro© (Redlands, CA, USA).The network analysis extension calculated the shortest way to get from home point to all destination points.The advantage of this method is mapping the spatial context of physical activity, in addition to merely estimating the transportation time and leisure physical activity time.
Average walking speed was set at 4 km/h based on previous studies.An estimated walking speed in English 7-12 year old children is 4. 1-4.5 km/h (Whittle, 2007).When considering crossing the roads, waiting in traffic lights, and the changing terrains, a value of 4 km/h has been previously used in investigations in Finnish children (Turpeinen et al., 2013).Average cycling speed was set at 10 km/h based on data from Swedish children, who were cycling at 13 km/h on a closed track (Briem et al., 2004).A value of 10 km/h has been previously used for Finnish children considering that free living cycling is slower than track cycling (Turpeinen et al., 2013).We used an estimated speed of 7 km/h for scooting, which is an average of walking and cycling.

Fixed effects
Objective bus accessibility was defined by using the YKR Urban Zones dataset provided by the Finnish Environment Institute (SYKE).YKR Urban Zones data is a 250 m × 250 m grid-based dataset in which all city regions in Finland are divided into different zones based on three main criteria: distance to the city center, public transport frequency and walking distance to public transport stops.Furthermore, these criteria are calculated for each YKR grid cell which is then assigned a value indicating if it belongs to walking, public transport, car zone, or has no YKR value indicating sparsely built and populated rural areas (Syke. Model of Three Urban, 2013).In this study, each child's home coordinates were assigned to one of the three objective bus accessibility levels: intensive public transport zone (YKR walking and public transport zones combined), car zone, or rural zone.
Actual bus use.Bus use for purposes other than school commuting was asked in order to separate free-fare transport service in Mikkeli from free school transport guaranteed by the Finnish Basic Education act.Children were asked whether they used bus for purposes other than school commuting daily, several times per week, once or twice per week (weekly bus users), less than weekly, or never (less than weekly bus users).The two level threshold was defined based on best model fit in linear mixed models.
Subjective bus accessibility is an average of ratings of accessibility of school, hobbies, best friend's house, and any other usual place by bus on a 7-level Likert scale (Lättman et al., 2016).A threshold value of more than 5 was used for good, and a threshold value of 5 or less for compromised subjective bus accessibility, based on best linear mixed model fit.

PPGIS data and spatial analyses
PPGIS data is characterized by two distinct types of data; spatial and non-spatial (Fagerholm et al., 2021).The spatial data is collected through mapping tasks in the PPGIS survey (Maptionnaire®), and can include data attributes such as points, lines, or polygons.The mapped spatial entries return a data set where each line represents one spatial entry mapped by a respondent.The non-spatial data is collected through traditional open or structured questions within the same survey, which returns an additional data set where each line of the data represents one respondent.
The spatial data cleaning and analyses were performed in ArcGIS Pro version 2.9.0 and QGIS version 3.10.9.The original data was downloaded in three intervals from Maptionnaire® software as a .csvfile and transformed into spatial data using Well-known text (WKT) coordinates.Initial data cleaning included removing entries made by researchers (who submitted an entry through demonstrating the PPGIS activity in class), removing duplicate entries (respondent duplicates and spatial duplicates), removing respondents who did not consent to participate in the final research (responses to basic survey questions and their spatial entries) and finally returning the spatial data into csv.and xlsx.formats for further statistical analysis.The final PPGIS data set included 427 respondents and 2445 points mapped.A detailed flow chart describing the data cleaning process is provided in Supplementary Fig. 1.For the active travel duration calculations, we further excluded 357 points which did not include information about the mode of transport, 106 points because the visitation frequency was not reported, 39 points which were marked outside the study area, or because of extremely long travel distances (e.g., >20 km by walking one way) and finally 603 points which were accessed by car.Thus, the final active travel measures included 1340 points.
After the data cleaning we performed Network Analysis between respondents' home and all of their mapped places, to calculate travel distances (Fig. 1).Network distance to all points from home was calculated using ArcGIS Pro and ArcGIS Online's network road dataset.Points in the analysis must be linked to locations that exist alongside existing roads and streets.In the data there were multiple destination points that were too far from the nearest available street network segment and in those cases the points were snapped to the nearest point in the network.The snapped Euclidean distance between the point location and the street network is then calculated to the total network distance.In addition, we calculated network distance to nearest public transit stop from home and from all destination points to which the respondent had reported traveling by public transit.This distance was used in calculating the active travel share of the public transit trips.
First we calculated total trip distance to each point by multiplying the total network distance by two to each point that were visited by walking and cycling, indicating a round-trip between home and each destination.Active travel part of public transport trips was calculated by multiplying the distance to nearest bus stop from home by two and to nearest public transport stop from destination points by two.The trip distance was multiplied by the reported visitation frequency and calculated separately for weekdays and weekends.The final active travel trip distance outcome variables were calculated separately for walking and cycling for each respondent as follows: where H1 is the respondents total active travel trip by walking and cycling, a is the visitation frequency and b is the distance between home and marked point.
The active travel of public transit trips was calculated as follows: where H2 is the respondents total active travel trip by walking to bus stop, a is the visitation frequency and b is the distance from home to nearest public transport stop and c is the distance from destination point to nearest PT stop.Similar process as with walking and cycling was used also for car trips.In addition, we assigned an urban structural zone for each respondent using the YKR zoning data by SYKE.

Sample size
Minimum difference of interest (MDI) is 15 min/day of active travel.We expect a 6% (ρ = 0.06) school-level intraclass-correlation and assume average n per cluster of 20 (design effect = 2.1 estimated with formula 1+ (n-1)ρ).A sample size of 210 for each city spread across 10 clusters is required to have ≥80% power to detect MDI at 5% alpha error level (two-tailed significance).The MDI of 15 min/ day of active travel is clinically meaningful and an approximate duration of a short (~1 km) walk to school (Kallio et al., 2016;Ekelund et al., 2012).

Statistical analyses
Statistical analyses were performed in RStudio Version 1.3.1093(RStudio, PBC, Boston, MA).Statistical significance was set at p < 0.05 (two-tailed).Background variables table and statistical comparisons were performed with compareGroups package (Subirana et al., 2014).A linear mixed effects model (lmer4 package fit with REML (Bates et al., 2015)) was used to compare the active travel duration between cities (fixed effect) using the pre-determined regions as a random effect.Square root transformed response variable was used to reach normally distributed residual variance based on visual inspection of base R Q-Q plots (The R Development Core Team, 2022).The main model was extended with pre-determined fixed variables, including objective bus accessibility (three levels: intensive public transit zone, car zone, rural zone), subjective bus accessibility (two levels: good accessibility, not good accessibility), and actual bus use (two levels: weekly, less than weekly).Models using different categorization levels (e.g. for more frequent bus use)

Table 2
Linear mixed models examining the differences in total active travel duration.for the fixed variables were compared and those models with a better fit based on Akaike and Bayesian information criteria are presented.Models for comparison were fit with maximum likelihood method.Random slopes were considered but were omitted due to lower model fit and model complexity.Because of the difference in distribution of classes between cities, class was considered as a covariate.However, the estimates were unchanged and the model fit decreased considerably, resulting in abolishing this covariate.Estimated marginal means were calculated using ggeffects package and visualized with ggplot2 package (Lüdecke, 2018;Wickham, 2016).Given a right-skewed distribution, density plots and boxplots were further used to illustrate distributions of active travel duration by travel mode between cities across public transport accessibility zones and across different destinations (school, hobbies and sport, other).Unpaired Wilcoxon tests were used to explore statistical differences between distributions within these factor levels.These visualizations and related statistical comparisons were performed using ggplot2 and ggpubr packages (Wickham, 2016;Kassambara, 2020).

Results
Background characteristics of children living in Mikkeli and Kouvola are presented in Table 1.The distribution of class varied between cities, such that there were more 5 th grade children in the Kouvola sample (P < 0.001).The distribution of school commute modes was different between cities (P = 0.003).During the day of data collection, 18% of children in the Mikkeli sample reported commuting to school by bus, compared to 8% of those living in Kouvola.A higher proportion of Mikkeli children reported using bus weekly during non-school time (P = 0.001).A total of 22% of Mikkeli children reported using bus during non-school time at least once per week, as compared to 10% reported by children living in Kouvola (Table 1).
The spatial analyses were based on a total of 2445 mapped places, of which 1340 places were visited with a travel mode including an active component (walking, cycling, scooting or walking to bus stop).On average, children mapped 5.7 ± 2.3 places and made 13.8 ± 7.3 trips per week.Children reported a total of 91 and 36 bus trips, of which 15 and 11 were for school commuting and 76 and 25 for non-school commuting in Mikkeli and Kouvola, respectively.Moreover, Mikkeli and Kouvola children reported a total of 377 and 361 walk trips, 428 and 749 cycle trips  1).
Linear mixed model results are presented in Table 2 and estimated marginal means are visualized in Fig. 2. Estimated total active travel duration was 3.19 h/week (95% CI 2.62-3.83) in Mikkeli and 3.33 h/week (95% CI 2.74-3.97) in Kouvola.There was no difference in total active travel duration between the cities (P = 0.749, Table 2 Model 1).The extended models revealed no significant interactions between city and objective bus accessibility (Mikkeli x intensive public transport zone P = 0.612; Mikkeli x car zone P = 0.207, Table 2 Model 2), subjective bus accessibility (Mikkeli x good accessibility P = 0.113, Table 2 Model 3), or actual bus use (Mikkeli x bus use weekly P = 0.305, Table 2 Model 4).Within Kouvola, active travel duration was longer in car zone (3.73, 95% CI 2.96-4.59h/week) as compared to rural zone (2.47, 95% CI 1.63-3.49h/week, P = 0.035, Fig. 2).To study the effect of free-fare policy alone without considering the free-fare school bus for children living more than 5 km away from school, which is available in both cities, sensitivity analysis was conducted by excluding the active travel share of school bus from total active travel Fig. 2. Estimated marginal means of total active travel duration in main and interaction models.A.J. Pesola et al. (Supplementary Table 1 and Supplementary Fig. 2).However, the main findings and interactions remained unchanged.
Fig. 3 shows walking (other than walking-to-bus-stop), walking-to-bus, and cycling duration within public transit accessibility zones between Mikkeli and Kouvola.Overall, children in Mikkeli were walking to bus stop more, and children in Kouvola were cycling more (P < 0.001, Fig. 3).Children were walking more in Mikkeli, but cycling more in Kouvola, in intensive public transport and car zones (P < 0.05, Fig. 3).Children living in Mikkeli were also walking to bus stop more in Mikkeli as compared to those living in Kouvola, in the car zone (P < 0.001, Fig. 3).
Fig. 4 shows the duration of active travel modes for school commuting, traveling to sport and hobbies, and for other reasons between Mikkeli and Kouvola.Children living in Mikkeli were walking and walking to bus stop more when commuting to school, and walking to bus stop more to access sports and hobbies, as compared to Kouvola children (P < 0.05, Fig. 4).Children living in Kouvola were cycling more to school, sport and hobbies, and other destinations, as compared to children living in Mikkeli (P < 0.05, Fig. 4).

Discussion
Free-fare public transport experiments are locally important endeavours to increase public transport use.Due to the potential benefits on city image and transportation behaviours, yet significant public costs, there is often political debate around these experiments (Cats et al., 2017;Fearnley, 2013).Advocates of Mikkeli's free-fare public transport system for children have commented that the free-fare transport would decrease the need for extra school transportation services, replace the need for parent's chauffeuring children to destinations, and increase equality of access to hobbies during leisure time, ultimately leading to increased total physical activity.Others have criticized the decision by pointing out that free-fare public transit would replace children's active commuting by bike and foot.The main finding of the present study was that children eligible for the free-fare transport have a similar total active travel duration as compared to children living in Kouvola, a reference city.While the findings do not support the hypothesis that free-fare transport would increase active travel, they also do not support the criticism that total active travel (including walking, cycling, walking to bus, and scooting combined) would decrease.On the other hand, the free-fare experiment was associated with less cycling, which can be one reason for the concerns.However, overall, children were using the bus more for school travel and for accessing hobbies, which resulted in more walking associated with bus use.
Previous investigations have shown an increase or no change in youth's bus ridership following introduction of free-fare public transport policies, with most changes evident for short trips in urban settings (Cats et al., 2017;Edwards et al., 1978).In the present cross-sectional study, the number of bus rides was on average 87% higher in Mikkeli as compared to Kouvola, of which 84% were non-school-related travels suggesting effectiveness of the free-fare policy, instead of government-backed free school travel for long travelers which is available in both Mikkeli and Kouvola.The difference in ridership as compared to the reference city is larger than a previously reported 21% increase in young people's ridership share in a longitudinal analysis of free-fare transport in Tallinn, Estonia, or no change following London's free-fare policies.Bus trips covered only 6% of all trips in Mikkeli, whereas the shares of walking, cycling and car trips were 26%, 30%, and 29%, respectively.The share of public transport trips was only 2% in Kouvola, similar to previous reports from Finland, England and USA (Edwards et al., 1978;Durand et al., 2016).However, the public transport trip share is higher in Finnish metropolitan area (34% for school commuting), and as high as 80% following the Tallinn free-fare policy, suggesting differences in data collection methodology, better public transport accessibility, or other reasons (Cats et al., 2017;Broberg and Sarjala, 2015).The present findings show that free-fare policy can considerably increase public transport use also in a small city, yet given a small trip share as compared to other travel modes, there is room for policies to further increase ridership.
Previous studies have reported that children replace short walking trips with bus traveling, with the number of young people's walking trips decreasing at half following the free-fare policy in Tallinn (Cats et al., 2017;Edwards et al., 1978).However, the previous studies have not reported changes in total active travel duration, including sum of walking to transport, walking, cycling and skating/scooting.(Cats et al., 2017;Edwards et al., 1978).The Tallinn study also used data from detailed one-day travel diaries in 15-19 year old youth, but active travel duration was not reported, and it is unclear if walking to transport stops were included as part of walking trip share (Cats et al., 2017).The London experiment was evaluated based on travel diaries where 12-17 year old young marked the start-point, interchange, end-point and travel mode of each trip, and individual stages of each trip were disaggregated, possibly including walking to transport stops (Edwards et al., 1978).This is a similar approach used in the present study, importantly considering the active share of all trips, yet in the present study we also measured total active travel.Furthermore, we studied interactions with factors possibly modifying the effect of free-fare policy on total active travel.When examining the confidence intervals in linear mixed effects models (Fig. 2), children in Mikkeli using bus weekly had a greater total active travel duration as compared to those using bus less than weekly, while such a difference was not evident within Kouvola.Therefore, free-fare bus can facilitate total active travel.Another explanation could be that children already having active travel habits are more likely to take advantage of this service.On the other hand, confidence intervals in linear mixed effects models show that within those having a good subjective public transport accessibility had a lower total active travel duration in Mikkeli as compared to Kouvola.This suggests that association between subjective bus accessibility and total active travel is not straightforward, particularly in interaction with the free-fare policy.This difference, however, is uncertain given wide confidence intervals.Taken together, the present detailed trip-based data, including the share of walking to transport, provides a more positive landscape for the free-fare public transport possibilities.The median distance walked to bus stop was 730 m, which was more than 200 m longer than other walking trip median, supporting the importance of this active component on total active travel duration (Voss et al., 2015).Despite travel mode differences between Mikkeli and Kouvola, including less cycling in Mikkeli, free-fare transport policy is not associated with decreased total active duration because of greater walking associated with bus use.
A theoretical background for free-fare policies has been well presented in transport economics literature (Cats et al., 2017).Previous research has shown that people gradually adjust their behaviour into pricing changes.Given that the Mikkeli free-fare service has been effective since 2017, the present results are likely depictive of habitual change.Yet, the effect is asymmetric such that fare increase causes a larger reduction in ridership as compared to smaller increases in ridership following price reductions (Chen et al., 2010).However, fare pricing is unlikely to be a direct barrier or facilitator of children's ridership but is a matter of how parents financially support their child's public transport use.In the case of free-fare transportation, it is conceivable that children can more independently select their travel mode without financial constraints.However, other factors like distance to bus stop, population density, access to car, cultural factors, or beliefs and attitudes, can also determine public transport use (Cats et al., 2017;Buehler, 2011).One of the unwanted outcomes from the free-fare transportation experiments has been unnecessary joy-riding and vandalism (Storchmann, 2003).The present data shows that primary school children reported their most rides being for school commuting, accessing hobbies and sport, and therefore such unwanted scenarios were not evident.
Previous free-fare and other public transport studies have been primarily conducted in urban settings, and less data exists from rural areas (Owen et al., 2012;Cats et al., 2017;Edwards et al., 1978;Green et al., 2014).It is likely that the need for public transport increases the further away children's places of interest are located.The median public transport distance was 6.4 km, similar to the median distance of 6.5 km travelled by car.However, there was no difference in public transport use between Mikkeli and Kouvola within the rural zone, suggesting that the free-fare policy does not increase public transport use in rural areas where the need is greatest.This can be partly explained by the fact that free school transport is already available for all children living 5 km or more away from school in both Mikkeli and Kouvola, leaving less space to see increased bus use following the Mikkeli experiment.Other reasons can be that the number of available trips and bus stop network area become sparser further away from the city center, making it more difficult to benefit from the free-fare buses (Pesola et al., 2020).Previous studies have shown that good public transport availability, like short distance to and a high density of bus stops, is associated with a greater public transport use and physical activity recommendation attainment in adults (Besser and Dannenberg, 2005;Rissel et al., 2012;Villanueva et al., 2008;Djurhuus et al., 2014).Although the number of bus stops near school is greater in rural zone regions (7-12 stops) than in car zone regions (1-12 stops), the number of public transport trips per weekday is lower in rural zone regions (8-16 trips per day) than in car zone regions (9-39 trips per day) (Pesola et al., 2020).One way to increase public transport use in rural areas would be to increase available bus trips which, however, is costly.Other possibilities would include behavioural interventions to promote public transport use, like modified versions of walking school buses and educational strategies that have been effective in increasing active travel to school (Jones et al., 2019).In the present data only a few children accessed bus stop by cycling.A possibility to carry one's bike at the bus (e.g. in San Francisco MUNI buses), secure bike parks close to the bus stops, or last mile bike rentals, would enable reaching bus stops that are further away, and destinations that are further than cycling distance away from home.For example, in the Netherlands everyone can rent a bike for last mile of their transit trip with a discount using their transit card.Such ideas should be refined together with children and families and tested in the field, together with other policies and interventions aiming at increasing active and independent travel to school and other destinations during the whole day.
This study is very policy-relevant and it provides the data to support what is currently happening in the free-fare bus experiment in Mikkeli, Finland.Other strengths of this study include a large sample of children, and a pre-determined cluster-based data collection scheme from two cities that are comparable in their public transport accessibility structure.Although we were forced to abolish the device-based measurements due to COVID-19 restrictions, public participatory mapping is an established methodology to investigate detailed travel pattern behaviour and has been previously used in children (Ikeda et al., 2019;Broberg et al., 2013;Kyttä et al., 2018) .While the trip distances are likely very accurate, we used the same estimated travel speed for all children, and assumed that all trips originated and ended at home.This approach does not consider individual and complex travel trajectory data, which would provide even more accurate space-time information of the individual travel behaviour and the origins of each destination (Yue et al., 2014).In future studies spatially explicit trip-chaining could be studied by having the respondents mapping all daily routes, with travel tracking mobile apps or using GPS devices.However, GPS studies are resource and data heavy, and using mobile tracking apps remains difficult due to questions of data ownership and privacy.We also could not disaggregate the possible active component of car trips, which however is likely small (Voss et al., 2015).The COVID-19 pandemic can influence children's travel patterns, and these effects will be reported in a separate manuscript.On average 41.8% and 24.9% of children in Mikkeli and Kouvola, respectively, reported reducing their public transport use during the pandemic as compared to time before, while about a half reported no changes in public transport use (manuscript under preparation).Therefore, these results may be an underestimation of the effects in pre/post pandemic and should be further evaluated in the future.The data collection was scheduled for non-snow season in order to provide generazible data.Active travel decreases during winter time in Finland, and free-fare public transport use can be more common and the positive effects on active travel amplified during snowy season (Kallio et al., 2016).Future studies should use longitudinal designs, device-based measures of total physical activity time, and consider different travel linking scenarios (like cycling to bus stop), and differences in urban, suburban and rural transportation needs and possibilities in order to capitalize and evaluate the potential of free-fare transport policies.

Conclusion
Free-fare public transport can have an effect on children's travel behaviour.The present study showed that children eligible for free-fare transport were cycling less, but walking more to bus stop as compared to children living in a reference city.The active component of accessing public transit should be considered as an active travel mode, because it can contribute to total active travel duration of children.These results were most evident in zones with good public transport accessibility.However, the total public transport use share was small (6%) and there were no differences in public transport use in rural areas, suggesting that future policies and interventions should focus on improving public transport access and use in children who most need it.

Fig. 1 .
Fig.1.An example of the spatial network analysis between one respondent's home and all mapped places.
, 22 and 22 skate trips, and 418 and 478 car trips, respectively.The median length of the marked trips were 521 m [258 m; 941 m] by walking, 730 m [505 m; 1017 m] by walking to bus stop, 872 m [507 m; 1692 m] by skating or scooting, 1794 m [1179 m; 2554 m] by bicycle, 6407 m [4233 m; 9364 m] by bus, and 6475 m [3859 m; 10430 m] by car.The distribution of bicycle trip and public transit trip numbers differed between cities, such that children in Kouvola marked more bicycle trips (P = 0.006), and children in Mikkeli marked more public transport trips (P < 0.001, Table

Fig. 3 .
Fig. 3. Comparison of walking, walking-to-bus and cycling between Mikkeli and Kouvola within different bus accessibility zones.Density plots show greater details on data distribution within both cities.

Fig. 4 .
Fig. 4. Comparison of walking, walking-to-bus and cycling between Mikkeli and Kouvola when traveling to school, sport and hobbies and to other destinations.Density plots show greater details on data distribution within both cities.

Table 1
Background characteristics and spatial variables by city.
Data are expressed as mean ± SD, number (%) or median [interquartile range] where specified.aSubjectivebusaccessibility is an average of ratings of accessibility of school, hobbies, best friend, and any other usual place by bus on a 7-level Likert scale.A threshold value of more than 5 was used for good bus accessibility based on best linear mixed model fit.A.J.Pesola et al.