1. INTRODUCTION
In many European countries, remote work and flexible work locations is increasing (Eurostat 2025), often attributed to shifts in work practices during the COVID-19 pandemic. However, remote work popularity was increasing in most European Union member states before the pandemic (European Commission 2020), attributable to technological advances (Zheng et al. 2023), cost-of-living and housing preferences (Böhnen & Kuhnimhof 2024; Marz & Şen 2022; Rafiq et al. 2022), socio-demographic factors (Balbontin et al. 2021; Zheng et al. 2023), commuting time and car availability (Balbontin et al. 2021), and employer and policy incentives (Rafiq et al. 2022; Zheng et al. 2023). In Finland, rates of working from home are amongst the highest in Europe (Eurostat 2025), and in 2023 remote work (or work performed at a location other than the employer’s premises) was done by 35% of all wage and salary workers, with 22% working at least half their time remotely (Statistics Finland 2024). While most remote workers in Finland work from home (Statistics Finland 2024), there is growing usage of alternative ‘third space’ locations such as co-working offices, libraries or vacation homes (Di Marino & Lapintie 2020).
The potential impacts of working from home on travel, energy use, greenhouse gas (GHG) emissions, physical activity, social interactions and urban planning have been widely studied (Sepanta & O’Brien 2023). One potential benefit of remote work is reduced commuting, and therefore transport-related GHG emissions (Hook et al. 2020). In Finland, transport accounts for around one-fifth of total GHG emissions, 90% of which are from road passenger vehicles (FTIA 2020)—around 8.8 million t of CO2 annually. Reductions in work-related travel could, theoretically, lead to significant reductions in transport emissions. However, such reductions are not straightforward. Studies indicate that rebound effects can offset some of the initial gains or even lead to a net rise in GHG emissions (Hook et al. 2020). Rebound effects include: increased non-work trips for activities that would normally occur as trip-chaining during commutes (Asgari et al. 2016; Kim 2016; Kim et al. 2015; Zhu 2012), more activities at the suburban level (Balbontin et al. 2021; Marz & Şen 2022), and increased household energy consumption (Larson & Zhao 2017). Another unintended consequence from remote work may be urban sprawl (Marz & Şen 2022), with individuals living further from employer’s premises and increased commute distances on office days (Goulias et al. 2020; Marz & Şen 2022).
Changes in work-related travel behaviour can also have potential health implications. Increased physical activity related to active commuting by walking or cycling has a positive impact on health, reducing instances of cardiovascular disease, cancer (Celis-Morales et al. 2017; Jarrett et al. 2012; Patterson et al. 2020), dementia (Jarrett et al. 2012), diabetes (Jarrett et al. 2012; Laverty et al. 2013; Saunders et al. 2013), and all-cause mortality (Celis-Morales et al. 2017). Cost–benefit analyses have concluded a measurable positive return on investment in active travel due to savings in national health services (Chapman et al. 2018; Jarrett et al. 2012). However, working from home can remove incidental opportunities for physical activity during commutes, and home-based workers may be more sedentary (Wells et al. 2023). Poor ergonomics are another potential shortcoming of remote work (Statistics Finland 2024). Indirect health benefits from remote working may include reduced traffic-related air pollution emissions which would improve public health (Stefaniec et al. 2024).
Due to the growth in remote work, there is increasing research interest in the implications of working across multiple locations. While many spatial analyses of employment in urban areas focus on a single work location (traditionally the employer’s premises), work can increasingly be considered to occur in a probability space depending on the time spent at different locations (Shearmur 2021), including third spaces such as co-working offices. Third spaces may be used by unaffiliated workers such as freelancers as an alternative to working from home. However, they are also increasingly used by remote working individuals affiliated with an employer; an estimated one-third of co-working space users are affiliated (Foertsch 2019), most of whom have their fees paid by their employer. For employers, co-working spaces can be a way to provide employees flexibility and autonomy (Spreitzer et al. 2020) and a sense of community and inclusion (Jeske & Ruwe 2019).
Multi-locational work may have implications for commuting modes, their environmental impacts and physical activity. Alternative workspaces may be chosen based on proximity to home, enabling affiliated employees to save commuting time, use active transportation and conduct work in environments with appropriate ergonomics. For unaffiliated workers, travelling to alternative workspaces may increase commuting-related emissions relative to working from home, but with opportunities for physical activity. Other alternative workspaces such as vacation homes or hotels may be significant distances away. While certain alternative workspaces may offer potential benefits for emissions and health, there is little evidence on how travel behaviours and emissions differ depending on work location.
This study examines travel behaviours of remote working individuals in Finland to assess differences in commuting-related GHG emissions and physical activity across work locations, characterise remote worker travel profiles and conduct an illustrative analysis of co-working locations in the Helsinki Metropolitan Area (HMA) that could support environment and health goals. A survey is conducted gathering data from both affiliated and unaffiliated remote workers across Finland on their home location (HL), employer’s location (EL) and alternative work locations (AL), and travel modes between them. Using network analysis, the commuting distance and times between different locations are calculated using their specified travel modes, the GHG emissions from this travel are estimated, and differences in travel behaviours and emissions between workspaces are analysed. Clustering methods are used to classify survey respondents into different types of remote worker categories based on commuting behaviours, revealing differences in their socio-demographic characteristics. Finally, using spatial data on the remote working population and co-working spaces in the HMA, an illustrative analysis of how co-working spaces could be strategically placed to leverage their potential benefits is conducted.
2. METHODS
2.1 SURVEY
Survey data were collected using a Public Participation Geographic Information Systems (PPGIS) questionnaire using Maptionnaire. The survey, in English and Finnish, was promoted via social media and posters displayed in various AL around Tampere and Helsinki, and via posters and intranet newsletters at university campuses, the sector where employees have the greatest possibility to influence their remote working hours (Statistics Finland 2024). The survey was restricted to working-age individuals residing in Finland. The survey received ethical and data protection approval from Tampere University Research Data Services before distribution, as despite the survey being anonymous, commuting data may be personally identifying. Respondents provided informed consent before beginning.
The survey (see Appendix 1 in the supplemental data online) asked for basic demographic data, including age range, household income and gender. Respondents were asked to provide the approximate location of their HL, EL and various AL on a map, their preferred travel mode to each work location, and the frequency with which they travelled to each location (per week, month or year) during summer and winter to account for seasonal differences in travel mode. Travel modes included walking, cycling, electric scooter, car, motorcycle, local or long-distance bus, local or long-distance train, tram, boat, or airplane. AL types included designated co-working spaces, public spaces such as a library, café or restaurant, shared workshops, and (to capture rebound effects) cottage or holiday home.
Data were then standardised, converting the frequency of work trips (reported as per week, month or year) into the number of trips annually, assuming five holiday weeks per year. The primary mode of travel was reclassified as public transit, driving, walking, cycling or other. Respondents who did not provide complete answers were removed.
2.2 DISTANCE AND GHG EMISSIONS
The Google Maps Distance Matrix application programming interface (API) was used to calculate travel distances and travel times between respondents’ HL, EL and AL using the gmapsdistance package for those driving, taking public transit, walking or cycling, assuming leaving at 08:00 hours and returning at 17:00 hours on weekdays. If the API failed more than five times, it returned no data, and assumed no clear route was available and the respondents were removed.
For land travel, emissions in passenger-km (pkm) were calculated based on travel distances, mode and mode-specific emission factors for Finland (Table 1), considering direct emissions only. CO2-equivalent emissions (CO2e) for road vehicles were obtained from the UK government’s GHG conversion factors for company reporting (DEFRA 2024), assuming an average of 1.6 persons per passenger car, 11.1 passengers per bus and a single passenger on motorcycles. Emission factors for passenger trains were obtained from VR, the national railway operator in Finland (VR 2025). The analysis was repeated with emission factors for buses reduced to zero to account for the rapid electrification of local buses in many Finnish cities.
Table 1:
Emission factors for the different modes of transport.
| TRAVEL MODE | EMISSION FACTOR (gCO2e/pkm) | REFERENCE |
|---|---|---|
| Car | 108 | DEFRA (2024) |
| Motorcycle | 101 | DEFRA (2024) |
| Bus | 27.2 | DEFRA (2024) |
| Long-distance train | 1.4 | VR (2025) |
| Tram or local train | 0.0 | – |
| Bicycle or e-bike | 0.0 | – |
| Scooter | 0.0 | – |
| Walk | 0.0 | – |
For those who flew, the R package geosphere was used to calculate the great circle distance between the home and work locations, while footprint was used to estimate the emissions of a flight between the two locations. For all respondents, emissions were calculated for a return journey, as well as the cumulative distance and emissions based on the self-reported number of trips made per year.
2.3 STATISTICAL ANALYSIS AND CLUSTERING
Descriptive statistics were calculated for respondents for whom travel distances and times were able to be calculated, including the numbers of respondents in different demographic categories, their rates of remote working, and median annual work-related travel distances and emissions. A demographic comparison between the survey respondents and the broader Finnish population examined survey representativeness.
Commuting distances, travel mode and GHG emissions between EL and AL were then compared. A logistic regression was used to examine differences in active travel by work location, including both trips that had at least one active travel leg, and those where the primary mode was active. The Mann–Whitney U-test was used to test for differences in single-trip distances and time travelled. For those using both EL and designated co-working spaces, a Wilcoxon signed-rank test was used to examine differences in individuals’ travel distances and times. For emissions, a generalised additive model (GAM) with a Tweedie distribution was used to account for zero inflation (i.e. many zero-emissions trips) and right skewed (e.g. a small number of high emissions trips).
Commuting distance, mode and frequency may exhibit different patterns. Typologies of remote workers were developed to understand different types of travel and the characteristics of people who undertake them. This followed a two-stage clustering approach. First, trips were clustered based on their transport mode combinations, then individuals were clustered based on the distance and frequencies of the different types of trips they made.
Multiple correspondence analysis (MCA) (Greenacre & Blasius 2006) was used to transform self-reported combinations of travel time for the modes of different legs of commutes into continuous variables using the Prince library for the Python programming language (Halford 2016/25). Hierarchical clustering was then applied to trips using Ward’s linkage method (Hastie et al. 2009; Ward Jr 1963), which minimises variance within clusters through iterative merging.
Once the trips were classified, individuals were characterised based on the trip clusters. Each individual was associated with two variables (normalised by z-score) per trip cluster that capture the frequency and distance of trips. Principal component analysis (PCA) was applied to reduce dimensionality and capture the main components of individuals’ travel behaviour. The PCA components summarise how strongly individuals are associated with different trip clusters, and thus the travel behaviour linked to work-related trips. Spectral clustering was then used on the PCA representation of individuals. This method was chosen as it can deal with non-convex clusters and higher dimensional spaces, which the data presented even with the dimensionality reduction methods applied.
The number of MCA components used was 10, selected based on increasing the number of components until the addition of an extra cluster had a minimal change in the explained variance of the new component (< 1%). The number of trip clusters, eight, was selected based on the silhouette score (SS) (Rousseeuw 1987) for different cluster numbers, which measures how distinct clusters are from each other. Additional clusters beyond eight had marginally higher SS, but also caused a loss of interpretability and resulted in clusters with very few members. The number of PCA components selected was four; additional components had small effects on the explained variance, and complicated interpretation of relationships between components.
2.4 POTENTIAL CO-WORKING LOCATIONS
Informed by the above methods, geographical information systems (GIS) was used to conduct an illustrative analysis of how co-working locations spaces could be located within 15-min walking distance of the potential remote working population in the HMA. This population was derived from gridded 250 × 250 m demographic data for individuals and their employment sector from SYKE & Statistics Finland (2023). For each grid cell, the number of potential remote workers was estimated based on the industrial sector of employment of the residents, and the estimated percentage of sector jobs able to be done remotely (Statistics Finland 2025a). Existing co-working spaces were identified from popular co-working service aggregators such as Spacent, with the locations geocoded using the geopy Python library. Coordinates of offices that could not be geocoded were added manually.
Potential locations for co-working spaces are based on the population of potential remote workers without easy access to current spaces. Areas currently within a 15-min walk of co-working space were identified by drawing buffers (1175 m, assumed equivalent to a 15-min walking distance) around existing co-working spaces and dissolved where overlaps occurred. Similarly, areas within a 15-min walk of the potential co-working population were estimated using the same size buffers around grid cells with populations employed in sectors that may work remotely. The buffers around current co-working spaces were subtracted from the buffers of remote working population, identifying areas where co-working spaces are currently absent but would be within a 15-min walk of potential remote workers.
A graph-optimisation approach was used to locate potential co-working spaces within these areas. Criteria were that they are no closer than a 30-min walk away from each other or an existing co-working space, and are current locations of office buildings in the HMA (from the same YKR dataset; SYKE & Statistics Finland 2023). This prioritises areas where existing offices could be adapted and excludes unsuitable areas.
Centroids of grid cells with offices were merged with point locations of existing co-working spaces. A conflict graph was created where nodes represented these locations, while edges connect pairs of points closer than 2350 m (an assumed 30-min walk), resulting in an undirected, unweighted graph. New co-working office locations that were too close to existing co-working spaces were removed. The remaining subgraph of new points was solved as a maximum independent set problem, resulting in largest possible set of locations where no two locations are closer than 2350 m. The output consisted of all existing co-working spaces plus the optimised subset of new co-working offices. Finally, the number of remote workers within a 15-min walk of current and future co-working spaces was calculated for visualisation purposes, and sites with low potential remote workers (fewer than 96) excluded from the final visualisation.
3. RESULTS
3.1 RESPONDENT CHARACTERISTICS
The total survey sample included 383 employed respondents, working from 444 unique work locations (excluding homes). Of these, 287 provided complete information on home and work locations and travel mode and were employed. From these, travel distance and time could be calculated for 268 respondents. Of these, 57 use AL while affiliated to an EL.
Female respondents are overrepresented in the survey (66%) relative to the current working-age population (49%) (Statistics Finland 2025b), although recent data indicate that 37% of women work remotely, compared with 33% of men (Statistics Finland 2024). Compared with statistical data, there is a slight overrepresentation of the 25–44-year age group (which represents 45% of the Finnish labour force) and underrepresentation of the 15–24-year age group (which represents 10% of the Finnish labour force) (Statistics Finland 2025b).
The median cumulative travel distances, emissions and commuting time for the surveyed population are also shown in Table 2. These values are low, an expected result due to the focus on remote workers. In addition, the relatively frequent use of public or active transit (62% of the trips), with low emissions/pkm, contribute to lower levels of annual emissions that would be expected from a regular commuter. Interquartile ranges (IQRs) of distances are significant, illustrating the right skew of the data, with long travel distances for a minority of respondents.
Table 2
Details of the survey respondents where complete travel information was provided.
| GROUP | COUNT | FREQUENCY (%) | % DAYS OF REMOTE WORK | MEDIAN CUMULATIVE ANNUAL TRAVEL-TO-WORK LOCATIONS (IQR) (km) | MEDIAN CUMULATIVE ANNUAL COMMUTE EMISSIONS (IQR) (kgCO2e) | MEDIAN CUMULATIVE ANNUAL COMMUTE TIME (h) |
|---|---|---|---|---|---|---|
| Gender | ||||||
| Female | 178 | 66.4% | 60% | 2,864 (1,296–5,834) | 51.7 (0–342) | 131 |
| Male | 85 | 31.7% | 40% | 2,422 (803–7,202) | 0 (0–117) | 145 |
| Non-binary | 3 | 1.1% | 40% | 1,547 (1,001–1,989) | 0 (0–6) | 88 |
| N/A | 2 | 0.7% | 15% | 25,269 (14,713–35,825) | 2,561 (1,337–3,785) | 469 |
| Age (years) | ||||||
| 18–24 | 10 | 3.7% | 12.5% | 1,322 (539–2,962) | 0 (0–11.3) | 169 |
| 25–44 | 134 | 50.0% | 60% | 2,407 (1,315–5,357) | 29.2 (0–108) | 135 |
| 45–64 | 122 | 45.5% | 60% | 3,165 (884–7,266) | 66.1 (0–408) | 133 |
| 65–70 | 2 | 0.7% | 20% | 14,705 (10,565–18,844) | 1,241 (621–1,862) | 895 |
| Household income (€) | ||||||
| €0–14,999 | 8 | 3.0% | 50% | 1,173 (351–1,593) | 0 (0–0) | 115 |
| €15,000–19,999 | 6 | 2.2% | 30% | 3,210 (1,567–3,348) | 0 (0–0) | 220 |
| €20,000–39,999 | 24 | 8.9% | 55% | 3,341 (1,658–5,477) | 44.4 (0–124) | 122 |
| €40,000–69,999 | 76 | 28.4% | 50% | 2,305 (795–5,586) | 0 (0–108) | 131 |
| €70,000–99,999 | 69 | 25.7% | 60% | 3,525 (1,642–7,359) | 30.1 (0–203) | 150 |
| €100,000–119,999 | 36 | 13.4% | 67% | 2,782 (1,230–6,140) | 44.7 (0–124) | 129 |
| €120,000–149,999 | 17 | 6.3% | 40% | 3,100 (2,171–5,844) | 0 (0–200) | 148 |
| ≥ €150,000 | 12 | 4.5% | 58% | 2,337 (793–4,052) | 4.8 (0–54.2) | 105 |
| N/A | 20 | 7.5% | 78% | 2,080 (863–4,154) | 20.1 (0–187) | 71 |
| Professional status | ||||||
| Entry-level employee | 62 | 23.1% | 6% | 2,328 (695–5,293) | 24.4 (0–233) | 118 |
| Senior employee | 109 | 40.7% | 60% | 3,100 (1,358–7,393) | 59.2 (0–362) | 132 |
| Worker | 54 | 20.1% | 54% | 2,814 (1,321–4,910) | 30.2 (0–131) | 157 |
| Manager | 14 | 5.2% | 40% | 2,997 (440–4,911) | 35.3 (2–179) | 144 |
| Other | 8 | 3.0% | 77% | 1,692 (509–14,704) | 0 (0–16) | 95 |
| N/A | 21 | 7.8% | 40% | 2,073 (1,411–5,545) | 44.5 (0–155) | 121 |
[i] Note: IQR = interquartile range; N/A = not available.
3.2 WORK LOCATIONS AND TRAVEL MODES
The majority of workers split time between working from home and EL, while around 20% made use of AL. The majority of those using AL (around 85%) were also affiliated to an EL. The most common AL visited by respondents were designated external co-working spaces (35%), followed by cottages or holiday homes (32%).
The modes of travel varied depending on work location (Figure 1), with trips to more distant AL such as cottages more likely to include travel by car, while most trips to designated co-working spaces and EL included some walking. Public transport has the highest rates of passive transport for both EL and AL. Slight decreases in the cycling rates were observable between seasons, with a shift towards public transit during winter. Two respondents took flights, one of which was to their EL once and one to an AL twice a year.
Figure 1
Percentage of individual trips to different work locations by different travel modes.
Note: Percentages within work location groups sum to more than 100% due to multimodal trips.
The logistic regression indicated an increased likelihood having an active leg during commutes to designated co-working spaces (odds ratio (OR) = 1.7, 95% confidence interval (CI) = 1.4–1.9) or public spaces (OR = 1.7, CI = 1.3–2.3) compared with those travelling to EL, while those travelling to cottages or holiday homes were significantly less likely to use active travel (OR = 0.4, CI = 0.3–0.4). Amongst affiliated workers with the option of both AL or EL, the odds of using active travel to a co-working space (OR = 1.6, CI = 1.3–1.9) or public space (OR = 1.4, CI = 1.1–1.9) were also significantly higher relative to travel to their EL.
The odds of a worker using active travel as the primary mode of their commute to designated AL workspaces is also significant (OR = 2.7, CI = 2.3–3.3) compared with EL, and comparatively higher than the odds of a single active leg. Active travel was not a primary mode for any travel to cottages or holiday homes.
3.3 TRAVEL DISTANCES AND GHG EMISSIONS
Accounting for the number of working days per week, the percentage of remote working days, and the modes and distances used to travel to AL and EL, remote workers in this survey collectively avoided 99,042 kgCO2e emitted by working at home or at AL, or around 259 kg per person per year.
The distributions of return and cumulative annual commute distances, emissions, and travel time for trips to the employer’s premises and alternative workplaces are shown in Figure 2 for the calculations assuming non-zero emissions from buses; Appendix 2 in the supplemental data online shows the same distributions assuming zero bus emissions.
Figure 2
Density plots showing the distributions of travel distance (top), greenhouse gas (GHG) emissions (middle) and travel time (bottom) to different work locations for single return trips (left) and the cumulative annual for all trips (right).
Note: Log scale.
Travel to both locations show a wide range of distances, including some outlier instances of travel by plane to AL and EL and trips to cottages or holiday homes. The median distance to EL was 8.7 km (IQR = 3.4–20.1 km), with a median commute of around 27 min. Designated workspaces were similar (median = 9.7 km, IQR = 4.8–11.7 km) with a median 24-min commute and no significant difference in distance or time indicated by the Mann–Whitney U-test. Café/restaurants and public spaces were significantly closer (2.8 and 3.1 km, respectively) than EL and took significantly less travel time to commute to (12 and 15 min, respectively). The Wilcoxon test indicates that, for those working at both EL and designated co-working locations, there were significant shorter commute distance (p = 0.005) and times (p = 0.001) to the co-working spaces.
Median CO2e emissions for a return trip were comparatively lower for designated workspaces (0.0 kg, range = 0.0–0.29 kg) and public spaces (median = 0.09 kg, range = 0–0.28 kg) compared with EL (median = 0.16 kg, range = 0–0.9 kg). The GAM analysis indicated significantly lower CO2e emissions for single trips to designated AL (p = 6.7e–05) and public spaces (p = 0.001) relative to EL, driven primarily by the increased probability of zero-emissions travel to local AL. Emissions of travel to cottages was significantly higher. The differences continued to be significant under the assumption of zero emissions from buses. For those affiliated with an employer, there were no significant differences between emissions between EL and designated workspaces.
3.4 REMOTE WORK TRAVEL TYPOLOGIES
Clustering provided a more in-depth analysis of typical travel and work typologies within the survey population. Table 3 briefly describes the cluster features and shows a descriptive label of the cluster that captures their travel behaviour and features. Table 4 shows more detailed quantitative features of the typology clusters. In contrast to the preceding analysis, disaggregate transport modes were used.
Table 3:
Typologies of travel patterns of remote workers based on travel mode, trip distances and frequencies.
| CLUSTER | TYPOLOGY |
|---|---|
| Cluster 0: Wealthy, regular car users (high emitters) | Dominated by car use (modal share = 0.76), resulting in the lowest mode diversity. Secondary mode is bus. Trips are consistent in length, with the second highest median distance. Members have the highest mean income by a large margin (about €2,000 above the next highest cluster). Emissions are high: second in mean trip emissions and third in cumulative emissions |
| Cluster 1: Commuting urban transit users (low emitters) | Main mode is bus (modal share = 0.26), followed by local tram. This cluster exhibits the highest trip frequency, indicating low remote work prevalence. Emissions are low due to a heavy reliance on public transport |
| Cluster 2: Active travellers with cars (low emitters) | Walking is the primary mode (modal share = 0.27), with car as the secondary mode. High mode diversity and moderate spread of trip distances suggest complex travel behaviour. Emissions are among the lowest, both per trip and cumulatively. Members have the lowest mean income, though still above the Finnish median |
| Cluster 3: High-emitting long-distance travellers (highest emitters) | Car is the main mode, followed by long-distance train. This group has the highest emissions per trip and cumulatively. Despite high emissions, mode diversity is notable due to train use. Median trip distance is the highest, more than twice that of the next cluster. The distribution of trip distance is the largest, capturing the long-distance journeys, but also the flexibility in commute distances associated with car usage |
| Cluster 4: Walking car users (low emitters) | Like cluster 2, but with a higher walking share (0.30) and slightly less mode diversity. Emissions are higher than cluster 2, both per trip and cumulatively |
| Cluster 5: Walking commuters (lowest emitters) | Smallest cluster and the lowest emitters overall. Walking is the dominant mode (modal share = 0.57), followed by bus. Trips are short (lowest median distance), and commuting frequency is high |
| Cluster 6: High-frequency car commuters (high emitters) | Car accounts for 0.43 of modal share. While average trip emissions are moderate, cumulative emissions are high due to frequent travel. This cluster has the lowest median trip length among car-dominant clusters (0, 3 and 6). Compared with cluster 3, trips are shorter and mode diversity is lower, though income levels are similar. Compared with cluster 0, car usage is less dominant and the relative spread of trip distances is higher |
| Cluster 7: High-frequency urban remote workers (low emitters) | Characterised by the lowest commuting frequency and a strong preference for tram (modal share = 0.56). Cumulative emissions are low (second lowest after cluster 5). Trip lengths are consistent with narrowest trip length distribution), and members have the second-highest income, typical of core Finnish urban areas |
Table 4
Statistics of the typology clusters.
| CLUSTER | N | MAIN MODE | MODE DIVERSITY | MEAN TRIP FREQUENCY (± SD) | MEDIAN TRIP LENGTH (± IQR) (km) | MEAN TRIP CO2 EMISSIONS (± SD) (kg) | MEAN CUMULATIVE CO2 EMISSIONS (± SD) (kg) |
|---|---|---|---|---|---|---|---|
| Cluster 0: Wealthy, regular car users (high emitters) | 35 | Car (0.76) | 1.12 | 85.05 (± 51.43) | 13.83 (± 14.67) | 2.16 (± 2.10) | 155.84 (± 170.78) |
| Cluster 1: Commuting urban transit users (low emitters) | 20 | Bus (0.26) | 2.36 | 144.42 (± 77.27) | 7.91 (± 8.98) | 0.29 (± 0.75) | 51.86 (± 139.16) |
| Cluster 2: Active travellers with cars (low emitters) | 39 | Walking (0.27) | 2.56 | 99.58 (± 87.91) | 5.50 (± 11.31) | 0.40 (± 1.19) | 20.71 (± 64.42) |
| Cluster 3: High-emitting long-distance travellers (highest emitters) | 72 | Car (0.35) | 2.59 | 99.28 (± 87.36) | 31.67 (± 160.90) | 6.19 (± 18.20) | 287.83 (± 991.73) |
| Cluster 4: Walking car users (low emitters) | 39 | Walking (0.31) | 2.43 | 117.44 (± 102.86) | 3.99 (± 7.83) | 1.69 (± 7.25) | 70.35 (± 700.37) |
| Cluster 5: Walking commuters (lowest emitters) | 13 | Walking (0.57) | 1.41 | 138.86 (± 54.97) | 2.60 (± 5.74) | 0.04 (± 0.11) | 5.66 (± 16.55) |
| Cluster 6: High-frequency car commuters (high emitters) | 35 | Car (0.44) | 1.98 | 138.39 (± 87.18) | 8.29 (± 12.88) | 1.52 (± 3.31) | 185.82 (± 346.32) |
| Cluster 7: High-frequency urban remote workers (low emitters) | 15 | Local tram (0.56) | 1.70 | 56.88 (± 35.45) | 9.23 (± 4.76) | 0.53 (± 1.82) | 14.72 (± 44.34) |
[i] Note: N is the number of members in the cluster. The main travel mode is the modal means of transport and is given along with the proportion of trips that use it in parentheses. Mode diversity is calculated as the entropy of the travel mode distribution. Uncertainties for means are taken to be the standard deviation (SD). For the median trip frequency, the uncertainty is given as the interquartile range (IQR).
Clusters can be split between car, walking and bus primary travel modes, which while it makes public transit use less visible, allows for finer analysis of the diversity of modal choice. Cluster 0 characterised regular car users, with the least diverse modal mix, highest median income, second highest mean emissions per trip and third highest cumulative emissions per individual.
Cluster 3 was the largest cluster with 72 members, had the highest CO2 emissions. It is distinguished by the high number of long-distance trips, with long-distance train being the second most used mode, and the median trip distance double the next highest cluster. However, car use was also dominant, contributing to high emissions per trip.
Walking was the most used mode of clusters 2, 4 and 5, which were labelled active travellers with cars, walking car users and walking commuters, respectively. Clusters 2 and 4 had a relatively diverse mode usage and represent car users that can shift towards active travel; these were otherwise similar, with cluster 4 presenting higher per trip and cumulative emissions. Cluster 5 had walking as a strongly dominant mode and was the cluster with the lowest carbon emissions.
Clusters 1 and 7 were characterised by public transport as the main travel mode and comparatively low emissions. Difference in these clusters are due to diversity of modes, with cluster 7 presenting more dominant use of the main transport mode with more than half the trips made by tram. The most notable difference, however, was the commuting frequency, reflected in the difference of cumulative emissions per individual. Cluster 1 presented almost double the median cumulative emissions, even though cluster 7 had higher median emissions per trip.
Cluster 6 was composed predominantly of car users with low modal diversity, and the second highest cumulative emissions per individual (only behind the long-distance travellers of cluster 3) attributed to the highest trip frequency. This was the cluster of remote workers who commuted the most.
3.5 OPTIMAL CO-WORKING LOCATIONS
Survey results indicated that co-working spaces could support increased active mobility in remote workers. Figure 3 shows the illustrative example of how existing and potential co-working spaces can support this. Most existing co-working spaces are concentrated in the city centre and its adjacent districts, while only a few are in the city of Espoo, to the west, and even fewer in Vantaa, north of Helsinki. New potential co-working locations are largely in non-central areas, since criteria meant they should be in areas unserved by existing spaces. These potential new locations can serve comparable numbers of residents with the existing central locations.
Figure 3
Current (blue) and optimised locations of potential new co-working spaces (red) that are within 15-min walks of potential remote working populations.
4. DISCUSSION
This study examines commuting behaviours to EL or AL by investigating travel modes, times, distances and emissions, and the characteristics of different commuters. Results indicate designated and public AL have a higher likelihood of walking and cycling compared with EL for both affiliated and unaffiliated workers. While public transport is relatively more common for trips to EL than to AL, this is largely due to the mode share for active travel being higher for AL. This suggests that AL can help mitigate some of the potential sedentary activity issues of remote work from home. The exception is work from holiday homes, which is entirely passive travel and demonstrates significant emission rebound effects. While such trips do not represent normal commuting patterns, they were included to capture multi-locationality of work and rebound effects, with AL analysed independently.
The highest proportion of zero-emission trips are made to AL, with median single-trip emissions for commuting to EL and designated AL of 0.16 kg (range = 0–0.861) CO2e and 0 kg (range = 0–0.29) CO2e, respectively. Average single-trip emissions to AL are higher than to EL, skewed by outlier trips to AL cottages or holiday homes. Amongst those with options of EL and designated AL, there was significantly less travel time and distance to the AL, suggesting that workers choose locations closer to their homes that save commuting time; this shorter distance may help incentivise increased active travel, with benefits for health.
Clustering gave nuance to the commuting behaviours. Travel typologies show high heterogeneity in remote workers’ work location type, travel mode choice and commuting distances. Clusters characterised by walking, cycling or public transport use (clusters 2, 4, 5 and 7) are associated with shorter commuting distances, lower cumulative emissions and more frequent use of nearby work locations, suggesting that the benefits of AL are most readily realised in relatively dense, urban environments that have either good transit service or are appropriate for active trips. In contrast, clusters dominated by car use and longer commuting distances (notably clusters 0, 3 and 6) demonstrate high cumulative commuting emissions, indicative of travel behaviours more commonly associated with low density or peripheral residential locations where longer distances and limited public transport provision increase reliance on private vehicles. For these groups, remote and hybrid work may reduce overall commuting frequency while simultaneously enabling residential relocation to more distant locations, increasing the emissions of the trips that do occur. This is consistent with concerns around rebound effects and urban sprawl associated with remote work; however, it could be ameliorated by improving the relative location of alternative workplaces, such as co-working locations. It also provides evidence on population groups (e.g. clusters 2 and 4) which may shift their travel behaviours from car to active travel, given appropriate work locations.
While the results suggest AL could support the reduction in GHG emissions and improve rates of active travel, currently co-working spaces are more available to higher income individuals, or as an employee benefit, as most of them are privately owned (Chanson & Sakka 2023). Through public investments, co-working space accessibility could be improved, with public savings on transport and improved population health potentially justifying this expenditure. The spatial analysis of potential co-working spaces is intended as an illustrative analysis of how strategic placement of co-working spaces in relatively dense urban locations could potentially increase active travel and align with the principles of the 15-min city. This would be particularly effective for workers in clusters that are susceptible to changing their commuting behaviour, with mixed-mode usage and moderate commuting distances. The potential impacts on GHG of this analysis are less clear: for those who normally work at home, these may present an increase in emissions, while there was a significantly lower emissions for trips to designated AL compared with EL; this was not the case for affiliated workers.
There is growing evidence that remote work is already restructuring cities by dispersing economic activity away from city centres and encouraging household relocation towards suburban areas within metropolitan regions (Ramani et al. 2024). Long term, these trends may contribute to increased suburban sprawl or a redistribution of the population towards smaller urban areas within the same regions (Biagetti et al. 2024). Remote work has been found to decrease transportation costs by 23%, which in turn increases employees’ flexibility in choosing housing locations further away from potentially expensive areas near employers’ offices (Larson & Zhao 2017). Frequent work from home may gradually increase the commute distance (Böhnen & Kuhnimhof 2024), leading to a de-densification of the city, and resulting in longer commutes to the EL when they occur (Marz & Şen 2022). Lower density suburban areas are typically less well served by public transport (Muñiz & Garcia-López 2019), and when combined with longer commuting distances, may incentivise car use. Urban sprawl can also evoke socio-economic challenges as lower income individuals may not have possibilities for remote work.
Thus, remote work may have implications for urban form, commuting patterns and CO2 emissions (Tao et al. 2023). However, a shift towards a more polycentric urban structure, not necessarily at odds with residential restructuring driven by hybrid work, has been associated with shorter commuting distances (Zhao et al. 2011). While transport mode choice is shaped by urban form and individual travel attitudes, travel time remains a key predictor of commuting mode choice (Frank et al. 2008). Reducing commute distances for at least some journeys through closer-to-home AL could therefore help counteract some of the negative transport and emissions impacts of residential relocation; a recent review finds that co-working spaces in residential areas can indeed encourage more sustainable commuting practices (Vogl & Orel 2024).
4.1 LIMITATIONS
There are important limitations to this study. First, 268 respondents provided sufficient data to calculate travel times and distances, a minority of whom use AL. Respondents are biased towards female workers and those between the ages of 25 and 44 years. While females in Finland tend to work remotely more often than males, this does not explain all the response bias. Therefore, it is possible that the results show some skewedness, as socio-demographic factors have previously been suggested to influence choices in commuting and remote working (Zheng et al. 2023). The survey is not representative, and as such has not been used to extrapolate to the population level. The analysis has assumed that AL are direct substitutes for EL. In reality, they are likely complementary spaces, with non-affiliated workers able to choose between HL and AL, while affiliated workers may choose between HL, AL and EL. Choices may not be fully flexible, but may be influenced by the mandated presence at EL on certain days. Around 58% of salary earners in Finland have reported that they are unable or uninterested in working remotely (Statistics Finland 2024), meaning even those who can work remotely would prefer not to do so.
Rebound effects from increased non-work trips, the correlation between the frequency of those trips and full- versus part-day remote workers, and the impacts of working from home were asked in the survey, but received low response rates and are therefore excluded from the study. Survey respondents can overreport the amount of active travel undertaken (Panter et al. 2014). The survey data used in this study are cross-sectional, meaning the results represent a snapshot of the modes and emissions of commuting to AL and EL. It has previously been shown that commute distances are likely to increase for employees who regularly work remotely (Böhnen & Kuhnimhof 2024), so estimated distances to EL may be longer than for typical workers.
The analysis of the survey results also has limitations. The network calculation estimates distances between the homes and work location, and it is assumed that the emission factor for the primary mode for this distance. This means that zero carbon legs of journeys, such as walking or cycling to public transit stops, are included in the emission estimate. The assumption was required, as the network calculation used did not break down the trips into separate legs. The calculations of physical activity consider self-reported active travel legs, as well as active legs as the primary mode. The approach assumes that commuters take the most direct route.
It was assumed that working from home has zero commuting-related emissions, thus commuting to AL represents a comparative increase in emissions for remote workers. However, prior studies have shown that working from home may increase overall travel emissions due to increased non-work trips (Asgari et al. 2016) that could otherwise occur as trip-chaining during commutes. The results are also sensitive to assumptions on CO2 emissions/pkm travelled. Data on the emissions of different modes were used, but particularly for buses there can be a wide variation in the occupancy rate which can impact pkm assumptions. In addition, there are rapidly growing numbers of electric buses in urban areas in Finland, while the emission factor used is inclusive of diesel buses. Emissions include only direct tailpipe emissions, and not indirect emissions from electricity generation.
The main limitations of the clustering are from the sample size which constrains the observed features of trips and individuals. While the self-selection of the respondents guarantees that remote workers are represented, the bias to high-income respondents will affect respondents’ observed travel behaviour. Another important factor is the use of a survey, which impacts the type of data that are obtainable. To reduce survey fatigue and increase completion rates, categorical variables were used to represent quantitative ranges. This data representation reduces the variation in modal distributions, making clusters resemble each other more and making classification harder.
The co-working space location analysis was intended to examine how spaces could be located to support climate and physical activity goals. Data on existing co-working spaces were acquired from co-working platforms, but it is acknowledged that there may be unregistered spaces and public spaces that were not included. The capacity of these spaces is not examined when assigning potential locations as these data were not available. When considering the travel distance, a simple buffer was used, while future work could use network distances. Calculations were performed for the total potential remote working population by job sector; however, there will continue to be individuals who prefer to work at EL, as well as those who continue to be car dependent regardless of work location.
4.2 FUTURE RESEARCH
Future surveys with more respondents should be employed to support the study’s conclusions. Research could include detailed analysis of the impacts of multimodal commuting, such as the use of private transport to access public transport hubs. This would address current limitations stemming from the assignment of a single transport mode per trip in the estimate of emissions. Additionally, further research would also help mitigate other biases such as gender imbalances.
From a spatial planning perspective, future research could assess the precise location of AL and their connectivity with public and active transport infrastructure to identify optimal locations for AL to reduce car dependence and enhance co-working accessibility. Furthermore, longitudinal studies on residency and travel mode behaviour could provide a more comprehensive understanding of how remote work influences both immediate commuting choices and long-term residential and travel patterns. This could be supported by the understanding of the types of people who work remotely and their travel behaviours. The method of identifying potential co-working space locations can be extended to other locations with spatial data on population by employment sector, office space and existing co-working locations. Transport emissions and transportation modes preferences can vary across regions, so research in different locations is necessary to understand local contexts.
5. CONCLUSIONS
This study has sought to examine the travel behaviours and characteristics of individuals working remotely in Finland, focusing on those using alternative work locations, while exploring how alternative work locations may be optimally located to take advantage of their benefits. Results from the survey indicate a greater likelihood of using active travel or zero emission transport to alternative work locations such as designated co-working spaces or public spaces, with implications for greenhouse gases emissions and health. The range of typologies of remote workers can help one understand their travel behaviours, and potential opportunities to use local co-working spaces as a policy tool to encourage active travel. Future work, building upon the same methodology and expanding the survey, will help the values to be representative for a wider population.
AUTHOR CONTRIBUTIONS
J.T.: conceptualisation, methodology, writing—original draft, formal analysis, funding acquisition; L.T.: conceptualisation, methodology, writing—original draft, formal analysis; A.E.M.d.V.; formal analysis, methodology, writing—original draft; P.A.: conceptualisation, formal analysis, writing—original draft; J.V.: conceptualisation, investigation; D.M.B.: conceptualisation, investigation; I.O.: methodology, original draft, investigation.
DATA ACCESSIBILITY
Data with personally identifying information were destroyed as per the data management agreement. Other data are available from the authors upon request.
ETHICAL APPROVAL
The survey received ethical and data protection approval from Tampere University Research Data Services before distribution (ethical approval number TUNI-1112302), as despite the survey being anonymous, commuting data may be personally identifying. Respondents provided informed consent before beginning.
SUPPLEMENTAL DATA
Supplemental data for this article can be accessed at: https://doi.org/10.5334/bc.789.s1